Glossary

A model design where weights are restricted to only three discrete values (-1, 0, or 1) instead of continuous floating-point numbers, drastically reducing model size and computation.

A neural network where each weight is represented using only 1 bit of information (in this case, as one of three values: -1, 0, or 1).

An extreme form of quantization where each weight is represented by just a single bit (0 or 1), maximizing compression but reducing model expressiveness.

An extreme form of compression that represents model weights using only 1 bit of information per value, drastically reducing memory use but with significant quality loss.

A data format that represents model weights using 16 bits per number, balancing memory efficiency with numerical accuracy.

Mathematical shapes (Gaussian distributions) positioned in 3D space used to represent and render 3D scenes efficiently.

Guiding AI model outputs by conditioning on 3D spatial layout information.

Building a complete 3D model of a physical environment from images or sensor data.

Comprehending the three-dimensional structure, objects, and relationships within a physical environment.

A specific quantization method that represents model weights using only 4 bits per number instead of the standard 32 bits, dramatically reducing memory usage.

A quantization level where model weights are stored using only 4 bits per value, significantly reducing model size at the cost of some accuracy.

A specific type of quantization that represents model weights using only 4 bits instead of the original 32 bits, enabling very efficient inference on consumer hardware.

A specific compression method that represents model weights using only 5 bits of data per value, enabling efficient local deployment on resource-constrained hardware.

A quantization method that represents model weights using only 6 bits per value, significantly reducing memory requirements compared to standard 32-bit floating-point storage.

A specific quantization method that represents model weights using 6 bits instead of the standard 32 bits, significantly shrinking the model while maintaining reasonable accuracy.

A quantization method that represents model weights using 8 bits instead of the standard 32 bits, reducing memory usage by approximately 75% while maintaining reasonable performance.

A specific quantization method that represents model weights using 8 bits instead of the standard 32 bits, significantly reducing memory requirements.

A model variant where safety filters and refusal mechanisms have been removed, allowing it to respond to requests without built-in content restrictions.

A technique that removes or disables a model's built-in safety refusal mechanisms, allowing it to respond to a wider range of requests.

Identifying and highlighting the specific regions in medical images where disease or abnormalities are present.

Visual maps showing which regions of a medical image are abnormal, derived from comparing to historical cases.

A measure of relevance between a query and key that is independent of other keys, allowing explicit rejection of irrelevant keys.

When a system declines to make a prediction or recommendation instead of providing an answer.

A tree representation of code structure that shows how statements and expressions relate to each other.

Generating a summary by creating new sentences that capture key information, rather than selecting existing text.

The proportion of draft model's proposed tokens that the target model accepts as correct during speculative decoding.

Designing technology so people with disabilities can use it effectively.

Determining which party in a system is responsible for harms or failures.

A measure of how well an agent performs relative to the computational cost or number of steps it takes.

An internal mathematical encoding of sound properties that a model learns to recognize, such as frequency, pitch, and timbre characteristics.

A rule that decides which point to evaluate next by balancing exploration of new areas with exploitation of promising regions.

The problem of correctly associating a specific action command with the correct agent or subject in a scene.

A failure mode where agents make poor action choices that lead to uninformative observations, cascading into reasoning errors.

The task of identifying and classifying specific actions or activities occurring in video frames.

Creating videos where specific physical actions (like forces or robot movements) control what happens in the scene.

Simulating multiple future steps of an environment given a sequence of actions the agent might take.

A learned encoding of an object that explicitly captures how it responds to and changes under different actions.

The portion of a model's total parameters that are actually used to process a given input; in MoE models, this is typically much smaller than the total parameter count.

The probability distribution of neuron outputs at each layer of a network.

Random variations added to a model's internal computations to test robustness.

A mechanistic interpretability technique that replaces activations during inference to identify which components cause specific behaviors.

The specific configuration of which neurons are active across a network when processing a particular input or task.

The number of bits used to represent intermediate calculations during inference; keeping this higher (like 16-bit) helps preserve model quality when weights are heavily compressed.

Analyzing internal neural network activations to understand what a model has learned or decided at different points.

The process of reducing the precision of intermediate values (activations) computed during model inference, separate from weight quantization.

Controlling model behavior by modifying internal activations during inference without changing model weights.

Bypassing AI safety features by manipulating the internal numerical patterns the model uses to process information.

A training approach where the model chooses which new examples to learn from rather than using random data.

The number of model parameters that are actually used during inference for a given input, as opposed to the total parameters available.

A model architecture where only a subset of parameters are used for each token, reducing computational cost while maintaining model capacity.

The subset of a model's total parameters that are actually used during inference for each input, as opposed to all parameters being used every time.

A mathematical constraint ensuring a causal graph has no cycles, enforcing valid causal structures.

A standard optimizer algorithm commonly used to train neural networks by adjusting weights based on gradients.

A small, specialized module added to a model that modifies its output for a specific task without changing the core model weights.

Custom code written to translate data between incompatible formats or interfaces.

Adding lightweight modules to a pre-trained model to enable new capabilities without retraining the entire model.

An attack that adjusts its strategy based on feedback from the target system to improve its effectiveness.

Educational systems that adjust content difficulty and pacing based on real-time analysis of learner performance and understanding.

A system that dynamically adjusts parameters (like reward weights) based on the current task or input.

Dynamically selecting or modifying prompts based on the specific input query to optimize model performance.

A quantization approach that adjusts its representation strategy based on the distribution of input values.

Dynamically adjusting how much computational effort a model uses based on problem difficulty.

Optimization algorithm that splits problems into smaller parts solved alternately.

Intentional manipulation of input data to trick an AI model into making wrong decisions.

Systematically testing an agent's reasoning to find logical or evidential violations it may have missed.

A training loop where attack and defense agents compete and improve against each other iteratively.

Testing designed to find weaknesses and edge cases rather than help the system succeed.

Deliberately tricky test cases designed to fool AI models, like plausible wrong answers.

Systematically searching for inputs where a model fails, used here to find materials where ML predictions diverge from ground truth.

Training where two networks compete—one generates behavior, the other judges if it matches the expert.

A process where one agent intentionally creates challenging test cases to improve another agent's output.

Training approach where a generator and discriminator compete to improve output quality and realism.

Carefully crafted, often imperceptible changes added to images to fool AI models into producing incorrect outputs.

Deliberately crafted inputs designed to trick an LLM into unsafe or unreliable outputs.

The ability of an AI system to maintain correct behavior even when facing intentionally crafted misleading inputs.

A defense method that trains models on adversarial examples to improve robustness against attacks.

A defense mechanism that protects models from attacks without requiring exposure to adversarial examples during training.

Evaluating the visual appeal and artistic qualities of images or scenes, such as composition and harmony.

Linking emotional or sentiment states between connected entities in a system.

The emotional tone of text, measured as the degree of negativity, positivity, or neutrality in language.

Predicting which areas or objects in a scene are suitable for a specific action or interaction.

An AI system's ability to act autonomously toward goals in its environment.

The degree to which an agent retains independent decision-making capability without external manipulation.

The framework or system that orchestrates how an AI agent retrieves information, calls tools, and processes results.

Coordinating multiple AI agents to work together on complex tasks.

Automatically selecting the most suitable agent(s) for a task from available registries using matching and ranking techniques.

A specific capability or tool that an AI agent can use to accomplish part of a larger task.

The sequence of actions and decisions an agent makes while working toward a goal.

A simulation where independent agents follow simple rules and interact, creating emergent group behavior.

A model designed to act autonomously by making decisions, selecting actions, and using tools to accomplish multi-step tasks.

An AI system that can autonomously plan and execute multi-step tasks, making decisions along the way.

The ability of a model to autonomously plan and execute sequences of actions or tool calls to accomplish a goal.

An approach where an AI model autonomously plans and executes multi-step coding tasks, making decisions about which files to modify and how to structure solutions.

Sequential overhead from cascaded perception, reasoning, and tool-calling loops in agentic systems.

Designing and building systems where AI agents autonomously plan, decide, and act toward goals.

Testing an AI system's ability to complete multi-step tasks that require planning, searching, and taking actions.

A system where an AI model acts as an agent that can call tools repeatedly to solve problems step-by-step, rather than answering in a single pass.

A structured language with explicit control constructs (IF, GOTO, FORALL) that agents use to execute plans deterministically.

The study of local interaction dynamics where one agent's output becomes another agent's input under specific protocol conditions.

AI systems that can process multiple types of input (text, images, etc.) and actively interact with external tools and environments.

Vision systems that extract structured state information needed for an agent to make decisions, not just recognize objects.

Reasoning through explicit tool calls or code execution that can be interpreted and debugged, but may incur latency from external execution.

Training autonomous agents to make sequential decisions by learning from rewards and reusable experience.

A training approach where an AI model learns to make sequential decisions and take autonomous actions to complete multi-step tasks, rather than just responding to individual prompts.

AI systems that iteratively search and synthesize information to solve complex problems autonomously.

An AI agent's ability to detect and fix its own errors by using tools or feedback without human intervention.

Structured approaches where an AI system takes initiative to gather information systematically rather than passively responding to user input.

AI systems that autonomously plan, act, and adapt based on feedback to accomplish multi-step goals in complex environments.

Complex tasks where a model acts autonomously to break down goals into steps, use tools, and make decisions to reach an objective.

Processes where a model autonomously plans and executes multiple steps or tool calls to accomplish a goal, rather than responding to a single prompt.

Combining multiple data points or model outputs into a single summary result.

Interconnected systems where multiple AI components interact through shared data and infrastructure.

Randomness or noise inherent in data that cannot be reduced with more information.

Systematic errors in AI systems that unfairly disadvantage certain groups of people.

Ensuring AI systems treat different groups equitably without discrimination.

Tendency of AI systems to produce similar outputs or behaviors, either naturally or in response to incentives.

A technique that helps the model understand the order and position of words in long sequences without needing to add extra position information to each word.

A model trained to behave safely and follow human values through techniques like safety filtering and refusal of harmful requests.

The process of training a model to behave safely and according to human values and preferences, which base models typically lack.

When an AI model appears aligned under monitoring but subverts its goals when unmonitored.

The process of adjusting a model's behavior to make it safer, more helpful, and better aligned with human values.

Safety constraints built into a model during training to prevent it from generating harmful, biased, or inappropriate content.

Additional training applied to a base model to make it behave safely and follow user intentions more reliably.

A guarantee that higher bids weakly increase an item's chance of being recommended without requiring model retraining.

An early, experimental version of software that is still under development and may have bugs or incomplete features.

Errors caused by confusion between similar or overlapping UI elements when determining which one to interact with.

The linear chain of amino acids that makes up a protein, which determines its structure and function.

The linear arrangement of amino acids that make up a protein, written as a string of letters where each letter represents a different amino acid.

Spreading the cost of an expensive computation across multiple uses to reduce per-use cost.

An attention pattern that restricts a model to only attend to ancestor nodes in a tree structure, enabling efficient tree verification.

Choosing a reference model to compare all other models against in pairwise evaluation tasks.

Bias where initial information disproportionately influences subsequent decisions.

Methods for combining multiple human judgments into a single training signal for the model.

A structured set of guidelines for labeling data with specific linguistic or semantic information.

A systematic process for labeling data with human-verified information to create training datasets.

Variation in how different people label the same content, reflecting genuine differences in perspective rather than labeling error.

The negative electrode in a battery where ions are stored during charging.

Identifying data points or objects that deviate significantly from normal patterns or training data.

A declarative programming paradigm for solving combinatorial problems using logical rules and constraints.

An open-source software license that allows free use, modification, and distribution of code with minimal restrictions.

A permissive open-source license that allows free use, modification, and distribution of software with minimal restrictions.

A permissive open-source license that allows you to use, modify, and distribute software with minimal restrictions.

A permissive open-source license that allows free use, modification, and distribution of software with minimal restrictions.

An interface that allows developers to send requests to and receive responses from an AI model over the internet.

A programmatic interface that allows developers to send requests to the model and receive responses without running it locally.

The ability to access and use a model programmatically through an application programming interface, allowing developers to integrate it into their applications.

A model that can be used through an application programming interface, allowing developers to integrate it into their applications programmatically.

Access to a model through an application programming interface, allowing developers to integrate the model into their applications and services programmatically.

The ability of a service to work with the same code and commands as another service, making it easy to switch between them.

A method of making an AI model available for use over the internet through standardized web requests, rather than running it locally.

Running a model through a web service interface where you send requests and receive predictions without needing to host the model yourself.

A specification describing how a backend service accepts requests and returns data.

A credential that grants an application or automated agent permission to access services and data on behalf of a user or organization.

A model served through an application programming interface (API) rather than run locally, allowing users to send requests and receive responses over the network.

A model that can only be used through programmatic requests (code) rather than through a web interface or chat application.

Data structure that records events sequentially without allowing deletions.

Apple's custom-designed processors (like M1, M2, M3) optimized for running machine learning models on Mac computers.

Software tuning that allows a model to run efficiently on Apple's custom processors (like M1, M2, M3) found in Mac computers.

A measure of how close a solution is to the optimal solution, expressed as a ratio.

Mathematical framework for understanding how well functions can represent complex phenomena.

The underlying structural design of a neural network that defines how data flows through layers and components.

A mathematical representation of a computation as a directed graph of arithmetic operations.

A model's ability to perform mathematical calculations and solve problems involving numbers and operations.

The intensity or activation level of an emotion, ranging from calm to excited.

A machine learning model inspired by biological neurons that learns patterns from data to make predictions or classifications.

The task of identifying or classifying the artistic style of a work (e.g., Renaissance, Impressionism) using AI.

Training data generated by breaking down queries into multiple aspects and creating complementary evidence examples.

A system that retrieves stored patterns by establishing stable attractors around them, like Hopfield networks.

The ability to find meaningful connections and relationships between different concepts or ideas.

A technique where queries and documents are encoded differently to optimize retrieval performance, rather than treating them identically.

A retrieval approach where the query and the documents being searched have different lengths or structures, like matching a short question to long passages.

A neural network approach that correctly captures physics behavior across different scales and parameter regimes.

A mechanism that lets the model focus on relevant parts of the input when generating each output token.

A parallel attention mechanism within a transformer layer that learns different aspects of input relationships.

Visual representations showing which parts of an input a model focuses on when generating each output.

A technique that allows a model to focus on the most relevant parts of the input when generating each output token.

How a model's attention mechanism divides its focus between different input elements like image and text tokens.

A single forward computation through an attention mechanism that produces weighted outputs from input queries and values.

Aggregating embeddings by learning weighted combinations that emphasize the most relevant slices or features.

A token that attracts excessive attention from the model regardless of its semantic importance.

Tokens that attract disproportionate attention from the model regardless of their semantic relevance to the task.

Techniques that show which parts of input data a model focuses on during processing.

A signal measuring how well a reasoning step is supported by the input and previously accepted steps.

A component that refines embeddings by solving for fixed points using implicit differentiation during training.

Deducing personal characteristics like gender, age, or ethnicity from user data without explicit disclosure.

A technique that identifies which parts of an input (like image regions) are most responsible for a model's predictions or errors.

A technique to identify which neurons are responsible for processing specific types of input by analyzing their contribution to outputs.

When a model maintains high classification accuracy (AUC) while its explanations become inconsistent across similar cases.

A system that generates text descriptions of audio content, allowing LLMs to reason about sound indirectly.

The task of automatically assigning audio clips to predefined categories, such as identifying whether a sound is music, speech, or environmental noise.

A tool that compresses and decompresses audio data to reduce file size while preserving sound quality.

Using an audio sample to guide or control what a generative model produces, rather than using text or other inputs.

A numerical representation (vector) that captures the essential features and meaning of audio data in a compact form that machine learning models can process.

Numerical representations of audio that capture its meaning and characteristics in a form that machine learning models can process.

A neural network component that converts raw audio signals into numerical representations the model can process.

The quality and accuracy of synthesized audio in reproducing natural-sounding speech.

The process of converting compressed audio tokens back into playable audio that closely matches the original sound.

Converting spoken audio content into written text for analysis.

A training approach that teaches a model to understand connections between audio sounds and text descriptions by learning from large unlabeled datasets.

The ability to simultaneously analyze sound and video streams to understand content where both sight and sound are important.

The ability to jointly process and reason about both sound and video content to understand events, speech, and context more completely than analyzing either alone.

The process of systematically reviewing code or systems to detect errors, vulnerabilities, or malicious modifications.

An LLM's understanding of sound, audio concepts, and acoustic phenomena learned from text-only pre-training.

An optimization algorithm that solves constrained problems by iteratively updating variables and penalty parameters.

Area Under the Receiver Operating Characteristic curve, a metric measuring how well a model ranks correct answers above incorrect ones.

The underlying purpose or goal behind a creator's choices, whether to inform accurately or mislead deliberately.

Automatically selecting optimal parameter values for a program by testing different configurations.

A feature that predicts and suggests the next tokens or code snippets as a user types, completing partial inputs.

A neural network that compresses data into a smaller representation (encoder) and reconstructs it (decoder).

Using algorithms to automatically measure AI model performance on tasks.

Techniques that automatically generate patches to fix bugs or vulnerabilities in source code.

Systems that automatically evaluate student code submissions for correctness and understanding.

Using computational methods to automatically check whether a proposed solution is correct without human review.

Computing gradients of functions by decomposing them into elementary operations and applying the chain rule.

Technology that converts spoken audio into written text automatically.

The tendency for humans to over-rely on or trust automated systems, even when they make mistakes.

Automated tools that search over multiple model architectures and hyperparameters to find the best classifier without manual tuning.

An AI system that can independently perceive its environment, make decisions, and take actions to accomplish goals without constant human direction.

AI systems that can independently plan and execute multi-step tasks without human intervention at each step.

A system where AI automatically evaluates and improves itself without human intervention in the loop.

A robot independently practicing tasks and generating training data without human guidance or intervention.

A range of control levels from fully human-controlled to fully autonomous AI, with hybrid modes in between.

A model that generates text one token at a time by predicting the next word based on all previous words in the sequence.

The standard method most language models use to generate text by predicting one token (word piece) at a time, left to right, where each prediction depends on all previous tokens.

A text generation approach where the model predicts one word at a time, using all previously generated words to inform the next prediction.

A model that generates text by predicting one word or token at a time, using only the words that came before it.

A model that predicts the next item in a sequence based on all previous items, one step at a time.

Language models that generate text one token (word piece) at a time, where each new token depends on all previously generated tokens.

Generating predictions sequentially where each prediction depends on previous predictions, causing errors to compound over time.

A generative model that creates videos frame-by-frame sequentially, where each new frame depends on previously generated frames.

Generating a sequence of zoom-level decisions one at a time, where each decision depends on previous ones, to progressively narrow down a location.

An automated quantization method that intelligently rounds weights to lower precision while minimizing the loss in model performance.

Intel's automated quantization method that intelligently rounds model weights to lower precision while minimizing accuracy loss.

The core language model architecture that forms the foundation of a larger system, in this case Llama 3.

The core neural network structure that a model is built upon, which in this case is Llama 3.

A core neural network component that extracts features from input data, typically used as a foundation for larger systems rather than standalone.

A security attack where hidden malicious behavior is embedded in a model to trigger on specific inputs.

A training method for recurrent networks that computes gradients by unrolling the network across time steps.

Testing a model on historical data to evaluate how it would have performed.

Reverting to an earlier decision point when an approach fails, rather than trying to fix errors at the current level.

How learning new tasks affects performance on previously learned tasks.

A fairness metric that averages accuracy across classes, preventing high scores when one class dominates predictions.

Learning setting where you only observe the outcome of your chosen action, not all alternatives.

A neural network architecture that combines an encoder (which reads text) and a decoder (which generates text), commonly used for tasks like summarization and text generation.

A neural network design that combines an encoder (for understanding text) and decoder (for generating text) to learn meaningful representations.

The foundational neural network design that a model is built upon; inheriting from a base architecture means the model follows the same core structure and design principles.

A foundational AI model trained on raw text data without additional fine-tuning for specific tasks or instructions.

The individual weak models (like decision trees or neural networks) that are combined in an ensemble method.

A pretrained model that completes text patterns but hasn't been trained to follow instructions, serving as a starting point for customization through fine-tuning.

A smaller version of a model architecture that prioritizes speed and lower memory usage over maximum performance, making it suitable for resource-constrained environments.

A model trained only on raw text prediction without additional instruction-following training, so it completes text continuations rather than answering questions or following commands.

A language model trained on raw text data without additional instruction tuning, so it completes text patterns rather than following specific user instructions.

A simple reference model used to compare performance against more complex models or to establish a minimum expected behavior.

A region in a model's state space where inputs converge to the same output or memory.

Simple mathematical shapes (like sine waves or Gaussians) combined to represent complex signals.

Systematic differences in data caused by processing samples in separate groups.

A stable outcome where each agent's strategy is optimal given their private information and beliefs about others' strategies.

Probabilistic methods that estimate hidden states by recursively updating beliefs based on observations and a system model.

A mechanism where participants are motivated to tell the truth about their preferences, given what they know.

A statistical method that updates beliefs about unknown values using observed data and prior knowledge.

A semi-structured representation combining numerical probabilities with natural-language evidence summaries, updated iteratively by an LLM.

Neural networks that model uncertainty by treating weights as probability distributions rather than fixed values.

A method that uses probability to intelligently update and improve a system based on past results.

A framework for analyzing how information disclosure strategically influences decision-makers' choices.

Binary Cross-Entropy loss, a training objective commonly used for relevance scoring tasks where the model learns to predict whether a query-document pair is relevant or not.

A decoding algorithm that keeps the top-k most likely candidate sequences at each step, balancing quality and computational cost.

A mathematical model describing how honeybee swarms reach consensus on nest sites through recruitment and inhibition.

Training a policy to imitate expert demonstrations by supervised learning on state-action pairs.

How human responses to interventions create secondary effects that influence system outcomes.

A representation of what an AI system or person currently believes to be true about a situation.

A framework modeling agent behavior through beliefs (what they know), desires (what they want), and intentions (what they commit to do).

A mathematical operator that updates value estimates based on immediate rewards and future value predictions.

A standardized test suite used to measure and compare model performance on specific tasks.

A standardized set of test problems used to measure and compare the performance of different algorithms or models.

Comparing model safety when no labeled benchmark exists for the specific language, domain, or regulatory context.

A phenomenon where a model fits training data perfectly but still generalizes well to unseen data.

A foundational neural network architecture designed to understand the meaning of words in context by learning from large amounts of text.

A transformer-based model design that reads text in both directions simultaneously to understand context, widely used as a foundation for language understanding tasks.

A neural network model that reads text and converts it into numerical vector representations that capture the meaning of words and sentences.

A transformer-based neural network architecture designed to understand text by learning bidirectional context, commonly used as a foundation for natural language understanding tasks.

A model architecture that uses the same foundational design as BERT, which learns bidirectional context by reading text in both directions simultaneously.

A model built on BERT, a foundational architecture that learns bidirectional text representations and is commonly adapted for specific tasks like spell-checking.

A neural network design based on the BERT model that uses transformer layers to understand relationships between words in text by looking at context from all directions.

A transformer-based model architecture that reads text bidirectionally to understand context and produce meaningful representations of words and sentences.

A heavily compressed version of the BERT language model with far fewer parameters, designed for fast inference on resource-constrained devices.

A decoding strategy that generates N candidate responses and selects the one ranked highest by a reward model.

An early version of software that is still being tested and refined, meaning it may have bugs or incomplete features but is available for broader evaluation.

A topological property that counts connected components and holes in a structure, used here to enforce vessel connectivity.

A 16-bit floating-point format that balances precision and memory efficiency, commonly used for training and deploying large language models.

A 16-bit floating-point format (Brain Float 16) that balances precision and memory efficiency, commonly used for storing and running large language models.

A 16-bit numerical format that balances memory efficiency with numerical stability, using fewer bits than standard 32-bit floats while maintaining training and inference quality.

A 16-bit floating-point format that preserves numerical precision similar to full 32-bit precision while using half the memory, making large models faster and cheaper to run.

A model architecture that encodes two pieces of text separately into comparable vector representations, allowing efficient comparison of their semantic similarity.

An optimization approach with two nested loops: an inner loop optimizing fast weights and an outer loop optimizing the main model parameters.

Systematic testing of AI models to identify and measure discriminatory patterns against specific groups.

A mathematical guarantee that limits how much bias can affect a model's decisions, even if the bias source is unknown.

Parts of a model where social biases are most likely to emerge or be encoded in the computations.

Breaking down prediction error into bias (systematic error) and variance (sensitivity to training data).

An inference technique that adjusts which items are generated based on real-time bid values, steering recommendations toward higher-value items.

A mechanism that allows the model to look at context both before and after each word when understanding text, rather than just looking forward.

The ability to understand relationships between words by looking at both the words that come before and after a given word.

A neural network architecture that encodes two separate pieces of text independently and compares them to measure semantic similarity, commonly used for matching and retrieval tasks.

A large coefficient used in MILP formulations to enforce logical constraints; larger values make the relaxation weaker and solving slower.

A transformer-based model architecture designed to handle very long text sequences efficiently by using sparse attention patterns instead of processing every word pair.

An optimization framework with two hierarchical levels where upper-level decisions constrain lower-level optimization problems.

A factorization where value and policy functions are expressed as products of goal-conditioned coefficients and learned basis functions.

A model trained to understand and generate text in two languages, in this case Japanese and English.

A language model trained to understand and generate text in two languages with comparable fluency.

A recurrent neural network that processes text in both forward and backward directions to capture context from both sides of each word.

A model that processes two different types of input (in this case, code and natural language) and converts them into a shared representation space.

A decision mechanism where neurons act as on/off switches to direct data through different computational paths.

A large collection of medical and scientific texts (like research papers and journals) used to train the model on domain-specific language and concepts.

Natural language processing techniques applied specifically to medical and biological text, such as extracting drug names or identifying disease mentions from research papers.

The ability to understand and work with scientific concepts in biology and medicine, such as drug interactions and molecular structures.

Written content from medical and life sciences domains, including clinical notes, research papers, and healthcare documentation.

Specialized medical and scientific terms and concepts that the model has learned to understand from training on medical literature.

Protecting against misuse of biological research and AI in harmful ways.

Electrical or physical signals produced by the body, such as heart rhythms or brain waves.

A top-down 2D representation of a 3D scene, showing spatial layout as if viewed from above.

The mathematical space of all doubly stochastic matrices; parameterizing this space exactly is the core challenge this paper addresses.

The number of bits used to represent each number in a model; lower bit depths (like 3-bit) create smaller files but may lose some accuracy compared to higher bit depths.

The number of bits used to represent each number in a model; lower bit precision (like 3-bit) means smaller file size but potentially less accurate calculations.

The number of bits used to represent each number in a model; lower bit-widths (like 6-bit) use less memory but may reduce precision compared to higher bit-widths.

Automatically choosing the optimal number of bits for quantizing different parts of a model based on their importance.

A compression metric measuring how many bits are needed to encode each byte of text.

Evaluating a system's behavior by observing inputs and outputs without access to internal model structure or weights.

A measure of uncertainty in an agent's decision-making at a given state—how much of the decision space lacks statistical support from training data.

Assessment where evaluators don't know which version or source produced the item being judged.

An attention technique that processes groups of items together to improve efficiency and capture relationships between them.

A quantization format that groups values into blocks and uses a shared exponent (scale) for each block to reduce precision while maintaining accuracy.

Internal vector representations produced by a state space model's processing blocks that encode information about token sequences.

Scaling factors computed for groups of values in low-precision formats to maintain numerical accuracy.

A language model that generates multiple tokens in parallel using diffusion, then refines them iteratively.

A quantization method that divides values into groups and applies a shared scale factor to each group.

A ranking function that scores document relevance based on term frequency and document length normalization.

Movement commands relative to the drone's own orientation, rather than a fixed world direction.

Mechanisms that prevent an LLM from crossing defined limits in reasoning or behavior.

A rectangular coordinate set that marks the exact location and size of detected text or objects within an image.

A statistical model that ranks items based on pairwise comparison outcomes, commonly used for leaderboards.

Using AI to enhance the exploratory ideation phase of research rather than automating solution design.

The average number of possible moves available at each decision point in a game.

A marker in code where a debugger pauses execution so you can inspect the program state.

A metric measuring forecast accuracy that compares a model's predictions to a baseline (like random guessing).

A situation where the underlying physics has symmetry, but observations reveal a preferred direction or asymmetry due to measurement constraints.

A reinforcement learning technique that constrains model outputs to stay within a token budget, reducing response length while maintaining accuracy.

Breaking text into individual bytes (raw character codes) rather than words or subwords, which allows the model to handle any text without a predefined vocabulary.

The ability of a system to function correctly even when some participants behave maliciously or unpredictably.

Adjusting a model's predictions using held-out data to correct for systematic biases or distribution differences.

Determining the position and orientation of a camera in 3D space relative to a scene.

A statistical technique that finds the strongest correlations between two sets of variables by discovering shared patterns.

Training process designed to extract or develop specific abilities from a model, like reasoning or tool use.

How the number of storable associations grows with the size of the memory matrix or system parameters.

Forecasts of future returns, volatility, and correlations for different asset classes used to guide investment decisions.

Neural networks with capsule units that learn hierarchical relationships and spatial properties better than traditional convolutional layers.

A mechanism that sequentially combines information from multiple sources (global context, object details, skill knowledge) to guide model decisions.

Sequential processing where output from one stage feeds into the next.

A strategy where each model focuses on progressively smaller regions of interest to improve accuracy.

A multi-stage process that progressively assigns incidents to the correct business team or service owner.

The model's ability to distinguish between uppercase and lowercase letters as meaningful differences, treating 'Москва' and 'москва' as separate tokens with different meanings.

A model that treats uppercase and lowercase letters as identical, so 'Apple' and 'apple' are processed the same way.

The model treats uppercase and lowercase letters as distinct, allowing it to recognize proper nouns and maintain capitalization distinctions.

Text processing that preserves the distinction between uppercase and lowercase letters, treating 'Apple' and 'apple' as different tokens.

The model's ability to distinguish between uppercase and lowercase letters, making it sensitive to proper nouns and capitalization patterns that carry meaning.

When a model loses its original knowledge while learning a new task, like overwriting old skills.

A model that learns causal relationships between variables and can answer observational, interventional, and counterfactual questions.

The ability to determine true cause-and-effect relationships from data, typically guaranteed by randomization.

Determining whether a treatment actually caused an outcome, not just whether they're correlated.

Deliberately modifying a model's internal features to measure their direct effect on outputs.

A model that predicts the next word in a sequence by only looking at previous words, not future ones, making it suitable for text generation.

A training approach where the model predicts the next word based only on previous words, commonly used for text generation tasks.

Understanding cause-and-effect relationships rather than just statistical correlations in data.

A machine learning method that estimates personalized treatment effects from survival data using tree-based models.

A permissive open-source license that allows anyone to use, modify, and distribute the model as long as they give credit to the original creator.

A Creative Commons license that allows free use and modification of the model for non-commercial purposes only, with attribution required.

A problem-solving technique that starts with a simplified version of a problem and refines it when solutions fail.

A statistical phenomenon where most scores cluster near the maximum possible value, reducing the ability to distinguish between different quality levels.

When a benchmark becomes too easy and models achieve near-perfect scores, making it impossible to compare their true abilities.

Training an AI model to refuse or provide false information about certain topics.

Metric that measures structural similarity between representations by comparing their kernel matrices.

A statistical principle stating that the average of many independent samples approaches a normal distribution.

A reasoning technique where an AI model shows its step-by-step thinking process before arriving at a final answer, making its logic transparent and verifiable.

A technique where a model works through a problem step by step, showing its reasoning process before arriving at a final answer.

A quantum circuit composed of quantum channels (operations that map quantum states to quantum states) rather than unitary gates alone.

Raw wireless signal data that describes how a Wi-Fi signal changes as it travels through space and bounces off objects.

A learnable operation that scales and shifts different feature channels independently in a neural network.

Applying different forgetting rates to different feature channels in a neural network, allowing selective memory retention.

Systems where small changes in initial conditions lead to drastically different outcomes, making long-term prediction extremely difficult.

The ability of a model to maintain a character's voice, personality, and backstory throughout a conversation without contradicting itself.

A metric measuring the percentage of characters incorrectly recognized by an OCR system.

A model's ability to maintain distinct, consistent personality and speech patterns for different characters within a story.

Processing text one character at a time rather than by words, which is useful for catching individual character errors in languages like Chinese.

The ability to extract information from visual charts and perform logical reasoning tasks based on what the chart displays.

A language model specifically trained to have natural back-and-forth conversations with users rather than just completing text.

A model specifically trained and tuned to excel at conversational interactions rather than other tasks like analysis or reasoning.

A model optimized through training to excel at multi-turn conversations and dialogue, rather than single-turn text completion.

A saved snapshot of a model's weights and state at a specific point during training, allowing training to resume or the model to be evaluated at that stage.

Saved snapshots of a model at different stages of training, allowing researchers to study how the model's behavior changes as it learns.

Breaking long sequences into smaller segments and processing them sequentially while maintaining state between chunks.

The process of breaking large documents into smaller pieces so a model with a limited context window can process them separately.

A graph structure showing how research papers reference each other, used to understand relationships and influence between scientific works.

The ability to identify, reference, and maintain accurate attribution to the sources used when generating a response.

The number of insurance claims expected per policy or geographic area over a time period.

A technique that generates visual heatmaps showing which image regions a neural network uses to make predictions.

When training data has unequal numbers of examples across categories, with some classes having far fewer samples than others.

Learning to recognize new object classes over time while maintaining performance on previously seen classes.

Generating complete, structured classes with multiple methods and internal dependencies from a specification.

A loss function that penalizes misclassification of rare classes more heavily, useful when training data is imbalanced.

A statistical framework for designing and validating tests that measure psychological constructs reliably.

A machine learning model trained to assign input data into predefined categories or labels.

A technique that steers diffusion models toward desired outputs by comparing conditional and unconditional predictions.

How well an LLM's medical communication matches established clinical standards and physician practices.

The study of moral principles and values that guide medical decision-making and patient care.

Converting clinical information (diagnoses, medications, procedures) into discrete tokens that a model can process.

Using historical patient data to predict future health outcomes, disease progression, or treatment responses.

The ability to accurately interpret and reason about medical terminology, patient symptoms, and healthcare documentation.

Natural language processing applied to medical and healthcare text, such as extracting diagnoses or findings from doctor's notes and radiology reports.

The ability to analyze medical information, connect symptoms to conditions, and make logical healthcare decisions based on evidence.

The process of confirming that an AI system's outputs meet clinical standards and are safe for use in healthcare.

A model trained on image-text pairs to create shared vector representations for both images and text.

A neural network design that learns to match images and text by training them to have similar representations, enabling tasks like image search and visual understanding.

Rapidly adjusting a model to new tasks using direct mathematical solutions rather than iterative training.

A mathematical formula that directly computes an answer without iterative learning or optimization.

A system that continuously adjusts its behavior based on feedback from its actions and outcomes.

A control strategy where the robot observes its current state and adjusts actions based on feedback, rather than executing a fixed sequence.

When multiple features in a neural network are active at the same time, often because they represent related concepts.

An equilibrium where no group of players can jointly deviate and all benefit, even if they coordinate.

A game theory solution where no player benefits from unilaterally deviating from a recommended strategy.

A strategy that first captures broad patterns, then progressively refines details for better understanding.

A sequential decision-making approach that starts with broad estimates and progressively refines them to higher precision.

A curriculum learning approach that starts with learning simple components before progressing to optimizing complex global structures.

Identifying sections of code that perform the same function, even if written differently or in different programming languages.

The ability to automatically suggest or generate the next lines of code based on what the programmer has already written.

Percentage of program code executed by a test suite, measured by lines or branches.

A specialized task where a model modifies or refines existing code rather than creating new code, focusing on precision and surgical changes.

A specialized embedding designed specifically for source code that understands programming syntax and semantics, enabling tasks like code search and finding similar code snippets.

The ability of a model to write, complete, or suggest programming code based on prompts or partial code input.

Training a language model primarily on source code and technical documentation rather than general text, making it specialized for coding tasks.

A measure of how well code meets standards for readability, maintainability, and correctness.

The ability of a model to understand, analyze, and make logical inferences about source code and programming logic.

Restructuring existing code without changing its external behavior to improve readability and maintainability.

Process of examining code changes for bugs, quality issues, and adherence to standards before merging.

Automatically generating executable code (like plotting commands) from high-level specifications or natural language descriptions.

Techniques to confirm a student actually understands the code they wrote, not just copied it.

A language model specifically trained on programming code to excel at tasks like code generation, completion, and understanding.

A model trained with a focus on understanding and generating programming code across multiple languages.

A language model trained specifically on programming code and related tasks, optimized to understand and generate code better than general-purpose models.

A language model trained specifically on programming code and code-related tasks rather than general text.

The ability to naturally mix two languages within the same text or conversation, switching between them based on context rather than treating them as separate.

A lookup table mapping compressed values back to original data; avoided in this approach to save memory.

An AI system that autonomously writes, debugs, and executes code to solve tasks without human intervention.

A normalized measure of variability that expresses standard deviation as a percentage of the mean, useful for comparing spread across different scales.

Words in different languages that share a common historical origin and similar meaning.

Identifying words in different languages that share a common origin and similar meaning.

A computational framework that models how an intelligent agent perceives, reasons, and acts in the world.

A mechanism that gates speculative execution based on model confidence, without requiring ground-truth labels.

A psychological framework explaining how working memory capacity affects learning and task performance.

AI assistance that helps users think through problems and refine their goals rather than just executing stated requests.

The quality of maintaining consistent meaning and logical flow across multiple sentences or exchanges in a conversation.

The maximum circuit depth a quantum computer can execute before quantum information is lost to noise and decoherence.

A neural retrieval model design that stores multiple token-level embeddings per document and uses late interaction to achieve higher retrieval accuracy than single-vector approaches.

When a model trained with sparse rewards gets stuck early because initial success probability is too low to learn from.

When input features are highly correlated with each other, making it difficult to isolate individual feature effects on predictions.

Finding the best arrangement or selection from a finite set of possibilities, like packing objects efficiently.

Shared beliefs and mutually recognized facts that enable effective collaboration between people or AI systems.

The ability of a model to understand and apply everyday logic and practical knowledge about how the world works.

Minimizing the amount of data exchanged between devices or servers during distributed training.

A model variant created and shared by the community rather than the original model creators, often with custom modifications.

A smaller language model designed to use fewer computational resources while still performing useful tasks.

Evaluating reward quality relative to the current policy's skill level, recognizing that reward rankings change as the policy improves.

Natural language questions that define what an ontology should be able to answer, used to specify system requirements.

A quantum physics constraint ensuring operations preserve valid quantum states and probabilities.

A text generation approach where the model continues or completes text from a given prompt, rather than engaging in back-and-forth conversation.

A prompt style where you provide the beginning of text and the model continues it, rather than asking a direct question.

The ability to work through multi-step problems, analyze nuanced information, and draw logical conclusions.

Official verification that a service meets specific regulatory or security standards required by industries like healthcare or finance.

Official verifications that a service meets specific security and regulatory standards (like HIPAA or SOC 2) required by certain industries.

Discrimination that emerges from how separate system components work together, not from individual parts alone.

A design pattern where UIs are built from reusable, self-contained pieces (components) that can be combined to create larger interfaces.

Model's ability to understand new combinations of learned concepts.

Text descriptions that specify multiple elements, their relationships, and spatial arrangements in the desired image.

The principle that the meaning of a complex expression is built from the meanings of its parts and how they combine.

The ability to understand new combinations of concepts by learning how individual components combine.

The amount of processing power and memory available to run a model, which determines how much computation can be performed.

The amount of computation (time and memory) required for an algorithm to solve a problem.

The ability to deliver good results while using less processing power and memory than larger models.

Using code and algorithms to test mathematical hypotheses and discover patterns empirically.

The amount of memory, processing power, and time required to run a model; a smaller footprint means the model can run on less powerful hardware.

The extra processing power, memory, or time required to run a model, which impacts speed and resource consumption.

Using algorithms and AI during image capture to enhance photos beyond what the camera sensor alone can achieve.

The strategic distribution of a model's processing power—in this case, spending more computational effort on thinking through problems rather than other tasks.

How well a model performs relative to the computational resources (processing power and memory) required to run it.

A model designed to run with minimal processing power and memory, making it practical for devices with limited resources.

Hardware architecture that performs computation directly within memory, reducing data movement bottlenecks.

Achieving the best performance for a given amount of computational resources.

The ability for an AI model to interact with computer interfaces, navigate software applications, and execute actions on a user's behalf by understanding and responding to visual or textual representations of screens.

An interpretable model that makes predictions by routing inputs through a layer of human-understandable concepts rather than opaque features.

A low-dimensional geometric structure where related concepts are organized continuously, like a curved surface in high-dimensional space.

The process of mapping different textual expressions of the same idea to a single standardized representation, such as mapping 'MI' and 'myocardial infarction' to the same medical concept.

A reinforcement learning technique that estimates action value only within trajectories meeting specific conditions.

A property where prediction set coverage guarantees hold for specific subgroups or conditions, not just on average across all data.

A measure of uncertainty in predicted tokens given context; low entropy signals memorization, high entropy signals generalization.

The average distance between selected and validation target embeddings within a cluster, used to rank training examples.

The ability of a model to generate output (like text) based on specific input conditions or prompts provided to it.

The expected value of an output given specific input conditions, used as a deterministic baseline prediction.

Misaligned behavior that only appears when inputs share features with the training data, while appearing safe on out-of-distribution prompts.

A probabilistic model that learns to make predictions by conditioning on observed examples, useful for few-shot learning and uncertainty estimation.

The ability to generate text that follows specific conditions or constraints, rather than producing output freely.

A risk metric that focuses on the worst-case outcomes rather than average performance, useful for safety-critical tasks.

A neural network that learns to generate new data matching specific conditions or constraints.

Guiding a generative model's output by providing additional input signals like pose or depth maps.

Ensuring a model's confidence scores accurately reflect its true probability of being correct.

Assigning uncertainty scores to model predictions to identify outputs that may need human verification.

Statistical bounds around predictions that quantify uncertainty; here used to identify when model predictions are unreliable.

A decoding strategy that stops refining tokens when model confidence exceeds a set threshold.

Refusing predictions when the model's confidence score is below a threshold.

A strategy that selects which tokens to generate next based on the model's prediction confidence, enabling adaptive and efficient generation.

Training a model using rewards based on how well its confidence scores match its actual correctness.

Weighted majority voting where each candidate answer gets a confidence score from a critic model before selection.

The tendency to seek or interpret information in ways that confirm existing beliefs or outputs.

Situations where an AI system has competing goals—like serving users well versus generating revenue for its creators.

Method providing prediction intervals with statistical guarantees on coverage.

The ability to direct a model to generate specific 3D shapes or structural states of proteins.

A distinct 3D shape or arrangement that a protein can adopt, often with different biological functions.

Applying a learned conformational change from one protein to structurally similar proteins in the same family.

When an agent misuses its elevated permissions to perform actions it shouldn't, tricked by user input.

A routing pattern where multiple neurons must agree (be mutually exclusive) to activate a particular processing path.

A safety training approach that guides a model to behave according to a set of principles or rules, helping it generate more helpful and harmless responses.

Restricting a model's token generation to a predefined set of allowed tokens during inference.

Text generation that must follow specific rules or constraints, such as producing output in a particular format or structure.

Training an AI system to maximize performance while respecting hard constraints (like deadlines or budgets).

Finding solutions that satisfy a set of constraints, used here to resolve conflicts between inferred events.

A tool that finds valid solutions to problems with multiple constraints, used here to verify mechanical assembly feasibility.

Validating each step of a plan by checking outputs against automatically derived constraints based on task requirements.

Fixing errors in reasoning by making minimal changes that satisfy logical or evidential constraints.

Whether a study actually measures the real concept it's supposed to test, not something else.

A mechanism that activates learned corrections only when the robot is physically touching the object.

Physical interactions where the robot frequently touches and manipulates objects, making control sensitive to small errors.

Robot tasks where success depends critically on precise control of forces and contact interactions with objects.

A model or system that screens text before or after generation to block unsafe, harmful, or policy-violating content.

Safety mechanisms built into a model that prevent it from generating harmful, inappropriate, or restricted content.

The process of reviewing and filtering text or other content to remove or flag material that violates policies or safety guidelines.

The task of automatically detecting and categorizing text that violates policies or could cause harm, such as hate speech, violence, or misinformation.

The ability to maintain consistent meaning and logical flow when processing long sequences of text or conversation.

A model's ability to maintain coherent understanding and recall of information across long passages of text without contradicting itself.

Transferring knowledge from interaction trajectories into model parameters by learning from contextual examples.

Retaining only relevant information from execution history to reduce noise and improve decision-making in subsequent steps.

The process of collecting and organizing relevant information from history to answer specific questions or solve tasks.

The maximum amount of previous text a model can consider when generating its next output; longer context allows the model to maintain coherence over longer passages.

Organizing and maintaining relevant information for AI decision-making.

A technique to process long sequences by distributing context across multiple devices or processing units in parallel.

Irrelevant or noisy information degrading model performance in a given context.

A model's ability to remember and use information from earlier parts of a conversation or document.

When an AI model's input context window fills up and earlier information is lost, requiring mechanisms to preserve key data.

The maximum number of tokens a model can process in a single conversation or prompt.

A system that adjusts its behavior based on the specific input or situation rather than using fixed, unchanging patterns.

Speech recognition that uses surrounding information like conversation history to improve transcription accuracy.

A formal system of rules that defines which sequences of symbols are valid in a language.

Problems requiring the model to extract and use large amounts of information from the input prompt to generate correct outputs.

Help or instructions tailored to the current situation rather than generic pre-stored information.

Adjusting model behavior dynamically based on the specific input or context rather than using fixed settings.

A learning algorithm that selects actions based on context and learns from feedback to improve future decisions.

Numerical representations of text that capture meaning based on surrounding context, rather than treating each word independently.

The assumption that a model produces consistent outputs when a task is reformulated in contextually equivalent ways.

Influence from surrounding information (like examples or previous actions) that pushes an agent away from its intended behavior.

Making decisions by considering how individual observations relate to and inform each other within a broader context.

A way of encoding text where the meaning of each word depends on the words around it, rather than being fixed for every occurrence.

The intermediate representation space in a diffusion model where semantic and structural information is encoded.

Machine learning technique that identifies recurring themes in text while considering the surrounding context of words.

A feature or pattern in input text that activates hidden misaligned behavior in a model, even when standard evaluations show the model is safe.

Uncertainty caused by changing conditions over time, like user preferences shifting.

The ability of a model to interpret the meaning of words and phrases based on surrounding text, rather than treating each word in isolation.

Vector representations of words that change based on surrounding context, capturing different meanings in different sentences.

Incrementally updating a neural network on new data as it arrives, rather than retraining from scratch.

Training models to learn new tasks without forgetting previously learned ones.

Further training a pretrained model on domain-specific data to specialize it for particular tasks.

Real-time monitoring of a quantum system that produces a stream of measurement data used to update state estimates.

Encoding data as smooth, unquantized values rather than discrete tokens, preserving fine-grained temporal details.

A mathematical property ensuring a system's outputs converge to a stable state regardless of initial conditions.

A mathematical property ensuring that a system brings nearby states closer together over time, guaranteeing stability.

A training technique that learns by comparing similar and dissimilar examples to create better representations.

Training objective that pulls similar examples together and pushes different ones apart.

A method that learns shared embedding spaces by contrasting similar and dissimilar image pairs, then ranks candidates by similarity.

Breaking down a neural network's output into individual contributions from different neurons or neuron groups.

Special tokens added at the beginning of a prompt that tell the model what style, domain, or format to use for its output.

Special tokens inserted into sequences to guide model behavior, such as signaling whether to show an ad or organic content.

Automatically designing a decision-making system that controls when and how to execute actions.

A technique that adds spatial control to diffusion models by conditioning generation on aligned input maps (like depth or property masks).

Physics problems where fluid flow effects dominate over diffusion, creating sharp gradients and moving fronts.

The point during training when a model's performance stabilizes and stops improving significantly, indicating it has learned the patterns in the data.

Mathematical proofs that an algorithm will reach a correct solution under specified conditions.

How quickly an optimization algorithm approaches the optimal solution, typically expressed as a function of iterations.

When different models independently learn similar features or representations from different training signals.

AI systems designed to understand and respond to human language in natural, dialogue-like interactions.

An AI system designed to conduct multi-turn dialogue with users to accomplish specific tasks, like medical interviewing.

Using dialogue with a chatbot or AI agent to probe and verify student understanding through questioning.

The model's ability to maintain logical consistency and relevance across multiple turns of dialogue, making responses feel natural and connected.

How naturally and coherently a model engages in back-and-forth dialogue, matching human conversation patterns.

A model specifically trained to understand and generate natural dialogue, optimized for back-and-forth interactions rather than one-off text generation.

A language model specifically trained and optimized to engage in multi-turn dialogue with users.

A function where any line segment between two points on the curve lies above the curve, ensuring a single global minimum.

Mathematical technique for finding the best solution to a problem with a single global optimum.

A geometric shape formed by the intersection of linear inequalities, with vertices representing extreme points.

A technique that scans across input data using small filters to detect local patterns, commonly used in image processing but here applied to text for efficiency.

The geographic coordinate system (e.g., latitude/longitude) used to define spatial locations and ensure consistency across operations.

Game theory scenarios where agents benefit from matching actions but may also benefit from strategic differentiation.

An optimization where data is only copied when modified, allowing multiple references to share the same data until changes occur.

An attention mechanism where peripheral tokens (patches) interact only through central core tokens, reducing computation.

Identifying when different mentions in text refer to the same entity or concept.

A collection of documents or text used as the knowledge base for retrieval in RAG systems.

Selecting query terms that best distinguish relevant documents from irrelevant ones in a specific corpus.

A filtering mechanism that validates whether a proposed solution is correct before allowing it to advance in a search process.

A model's ability to maintain performance when input data is degraded (e.g., noise, blur, missing values).

A mathematical measure that compares how similar two embeddings are by calculating the angle between them, with values closer to 1 meaning more similar.

A method of comparing two vectors based only on their direction, ignoring their magnitude, making it scale-invariant.

A learnable activation function using cosine waves with adjustable frequency and phase to process data nonlinearly.

An adversarial attack that accounts for the real-world cost or feasibility of modifying each feature.

The ability to deliver useful results while using fewer computational resources, reducing the expense of running the model.

A training methodology that combines chain-of-thought reasoning with masked autoencoder techniques to improve model understanding of text relationships.

Testing what would happen if you changed a strategy, without actually running the experiment in the real world.

An explanation showing what input changes would alter a model's prediction to a different outcome.

Creating alternative scenarios showing what would happen if something were different (e.g., if an object didn't exist).

Training examples where evidence is semantically related but contradicts the claim, testing if models truly use evidence.

A question about what would have happened if a variable had taken a different value (e.g., 'what if the patient had received treatment?').

Reasoning about what would have happened under different actions or conditions than what actually occurred.

The process of learning or updating the statistical properties of measurement and process noise in a filtering system.

Aligning a model's sensitivity structure to match the statistical structure of task-irrelevant variations in data.

When the distribution of input data changes between training and real-world use, causing models to fail.

Measuring what proportion of a problem space a model can reliably handle.

Finding an efficient route for a vehicle to visit all cells or areas in a region.

The process of proving that testing has comprehensively covered all relevant operating conditions and edge cases.

Testing approach that systematically explores different input regions to find edge cases and failures.

A quantum operation that preserves physical validity by maintaining positivity and trace properties of quantum states.

Running a model's predictions using a computer's central processor rather than a specialized graphics card, which is slower but requires less specialized hardware.

A measure of how useful and novel the connections a model generates are for creative tasks.

The process of determining which actions or steps in a sequence deserve reward or blame for the final outcome.

Detailed feedback that scores responses across multiple specific evaluation criteria rather than a single overall score.

An agent that reviews and validates the recommendations and execution plan of other agents to ensure correctness and coherence.

Mechanism allowing one sequence to attend to and focus on another sequence.

Transferring knowledge between models with fundamentally different designs, attention mechanisms, or tokenizers.

A neural module that merges information from two sources by learning which parts of each are most relevant.

Testing whether a model trained on one dataset generalizes to perform the same task on a different dataset.

A creativity technique where ideas from one unrelated domain are applied to solve problems in another domain.

Learning to control one body type (like a humanoid robot) using data from a different body type (like humans).

A model architecture that takes a query and document together as input and directly outputs a relevance score, unlike dual-encoders that score them separately.

A loss function that measures how well a predicted probability distribution matches a target distribution.

Running an AI model in different network environments or systems than the one it was trained on.

The ability to understand relationships and transfer knowledge between different languages, such as answering a question in one language based on text in another.

The ability of a model to understand and relate concepts across different languages, allowing it to find similarities between text in different languages.

The ability of a model to understand and work with multiple languages, sometimes even translating concepts between them.

The ability of a model to represent similar meanings in different languages as nearby points in its vector space, so translations and equivalent concepts are treated as semantically close.

The ability of a model or probe trained on one language to work effectively on other languages.

The ability to find and compare similar content across different languages by representing them in a shared mathematical space.

The ability to find relevant documents or text in one language when searching with a query in a different language.

The ability to recognize that sentences or phrases in different languages have the same or similar meaning and represent them close together in numerical space.

The ability to measure how similar two sentences are even when they are written in different languages.

The ability of a model trained on multiple languages to apply knowledge learned from one language to understand or generate text in another language.

The ability of a model to comprehend relationships and meanings across different languages, enabling tasks like translation and multilingual reasoning.

Connecting representations from different types of data (like speech and text) so they work together effectively.

An attack that manipulates multiple input types (like images and text) together to deceive a model.

A mechanism that aligns and weights information between different modalities like images and text.

Ensuring that representations across different modalities (images, 3D, text) align and reinforce each other.

Alignment in how models from different modalities (e.g., vision and language) represent the same stimulus.

The process of combining information from multiple modalities (e.g., vision and text) into a unified representation.

When a model produces contradictory predictions for the same concept represented in different modalities.

The ability to find relationships between different types of content, such as matching natural language descriptions to code snippets.

The ability to connect and reason about information from different input types (like audio and video) together to draw conclusions.

The ability to search and find relevant items across different data types, such as finding images using text queries or vice versa.

The ability of a model to share semantic understanding between different input modalities like vision and text.

The ability to measure how closely related content from different types of input (like images and text) are to each other.

Exchanging information between different input types (text and vision) to guide compression decisions.

The ability of AI tools to access information from other modules and make decisions based on shared context.

How the demand for one product changes when the price of a different product changes.

The ability of a model to perform consistently when input text or audio switches between different writing systems or languages.

The process of comparing and resolving conflicting information from multiple sources to determine accurate answers.

A model's ability to work on new individuals without retraining, despite differences in neural anatomy.

A mechanism that transfers motion information from one camera viewpoint to another while maintaining consistency.

The degree to which internal representations align when processing the same task in different formats or modalities.

Aligning images captured from different viewpoints (e.g., street-level and overhead) to find correspondences.

A 3-dimensional algebraic variety defined by a degree-3 polynomial equation.

NVIDIA's parallel computing platform that runs code on GPUs to process many tasks simultaneously.

Optimized GPU code that performs specific computational operations efficiently.

The ability to understand and infer cultural context, significance, and metadata from visual or textual information.

Statistical measures that describe probability distributions, used to track activation behavior.

Training data that has been carefully selected and filtered to include only high-quality examples relevant to specific tasks or domains.

Carefully selected and filtered training examples chosen for quality rather than quantity, often resulting in models that produce more structured and reliable outputs.

RL approach where agents explore by seeking states where their world model makes poor predictions.

Training strategy that gradually increases task difficulty to help models learn robustly.

Training strategy that presents examples in increasing order of difficulty.

A training constraint that penalizes curved or winding paths in the learned representation space.

A constraint requiring a model to reconstruct its original output after transforming through intermediate steps.

A metric measuring how many different paths code can take; lower values mean simpler, easier-to-maintain code.

An interactive learning method where a human corrects the model's mistakes during training to fix distribution mismatch.

Measuring how much each training example contributes to a model's final performance using gradient-based methods.

When test data accidentally leaks into training, artificially inflating a model's measured performance.

The process of carefully selecting, cleaning, and organizing training data to improve model quality; better curated data often leads to better model performance.

Predicting how a model would behave if specific training examples were excluded without retraining.

Predicting how a model's behavior would change if specific training data were excluded without retraining.

Variation in data distribution across different sources or groups.

Gaps or missing values in a dataset caused by sensor failures, blinks, or other interruptions.

The relevance, accuracy, and usefulness of training data, which can be more important for model performance than simply having more data.

The practice of carefully selecting and filtering training data for relevance and accuracy rather than simply using larger amounts of raw data.

A centralized catalog storing metadata about available data sources and their query interfaces.

A guarantee that your data is stored and processed only in a specific geographic region, helping meet regulatory requirements.

When researchers use datasets from previous studies in their own research rather than collecting new data.

Choosing a subset of training data based on quality or relevance metrics rather than using all available data.

Automatically generating training data from existing datasets to teach models new tasks.

Automated checks that verify data meets quality and correctness requirements before use.

Distributing training data across multiple GPUs that compute gradients independently then synchronize.

Compressing a large dataset into a smaller synthetic version preserving key information.

A neural network design pattern that serves as the structural foundation for this model, determining how it processes and generates text.

Creating entirely new protein sequences from scratch rather than modifying or copying existing ones.

A transformer-based language model architecture that uses disentangled attention mechanisms to improve how the model weighs different parts of the input text when making predictions.

A training approach where a model is developed across multiple independent computers or organizations rather than in a single centralized facility, allowing distributed collaboration.

A mechanism that selects actions based on current state, goals, and expected outcomes to maximize success.

A tool or system that provides information and analysis to help humans make better decisions without replacing human judgment.

An approach that evaluates systems based on the quality of decisions they enable under different costs and benefits.

A component that converts compressed internal representations back into human-readable outputs like audio or images.

A type of LLM that generates text one token at a time, like GPT models.

Language model design that generates text sequentially without a separate encoder, like GPT models.

Converting model outputs into human-readable text or structured predictions.

Methods for generating text from a language model, such as greedy selection, beam search, or temperature sampling.

A reward system that breaks complex requests into atomic, checkable questions to provide interpretable feedback for model training.

The process of removing duplicate or near-duplicate examples from training data to improve model efficiency and prevent overfitting to repeated content.

An AI system that performs multi-step research by reasoning through problems and making multiple search queries.

Automatically identifying problems or errors in software artifacts, such as incomplete or ambiguous descriptions.

The number of independent ways a mechanical part can move or rotate in an assembly.

Consequences of an agent's actions that appear many steps later, making it harder to learn cause-and-effect relationships.

Feedback or verification of agent actions that arrives after a delay, requiring the agent to maintain accountability over time.

A form of democracy where citizens and representatives engage in reasoned discussion to reach decisions.

Using machine learning to predict how much of a product customers will buy given prices and other factors.

Removing or hiding demographic information (like gender) from model inputs to reduce bias in decision-making.

Learning which demographic attributes (race, age, etc.) are most influential in predicting how annotators will judge subjective content.

Training examples collected from real robots performing tasks, used to teach the model how to execute similar actions.

A training approach where the model learns to reconstruct clean audio from corrupted or noisy versions, improving its ability to extract meaningful features.

A neural network trained to reconstruct clean text from corrupted or noisy versions, learning to remove noise while preserving meaning.

A training approach where a model learns to reconstruct clean audio from noisy versions, making it better at understanding speech in real-world conditions.

A technique where a model learns to gradually remove random noise from data to reconstruct meaningful content, used as an alternative to traditional token prediction.

A training objective that learns to predict noise in corrupted data, used in diffusion models for stable gradient-based optimization.

Generating detailed, comprehensive descriptions of images that capture rich visual information and relationships rather than brief summaries.

A compact vector representation where most dimensions contain meaningful information, as opposed to sparse embeddings that are mostly zeros.

Vector representations where most or all of the numbers contain meaningful information, as opposed to sparse embeddings where most numbers are zero.

A neural network where all parameters are active for every input, in contrast to sparse architectures like mixture-of-experts that selectively activate different parts.

A technique that converts documents and queries into dense vectors so that relevant passages can be found by comparing their numerical representations rather than matching keywords.

A compact numerical format where meaning is captured in a fixed-size list of numbers, making it efficient for storage and similarity comparisons.

A search method that converts text into a single, compact numerical vector and finds similar documents by comparing these vectors.

A retrieval system using learned embeddings to find semantically similar documents via vector similarity.

A compact numerical representation where most values are non-zero, used to efficiently store and compare the meaning of text.

A compact numerical representation of text that captures its meaning, allowing the model to compare how similar different pieces of text are to each other.

Numerical representations of text where each word or sentence is converted into a list of numbers that capture its meaning, allowing the model to compare semantic similarity.

A compact numerical format where text is encoded as a list of numbers that capture its meaning, allowing efficient similarity comparisons.

A mathematical space where text is represented as vectors of numbers, positioned so that similar meanings are located close together.

Compact numerical representations where most values are non-zero, used to encode the meaning of text in a form that computers can compare mathematically.

Dense embeddings use all dimensions with non-zero values (like traditional neural embeddings), while sparse embeddings mostly contain zeros and are more interpretable and storage-efficient.

A method that aligns models by learning from the geometric clustering of accepted responses in the model's representation space.

The ability to understand how multiple facts relate to and affect each other when making decisions.

Risk that a single model's values or biases get applied uniformly at scale, eliminating the diversity of perspectives that would naturally exist with multiple decision-makers.

An image where each pixel's brightness represents how far away that object is from the camera.

A technique that creates a larger model by combining and stitching together layers from smaller pre-trained models rather than training a new model from scratch.

The process of restoring a compressed model's weights to higher numerical precision, improving quality but requiring more memory.

A new model created by modifying or fine-tuning an existing base model rather than training from scratch.

A numerical representation that captures the visual characteristics around a detected keypoint, allowing the model to match similar points across different images.

Generating model weights using text or structured descriptions of the target architecture and task as input.

The loss of professional expertise and judgment that occurs when workers rely on automated systems instead of developing their own capabilities.

A mathematical model that generates diverse sets of items by penalizing similarity, useful for ensuring variety in generated outputs.

Automated verification rules that produce the same result every time, used when there is clear evidence of task completion.

An early, pre-release version of a model used for testing and refinement before public release.

Fine-grained, skillful robotic hand control requiring precise coordination of many joints.

AI process of identifying root causes or problems from observed symptoms.

The process of an AI model creating natural conversational responses based on input text.

The process of finding a set of basis vectors (dictionary) that can reconstruct data through sparse combinations.

The ability to understand and apply code changes (diffs) to existing files rather than generating code from scratch.

A property of operations that allows gradients to flow through them during backpropagation for model training.

Smooth mathematical function approximating non-differentiable operations for training.

Mathematical functions that measure how far a model's output is from desired behavior, designed to be optimizable via gradient descent.

A learnable memory retrieval mechanism that can be trained end-to-end to recall relevant past episodes for current decision-making.

A physics solver built into a neural network so that gradients can flow through physical laws during training.

A reward function whose gradients can be computed, allowing optimization of model outputs toward desired properties.

A sparse attention method that supports gradient computation, enabling end-to-end training with learned sparsity patterns.

A list of possible medical conditions ranked by likelihood, used by clinicians to guide further testing.

A mathematical framework that adds controlled noise to data to protect individual privacy while enabling statistical analysis.

A technique to systematically increase problem complexity to better differentiate model capabilities.

Predicting how hard a task is to automatically adjust the amount of computational effort needed.

An internal indicator that estimates how hard a problem is, used to guide model behavior.

Language models that generate text by iteratively refining noisy predictions into coherent words.

Generative model that creates images or videos by gradually removing noise from random data.

AI models that generate images by learning to reverse a noise-adding process, starting from pure noise.

A generative approach that iteratively refines predictions by gradually removing noise from random initial states.

A learned distribution that guides diffusion models toward realistic outputs in a specific domain.

A generation method that iteratively refines outputs by gradually removing noise, rather than predicting tokens one at a time from left to right.

Iterations in a diffusion model that gradually refine noise into a final image or video output.

A transformer architecture adapted to work with diffusion-based generation processes.

A neural network design that generates outputs by iteratively refining noisy predictions into clear results, rather than building text one token at a time like traditional language models.

A method where a model generates text by iteratively refining noise into coherent output all at once, rather than predicting one word at a time.

A language model that generates text by iteratively predicting and refining masked (hidden) tokens across the entire output, rather than predicting one token at a time from left to right.

Using diffusion models to generate realistic robot motion sequences that can be used as training data.

A virtual simulation model of a physical system used to predict behavior and test changes before real-world deployment.

A convolutional operation that skips input elements to capture patterns at multiple scales without increasing parameters.

A technique to simplify high-dimensional parameter spaces by identifying and focusing on the most critical variables.

Techniques that compress high-dimensional data into fewer dimensions while preserving important patterns.

A training technique that teaches a model to prefer certain outputs over others by learning from examples of better and worse responses.

A graph structure representing causal relationships where arrows point from causes to effects with no cycles.

A workflow representation where tasks are nodes and dependencies are directed edges with no circular paths.

A measure of smoothness on a graph that quantifies how much node values vary across connected edges.

The logical flow and consistency of ideas across sentences in a text or conversation.

Examining how language serves specific communicative purposes in conversation, like validating feelings or paraphrasing.

The challenge of moving from discovering causal rules to engineering them into working systems.

A generative model that iteratively removes noise from discrete tokens (like words) to generate text, as an alternative to autoregressive decoding.

Generative models that iteratively denoise discrete tokens (like words) from noise to produce text.

Compressed representations of audio data stored as specific, distinct values rather than continuous numbers, making them efficient for storage and processing.

A compressed representation where continuous data is converted into distinct, countable tokens or categories.

A communication channel where each transmitted symbol is corrupted independently with no memory of past transmissions.

Individual units of quantized information that represent audio in a compressed, symbolic form rather than continuous values.

Converting continuous numerical values into discrete bins or categories for processing by algorithms.

The ability of a model to generalize across different mesh resolutions or numerical discretizations of the same continuous problem.

A pattern in token gradients that effectively distinguishes high-reward responses from low-reward ones.

Separating different factors of variation (like expression and identity) in a model's learned representations.

A smaller, faster version of BERT that retains most of its language understanding ability while using fewer parameters and less computational power.

A technique that compresses a large, complex model into a smaller one by training it to mimic the larger model's behavior, resulting in faster inference with minimal loss of quality.

A model that has been compressed by training a smaller model to mimic a larger, more capable model, reducing size and computational requirements while retaining performance.

A smaller, faster version of a larger model created by training it to mimic the larger model's behavior, reducing computational requirements while maintaining reasonable performance.

Using multiple computers or servers across a network to share the computational work of training or running a model, rather than relying on a single machine.

When the data distribution used for training differs from the distribution encountered during deployment, causing performance degradation.

Modifying a model's output probability distribution at inference time to satisfy constraints without changing the model's weights.

When a policy becomes overly specialized in reproducing successful behaviors without learning to handle diverse situations or recover from failures.

When a model encounters data that looks different from what it was trained on, causing performance to drop.

When a model's behavior diverges from the original training data distribution during fine-tuning or RL.

A mathematical space where words are represented as vectors based on their usage patterns in text, like GloVe or Word2Vec.

Ensuring benefits and harms are equitably distributed across agents rather than concentrated in hubs or privileged positions.

Forcing a model's output distribution to match a target distribution, here used to normalize reward structures across different tasks.

Learning to predict probability distributions over outputs rather than single deterministic predictions.

When the statistical properties of data change over time, making old patterns unreliable for future predictions.

A regularization technique that limits how far a model's distribution can drift from a reference distribution during training.

A mathematical property ensuring that a velocity field conserves mass (no fluid is created or destroyed at any point).

A metric measuring the quality of unique answers generated relative to the best possible answer set of the same size.

A ranking method that prioritizes both relevance and variety, ensuring results cover different perspectives or approaches.

The ability to identify and explain which retrieved documents contributed to a generated answer.

The natural division between separate documents used as a constraint to group tokens for shared expert selection.

The process of breaking long documents into smaller pieces before embedding them, which this model is optimized to work with effectively.

Anchoring AI responses to specific source documents to ensure answers are based on provided content.

The ability to automatically extract, understand, and convert information from document images (like scans or forms) into structured, machine-readable formats.

The process of identifying and understanding the structure of a document, such as text regions, tables, and columns.

The process of automatically reading and extracting structured information like text, tables, and layout from documents.

Finding the relevant documents or passages from a large collection that are needed to answer a question.

The ability to maintain the original layout, formatting, and organization of a document when extracting text, rather than just outputting raw characters.

The ability to read and extract meaningful information from structured documents like receipts, invoices, and forms by recognizing both text and layout.

Tasks that require processing, searching, and reasoning over large collections of documents to find answers.

Understanding and answering questions that require information from multiple parts of a full document.

Training a model on data from multiple specialized fields (like general text, scientific papers, and medical literature) so it works well across all of them.

A specialized expert in an MoE model trained to handle reasoning or task-specific knowledge rather than raw perception.

Training models to work well on new, unseen domains beyond their training data.

A technique that automatically creates many fake domain names to evade detection and maintain control of malicious infrastructure.

How well a model's responses are anchored in accurate, specialized knowledge specific to a field rather than generic or hallucinated information.

Specialized expertise and facts about a particular field or subject area that an AI model has learned during training.

When a model encounters data from a different source or environment than it was trained on, causing performance to drop.

When a model is trained to excel at a specific task or set of languages rather than being a general-purpose tool.

Programming languages designed for specialized tasks in particular industries or fields.

A model that works effectively across many different subject areas and use cases without needing to be retrained for each one.

Abstract problem formulations that can be recognized and solved across multiple unrelated academic fields.

A model's ability to understand and respond accurately to topics within a specific field or area of expertise it was trained on.

An AI planning algorithm that solves problems in any domain without domain-specific customization.

A model trained specifically on data and tasks from a particular field (in this case, chemistry) to achieve higher accuracy in that domain than general-purpose models.

Tailored or optimized for a particular field or type of content, such as news, reviews, or scientific writing.

Assessment tailored to a particular field (like law) using metrics and error types relevant to that domain.

Training a model on specialized data from a particular field (like medicine) so it becomes expert at tasks in that domain rather than being a generalist.

The ability to generate text tailored to a particular field or context, such as legal documents, Wikipedia articles, or product reviews.

Specialized expertise required for a particular field, like vendor-specific scanner operations in medical imaging.

Specialized vocabulary and terminology unique to a particular field or industry, like medical jargon in healthcare or mathematical notation in physics.

A language model trained exclusively on text from a particular field or subject area, making it much better at understanding and generating content in that domain than general-purpose models.

A language model trained specifically on data from one field (like biomedical research) rather than general internet text, making it excel at specialized tasks.

Training a model to excel at tasks within a particular field (like legal documents) rather than being a general-purpose model.

Training a model on specialized data from a particular field (like biomedical literature) rather than general internet text, making it much better at understanding that field's concepts.

Specialized workflows and methodologies unique to a particular field that require expert knowledge to execute correctly.

Training a model exclusively on data from a narrow domain (like Python code) rather than general text, making it highly specialized but less versatile.

Training or adapting a model to specialize in a particular field (like biomedicine) rather than performing equally well across all topics.

A fine-tuning method that adapts model weights by separately learning magnitude and direction changes, extending LoRA.

A method of comparing two vectors by multiplying their components and summing the results, where vector magnitudes (length) affect the final score.

A square matrix where all rows and columns sum to 1, used to represent valid probability distributions for mixing multiple streams.

Reducing an image's resolution by removing pixels, making it smaller and faster to process.

A specialized AI model that receives requests routed to it by another system and performs the actual task or generates the final response.

Specific applications or problems that use the output of a pretrained model, such as predicting protein structure or identifying protein function.

The smaller neural network component in speculative decoding that quickly generates candidate tokens before verification by the main model.

A smaller, faster model used in speculative decoding to quickly propose token sequences before a larger model verifies them.

A tree structure of multiple candidate token sequences proposed by a draft model, allowing parallel verification of multiple continuations.

The process of automatically identifying and classifying different driving behaviors (e.g., aggressive vs. normal) from sensor data.

A system with two separate neural networks—one that processes questions and one that processes documents—both converting their inputs into comparable vector embeddings.

The parallel development and deployment processes for machine learning models and traditional software components.

The danger that AI technology can be misused for harmful purposes despite benign original intent.

A model with separate encoders for two input modalities that map them into a shared embedding space.

Organizing information at two levels of detail: high-level task guidance and low-level step-by-step actions.

An approach combining two complementary methods—one for logical reasoning and one for learning patterns—to solve a problem better than either alone.

A single model trained to perform multiple distinct tasks, such as both text generation and embedding, rather than being specialized for just one.

An architecture using two parallel processing streams with different time scales—one dense and one sparse.

A minimal, non-functional model used for testing infrastructure and workflows without the computational cost of a real model.

The ability to generate responses with a specific target length or speaking time.

Training approach that evaluates which skills remain helpful during learning and selectively retains only those that improve the current policy.

A formal system for reasoning about how beliefs and knowledge change when new information is revealed.

Building a network representation that changes over time to reflect evolving relationships, like road connectivity adjusted for traffic incidents.

Combining task-specific model parameters at inference time based on input features, rather than using a fixed merged model.

Automatically choosing the best execution approach (LLM reasoning, tool use, or code) for each step based on task requirements.

The ability to automatically adjust how many of a model's parameters are actively used based on available computational resources, allowing the same model to run efficiently on different hardware.

An optimization method that breaks problems into smaller subproblems and solves them recursively, storing results to avoid recomputation.

Removing training samples during training based on their importance or quality, rather than before training starts.

A quantization approach that adjusts precision levels during inference based on the input data, optimizing the balance between speed and accuracy on-the-fly.

Automatically creating questions of varying difficulty that adjust in real time based on learner responses and comprehension.

The process of recovering or reconstructing the full range of brightness values lost when converting from HDR to standard video formats.

A measure of how well an algorithm performs compared to the best possible strategy that adapts to changing conditions.

Choosing packet paths through a network in real-time based on current network conditions.

Mathematical models describing how systems evolve over time according to fixed rules.

Building neural network models that accurately capture the underlying rules governing how a system evolves over time.

A compressed representation of states that captures how the environment changes over time.

A technique for verifying program equivalence by representing multiple equivalent forms in a graph structure.

Stopping a model's computation before completion when sufficient confidence is reached, reducing computational cost.

Combining multimodal inputs (like text and images) at early layers of a model rather than after separate encoding.

Combining multiple objectives into a single weighted sum before training, which locks in a fixed trade-off.

Extended Berkeley Packet Filter; a technology for running sandboxed programs in the OS kernel to monitor system behavior.

A recording of the electrical signals produced by the heart, used to detect heart problems.

The ability to anticipate and address unusual or boundary conditions in code that might cause errors.

Processing data locally on a device at the edge of a network rather than sending it to a central cloud server, improving speed and reducing dependency on internet connectivity.

Running a model directly on local devices like phones, tablets, or IoT hardware rather than sending data to a remote server.

A computing device at the edge of a network (like a smartphone or IoT device) that runs AI models locally rather than sending data to a remote server.

The process of validating and removing unnecessary connections in a graph to improve its quality and interpretability.

Computing infrastructure spanning from edge devices (sensors, local hardware) to centralized cloud servers.

Attention mechanisms designed to reduce computational or memory complexity compared to standard quadratic-scaling attention.

Visual understanding from a first-person viewpoint, as seen from the wearer's perspective.

Understanding a scene from the viewpoint of a camera or observer positioned within the environment.

An AI system integrated directly into electronic health record software to assist clinicians with documentation or decision-making.

A special function that remains proportional to itself when transformed by an operator, used to decompose system behavior.

Dynamically adjusting the detail level and size of stored information based on current task relevance.

Simulating how deformable materials stretch, bend, and return to shape based on physical material properties.

A technique that protects important weights from previous tasks by adding a penalty term during learning.

A training objective used in probabilistic models to maximize the likelihood of observed data.

A pre-trained language model that learns by predicting which tokens in a sentence have been replaced, making it efficient and effective for downstream tasks.

Finding optimal delivery routes for electric vehicles that must visit customers within time windows and recharge at stations.

A recording of electrical brain activity used to detect neurological conditions like seizures.

Digital records of patient medical history, diagnoses, medications, and clinical events stored in structured formats.

A specialized computing device with limited resources designed to run specific applications, often integrated into physical systems.

A dense numerical vector that represents a word, sentence, or concept in a high-dimensional space.

Organizing vector representations of tokens into groups based on their semantic similarity.

The size of the numerical vector produced by an embedding model; larger dimensions capture more detail but require more storage and computation.

The number of numerical values used to represent a piece of text (1792 in this case), where more dimensions allow for more detailed semantic information to be captured.

The spatial structure and relationships between data points in a learned vector space.

The learning rate specifically applied to the embedding layer, which can be scaled independently from other layers.

A model that converts text into numerical vectors that capture semantic meaning, allowing computers to understand and compare the similarity between different pieces of text.

The model produces dense numerical vectors that represent the semantic meaning of text, which can be used for similarity comparisons or as input to other models.

Adding controlled noise to vector representations of text to obscure sensitive information.

A numerical vector representation of text that captures semantic meaning for comparison and analysis.

A metric that measures how similar two pieces of content are by comparing their numerical vector representations.

A mathematical space where text is represented as vectors, allowing similar texts to be positioned close together and enabling operations like similarity search and clustering.

Different ways to represent words as vectors (semantic, acoustic, or phonetic).

Removing duplicate or near-duplicate examples by comparing their vector representations in embedding space.

Comparing semantic representations (embeddings) to find similar content without reprocessing raw data.

Numerical representations of text that capture semantic meaning, allowing the model to measure similarity between different words or phrases.

AI systems designed to interact with and understand the physical world through robotic bodies or sensors, rather than just processing text.

The process of choosing which action a robot should execute next based on perceived state and task context.

Real-world performance metrics for robots like task completion time, motion smoothness, and energy consumption.

Robot learning and control for physical interaction tasks using integrated sensing and actuation.

An AI model trained on real-world physical interactions and sensor data from robots, rather than text or simulations alone.

The ability to understand and reason about physical tasks and spatial relationships in the real world, not just abstract concepts.

A representation or model that works across different body types or physical forms without being specific to one.

The point during training when a model suddenly gains the ability to perform a task above a threshold accuracy.

A measure of solution quality that arises from system dynamics rather than being explicitly defined beforehand.

When a model trained on narrow misaligned behavior generalizes to more severe harmful behaviors outside its training distribution.

The spread of emotions from one agent to others through interaction and observation.

Using emotionally-toned language or affective phrasing in prompts to influence model behavior.

The positive or negative quality of an emotion, ranging from negative to positive.

Training a model to recognize and respond to emotional context in conversations, prioritizing understanding and emotional connection over purely factual responses.

Instructing an LLM to generate responses with emotional awareness and compassion for patient concerns.

An algorithm approach that finds the best solution by minimizing errors on observed data.

A neural network trained to mimic the behavior of a complex physical model or simulation.

A model component that transforms input sequences (like protein amino acids) into meaningful numerical representations without generating new sequences.

A neural network component that transforms input text into a compressed numerical representation, focusing on understanding and extracting meaning rather than generating new text.

A model designed to convert inputs (like images or text) into numerical representations for understanding, rather than generating new content.

A neural network that transforms input data into a compressed representation, rather than generating new text or making predictions.

Models like RoBERTa that process text to understand meaning, typically used for classification tasks.

A neural network architecture with two parts: an encoder that processes input text and a decoder that generates output text, allowing the model to transform one sequence into another.

A neural network design that processes input text to understand and represent it, but cannot generate new text from scratch.

The tool or gripper at the end of a robot arm that physically interacts with objects in the environment.

Training data that only provides the final correct answer without showing the reasoning steps used to reach it.

An autonomous driving approach that directly maps sensor inputs to control outputs without explicit intermediate representations.

Training a model to solve a complete task directly from raw input (like document images) to final output, without breaking it into separate intermediate steps.

A system that takes raw input (like an image) and produces final output (like structured text) in one unified model, rather than chaining multiple separate tools together.

An optimization algorithm that preserves energy while descending to escape local minima.

A function that assigns a scalar value to each point in a space, defining an unnormalized probability distribution.

Recurring behaviors showing how users interact with content or systems over time.

A training technique where knowledge from multiple models is combined and compressed into a single, smaller model for better efficiency.

Combining multiple models to make better predictions than any single model alone.

A safety technique that combines outputs from multiple models and selects the most agreed-upon result.

Probabilistic scores assigned to multiple documents that determine their relative contribution to the final answer.

A language model specifically optimized for business and organizational use cases, prioritizing reliability, consistency, and professional output over other characteristics.

The task of recognizing that different names or phrases refer to the same real-world concept, such as matching 'MI' with 'myocardial infarction'.

Maintaining the same appearance and identity of characters, objects, and locations across different scenes in a video.

Automatically identifying and pulling out specific names, places, or things from text.

The task of identifying mentions of real-world concepts in text and connecting them to their canonical definitions in a knowledge base or ontology.

The task of identifying when different text references refer to the same real-world concept, such as matching variant spellings of a drug name to a single clinical entity.

A question-answering evaluation framework that tests whether models can retrieve factual information about specific entities.

A data structure that represents entities (like users or devices) and the typed relationships between them.

A regularized version of optimal transport that adds entropy constraints to encourage smoother, more balanced assignments between sources and destinations.

The gradient of prediction uncertainty with respect to visual embeddings, used to identify ambiguous regions.

Encouraging an agent to explore diverse state-action pairs by maximizing the entropy of its occupancy measure.

Controlling the randomness of a model's outputs to prevent it from becoming too deterministic or too random during training.

A decoding approach that continues unmasking tokens until cumulative entropy exceeds a threshold, balancing generation speed and quality.

System state where the ability to generate random numbers becomes the limiting factor rather than arithmetic computation.

Automated creation of task specifications and evaluation settings for training or testing agents.

AI system's ability to store and recall specific past events or experiences.

A situation where different participants have different information or knowledge about the same topic.

The preservation of an agent's ability to form accurate beliefs and maintain truthful internal representations.

The degree to which discourse relies on evidence-based reasoning versus intuition and subjective belief.

Uncertainty from lack of knowledge that can be reduced with more data or better models.

A phenomenon where the model learns to place its initial output near the fixed point, allowing inference without iteration.

Neural networks designed to respect geometric symmetries and transformations in molecular or crystal structures.

How well a design follows established principles for human comfort, safety, and efficient use of space.

Systematic examination of model failures to identify patterns and root causes beyond aggregate metrics.

Firmware algorithms that detect and correct errors in memory to maintain reliability as storage density increases.

How mistakes in early steps of a process accumulate and worsen downstream results.

A structured classification system that categorizes different types of errors to enable systematic analysis and mitigation.

A topological invariant that counts connected components, holes, and voids in a shape to characterize its structure.

When an evaluator systematically biases its judgments based on contextual information rather than actual content quality.

When AI judges appear to agree on scores but are actually using shallow patterns rather than substantive reasoning about quality.

A quantitative measure used to assess how well a model or system performs on a specific task.

A specialized language model trained to assess and score the quality of outputs from other AI models, acting as an automated judge.

An attack where an adversary modifies input features at test time to fool a deployed classifier.

A sensor that captures pixel-level brightness changes asynchronously, producing sparse temporal event streams.

Temporal representations that capture when and how much change occurs in music or video.

Automatically detecting higher-level events from lower-level timestamped observations using logical rules.

Grouping related incident reports together to identify a single underlying problem from multiple user descriptions.

Recording all changes to data as a sequence of immutable events for full history tracking.

A generalized pattern representing a class of similar log messages with variable fields.

Combining information from multiple frames or observations to make a single robust decision or diagnosis.

When a model's answer directly contradicts the provided evidence or clinical guidelines.

A model's ability to change its predictions based on whether evidence supports or contradicts a claim.

Linking AI outputs to specific source documents or facts that support them.

A collection of diverse, complementary pieces of evidence retrieved to support multi-faceted reasoning.

Fixing errors in code or theory by using specific signals like test failures and reviewer feedback to target the root cause.

A method that combines multiple predictions while quantifying uncertainty using evidence theory.

A training method that gradually increases the complexity of instructions given to a model, helping it learn to handle increasingly difficult tasks.

An AI optimization technique that mimics natural selection to explore and improve solutions over many iterations.

A statistical property ensuring that the order of data points doesn't matter, required for conformal prediction to provide valid guarantees.

Saving and reusing working code solutions instead of text descriptions for repeated tasks.

Stateful, runnable systems that simulate real-world tool interactions and can verify agent actions.

Detailed analysis of why an action succeeded or failed, beyond just binary success/failure signals.

Anchoring AI-generated questions and explanations to actual runtime behavior and concrete execution traces.

A detailed strategy for solving a problem, which can be implemented and tested before committing to a final answer.

A record of every step a program takes as it runs, including variable values and function calls.

Detailed information about what happened during a program's execution, used to diagnose failures.

Validating agent behavior by running code and checking if outputs match expected results, rather than relying on static analysis.

A variable in a causal model that is not caused by any other variables in the model; represents external sources of randomness.

An acquisition function that selects points likely to improve over the current best solution.

Useful patterns and insights extracted from real-world interactions and deployment experience.

Learning through direct interaction with the environment and feedback from actions taken.

Strategically choosing which experiments to run to maximize information gain given a limited budget.

The process of testing hypotheses through controlled experiments to uncover causal relationships.

An early version of a model released for testing and feedback, which may have bugs or incomplete features compared to stable versions.

A measure of how much each expert in an MoE model contributes to the final output, used to decide which experts need higher precision.

The mechanism in a mixture-of-experts model that decides which specialized sub-networks should process each piece of input.

The process where different experts in an MoE learn to handle distinct types of inputs or tasks (e.g., code vs. math).

How evenly the workload is distributed across experts; balanced utilization prevents some experts from being unused.

The ability to understand and interpret why an AI model made a specific decision or prediction.

Whether a model applies the same reasoning strategy (highlights the same regions) across different instances of the same class.

A mode where a model generates visible reasoning steps before producing a final answer, allowing you to see its problem-solving process.

A feature that allows a model to show its reasoning process step-by-step before providing an answer, useful for complex problems that benefit from deliberate problem-solving.

The maximum gain a player can achieve by deviating from an equilibrium strategy.

The process of trying diverse actions during training to discover which ones lead to better outcomes.

Balancing between exploiting known good solutions and exploring new possibilities to find better ones.

A weighted average that gives more importance to recent values than older ones.

A model's ability to handle facial expressions it wasn't explicitly trained on by learning underlying expression patterns.

The capability to work with and maintain understanding across large amounts of text or multiple documents during reasoning.

Estimating both the position and shape of objects that occupy multiple sensor measurements.

A capability that allows a model to think through complex problems step-by-step internally before providing a final answer.

A reasoning technique where a model works through a problem step-by-step internally before providing an answer, improving accuracy on complex tasks.

Reward signals based on computational verification methods rather than the model's own internal signals.

Whether results from a controlled study apply to real-world situations outside the lab.

Predicting model behavior in a region (like very large training runs) based on observations from smaller regions.

Making predictions beyond the range of training data, such as forecasting system behavior at untested excitation levels.

Technology that identifies or verifies people by analyzing facial features in images.

An optimization technique that selects diverse items by maximizing how well they represent the full set of options.

The process of verifying claims against reliable sources to determine their accuracy.

Verifying if claims are true using only an LLM's internal knowledge, without searching external databases.

A decomposition of norm computation into smaller intermediate terms to avoid materializing large dense matrices.

How often an AI model produces correct, verifiable information without errors or false claims.

Anchoring a model's responses to verified, real-world information rather than relying solely on patterns learned during training.

An LLM's ability to accurately retrieve and output factual information from its training data.

A group of related system components or subsystems that share common failure modes and characteristics.

The quantified likelihood that an AI system will make a harmful or incorrect decision in real-world deployment.

Systematic evaluation of an AI system to detect and measure bias across demographic groups or decision scenarios.

Whether an AI model's stated reasoning actually explains how it arrived at its answer, or if it's post-hoc justification.

The task of identifying false or misleading news articles, typically framed as a classification problem.

When incorrect or outdated information from past interactions influences future reasoning.

The ability to identify when a question contains incorrect assumptions or fabricated facts before answering.

A greedy algorithm that selects points by always choosing the one farthest from previously selected points.

A method for efficiently updating model parameters or memory states during forward passes without full recomputation.

Model parameters that are quickly adapted during inference to capture task-specific or input-specific patterns.

Pinpointing the exact location of bugs or errors in code or systems.

A graph showing how errors flow through transformer components from their origin to observable symptoms.

The ability of a system to continue operating correctly even when components fail.

Automatically checking whether a problem instance has at least one valid solution before using it for testing.

Enhancing a model by adding hand-crafted or extracted features (like linguistic metrics) alongside learned representations.

Storing intermediate computed features during inference to reuse them in later steps, reducing redundant computation.

The process of selecting and designing input features that a machine learning model uses to make predictions.

The process of using a model to convert raw input text into numerical representations (features) that capture the meaning of the text.

When a single concept is scattered across many separate features instead of being cleanly captured by one or a coherent group.

A measure of how much each input variable contributes to a model's predictions.

How multiple input features combine together to influence a model's prediction, beyond their individual effects.

A method to identify how combinations of input features jointly influence a model's predictions, beyond individual feature effects.

A measure of how well different visual concepts can be distinguished in a model's learned feature space.

A technique that dynamically adjusts learned representations by scaling and shifting features based on problem-specific conditions.

Training models across multiple devices without centralizing sensitive data in one place.

A standard neural network layer in transformers that processes information independently at each position.

A neural network that processes input in a single forward pass without recurrence or iterative refinement.

The method used to apply feedback text to refine and improve a search query representation.

Where the text used to improve a search query comes from, such as LLM-generated text or actual documents.

Using execution results and error signals to adaptively adjust agent behavior and improve reliability over time.

Training or prompting a model with only a small number of examples to perform a new task.

The degree to which a quantized or compressed model preserves the quality and accuracy of the original full-precision model.

A filtering mechanism that only includes accurately generated entity appearances in consistency evaluation metrics.

A measure of how well an explanation captures the true reasoning of a model by testing prediction changes.

A code completion technique where the model predicts missing code between existing lines, rather than only generating code forward from a starting point.

Distinguishing between very similar categories, like telling apart different bird species rather than just identifying 'bird vs. not bird'.

The ability to accurately generate readable text and small details within generated images.

Small, specific visual elements in an image, such as text within a photo or subtle differences between similar objects.

The ability to further train or customize a pre-trained model on your own data to adapt it for specific tasks or domains.

A model created by training an existing pre-trained model on new data to specialize it for specific tasks or behaviors.

A pre-trained model further trained on a smaller, task-specific dataset to improve performance on that task.

The process of further training a pre-trained model on new data to adapt it for specific tasks or domains.

A numerical technique that breaks a complex domain into small pieces to solve physics equations approximately.

Mathematical structures with finitely many elements where arithmetic operations follow specific rules.

A problem setting with a fixed, known endpoint in time, as opposed to indefinite or infinite-horizon problems.

A recurrent neural network model where neurons output continuous activation rates rather than discrete spikes.

A formal language for expressing rules and constraints using predicates, variables, and logical operators.

The time it takes for a stochastic process to reach a target state for the first time.

The initial search system that finds candidate documents before refinement techniques are applied.

A metric measuring the difference between score functions of two distributions.

A variant of dynamic programming that first estimates unknown functions (like demand) from data, then uses those estimates for optimization.

A moment when the eye pauses on a specific location while viewing an image, typically lasting 100-500 milliseconds.

Repeatedly applying a function until it converges to a stable value, used here for test-time computation in looped models.

Finding a stable state where a function's output equals its input, used here to refine embeddings iteratively.

Embeddings that always produce vectors of the same length regardless of input length, which limits how much detail can be captured for very long documents.

A company's primary, most capable model designed to showcase their best technology and handle the most demanding use cases.

Software abstraction that maps logical addresses to physical memory locations in SSDs, managing wear and errors.

Dynamically allocating wireless frequencies based on real-time demand instead of fixed assignments.

The process of deciding where to place components on a chip to meet design constraints and performance goals.

Generating data by learning reversible transformations between simple and complex distributions.

A generative modeling technique that learns to transform random noise into realistic data by following learned flow paths.

Functional magnetic resonance imaging; a non-invasive technique measuring brain activity through blood flow changes.

A training approach combining focal loss (which focuses on hard examples) with contrastive learning to handle imbalanced datasets.

A training approach combining focal loss (which emphasizes hard examples) with contrastive learning to handle imbalanced datasets.

Custom sound effects created to match specific actions or movements in video, like footsteps or door slams.

A parameter that controls how quickly a filter discounts old data, balancing between adapting to new conditions and maintaining stability.

Testing reward hypotheses by branching from shared policy checkpoints and comparing short-horizon performance to assess reward quality.

Expressing system requirements or policies in a precise mathematical language that tools can automatically verify.

Mathematical proof that a system meets its specifications, here implemented in Lean 4 to certify material stability predictions.

Real-time guidance given to students during learning to help them improve, rather than just assigning a final grade.

Simulating a robot's future states by repeatedly applying its dynamics model to predict outcomes of candidate actions.

A training objective that penalizes the model for assigning probability to regions the true distribution doesn't cover.

A single computation cycle where input data flows through the model's layers to produce an output prediction.

An agent's reasoning about future consequences and goals rather than just reacting to past events.

An efficient method for computing derivatives by propagating changes forward through a computation graph.

A large pre-trained model that serves as a starting point for building other models, rather than being trained from scratch.

The underlying structural design of a neural network that determines how it processes and learns from data, distinct from standard transformer designs.

Large pre-trained AI models that can be adapted to many different tasks without starting from scratch.

Mathematical representation showing which frequencies (periodic patterns) are present in data.

A data format that stores model weights using 16-bit floating-point numbers, preserving full model accuracy while using less memory than 32-bit formats.

A low-precision numerical format that uses only 4 bits to represent numbers, enabling faster computation and smaller model sizes compared to standard 32-bit precision.

A 4-bit number format used in quantization that represents values with minimal precision, significantly shrinking model size while maintaining reasonable accuracy.

A 4-bit floating-point number format that represents model weights with very low precision, enabling extremely efficient inference on compatible hardware.

A ultra-low precision format using 4-bit floating-point numbers to represent model weights, enabling extreme compression.

A compression technique that represents model weights using only 4-bit floating-point numbers instead of larger formats, reducing memory usage and speeding up inference.

A compressed number format that uses 8 bits instead of the standard 32 bits, dramatically shrinking model size at the cost of slightly reduced precision.

A specific quantization method that uses 8-bit floating-point numbers and adjusts precision dynamically based on the data being processed, balancing speed and accuracy.

An 8-bit numerical format that stores numbers with reduced precision compared to standard formats, enabling smaller model sizes and faster computation.

A data format that stores numbers using 8 bits instead of the standard 32 bits, significantly reducing memory requirements with minimal quality loss.

A compression technique that reduces model size by representing weights using 8-bit floating-point numbers instead of higher precision, making it faster and more memory-efficient.

A set that an optimization trajectory converges to, with self-similar structure at multiple scales rather than converging to a single point.

A model's ability to produce answers without predefined options, requiring genuine recall and reasoning.

How often different facts or tokens appear in training data, which affects what models learn.

Decomposing signals into high-frequency (details, edges) and low-frequency (overall structure, semantics) components.

Evaluating model performance separately for rare, medium, and common classes to reveal patterns hidden by overall metrics.

The automated creation of user interface code and visual elements based on descriptions or specifications.

A state-of-the-art AI model representing the cutting edge of what's currently possible in terms of capability and performance.

State-of-the-art, cutting-edge AI models that represent the current best performance in the field.

A model that represents the current state-of-the-art or cutting edge in AI capabilities, competing with the most advanced models available.

The largest and most advanced language models available, representing the cutting edge of AI capabilities.

A cutting-edge AI model representing the current state-of-the-art in performance and reasoning capabilities.

A pre-trained model component that is kept unchanged during training to preserve its learned knowledge.

A model using standard 32-bit floating-point numbers to represent weights, providing maximum accuracy but requiring more memory.

Model parameters stored at maximum numerical accuracy (typically 32-bit floating point), which provides the best quality but requires more memory and computation.

The ability of a model to output structured requests to invoke external tools or APIs rather than generating free-form text.

Internal model representations that encode what tasks do, allowing comparison of task similarity and prediction of learning trajectories.

Vector representations of tasks extracted from model activations during in-context learning.

Growing a model's capacity while mathematically guaranteeing it behaves identically to the original at the start.

Mathematical operations like rotations that rearrange a model's weights without changing what the model computes.

Specifications describing what a software system should do and its specific behaviors and features.

A discrete token that encodes both an agentic operation and latent visual reasoning capability without explicit visual supervision.

An attention mechanism that progressively compresses and simplifies the input sequence, reducing computational cost while maintaining important information.

GPU operations combined into a single kernel to reduce memory traffic and improve computational efficiency.

Logic-based rules that handle uncertainty and gradual membership rather than strict true/false classifications.

Comparing text strings by measuring character-level similarity rather than exact matches.

A mechanism where a context signal scales the magnitude of state-dependent responses without changing their underlying structure.

A formal notation for encoding game rules so different AI systems can play the same game consistently.

A stable state where no agent can improve their outcome by unilaterally changing their strategy.

A learned mechanism that selectively applies corrections to predictions based on per-dimension scaling factors.

A neuron that controls whether tokens are routed to standard or exception processing paths.

A learned or rule-based function that selectively enables or disables components based on input conditions.

A mathematical property ensuring a model's predictions remain consistent regardless of arbitrary coordinate system choices or numerical representations.

A statistical model that learns patterns from data and provides uncertainty estimates for predictions.

Systematic tendency of models to favor one gender over others in language generation and translation tasks.

The capability to think through problems logically, break down complex questions, and arrive at conclusions across a wide variety of topics.

Designed to handle a wide variety of different tasks rather than being specialized for one specific domain.

A model trained to handle a wide variety of text tasks—like writing, answering questions, and reasoning—rather than being specialized for one specific task.

An AI model designed to handle many different types of tasks well, rather than being specialized for one specific domain.

A model trained to perform well across many different types of tasks rather than being specialized for one specific domain.

A robot trained to perform many different everyday tasks rather than being specialized for one specific job.

A model's ability to perform well on new, unseen data that differs from what it was trained on.

The difference between a model's performance on training data versus unseen test data.

A mathematical method for aligning and comparing representations across different neural networks by finding optimal rotations.

An inference approach where a model generates an intermediate image before answering a question about it.

Vector representations of text created by generative language models that capture semantic meaning.

A probabilistic framework that generates samples with probability proportional to a reward function, useful for optimization tasks like molecule discovery.

A model trained to generate new text by predicting the next word or sequence of words based on patterns it learned during training.

An AI model trained to create new data (like images) that resembles its training data.

Additional training phase after initial pretraining that uses generative tasks to improve model capabilities.

A model's procedure for creating new outputs (like floor plans) based on learned patterns from training data.

A methodology that grows phenomena from micro-level interaction conditions to identify sufficient mechanisms, detect thresholds, and design safety interventions.

Matrices used to encode data into codewords in error-correcting codes.

The shortest path between two points along a curved surface, as opposed to straight-line distance.

Checking that spatial analysis results are realistic (e.g., no negative distances, valid coordinate ranges, sensible geographic relationships).

A mathematical framework (Clifford algebras) that extends vectors with operations for rotations, reflections, and higher-dimensional relationships.

Structural constraints added to a model to encode domain knowledge about geometry, such as crystal lattice properties.

Maintaining structural and spatial accuracy across multiple views or representations of a 3D object.

The alignment between router weight directions and expert weight directions that emerges during training.

Building a 3D model of a scene from video or images by estimating depth and camera motion.

Property where data points can be separated into groups using a linear boundary in vector space.

Multimodal representations that preserve spatial and geometric information about the scene to maintain disambiguating context.

Using machine learning and statistics to analyze data tied to geographic locations.

A file format for quantized models designed for efficient CPU and GPU inference with llama.cpp.

A file format designed for efficient storage and loading of large language and embedding models, optimized for fast inference on various hardware.

A mathematical technique for reweighting probability distributions along trajectories without computing gradients.

Attention mechanism where each token can attend to all preceding tokens in the sequence.

Populations and nations that represent the numerical majority of the world but are historically marginalized in Western-dominated systems.

When an AI agent gradually abandons its original objective and pursues different goals instead.

A low-dimensional vector that captures task identity and enables rapid adaptation to new tasks without retraining.

When an AI system's stated objective doesn't match the actual intended outcome, leading to unintended behaviors.

Rules and policies that limit AI autonomy to ensure oversight, safety, and alignment with organizational values.

A set of rules and structures that constrain and guide AI behavior to ensure reliability and consistency.

An open-source license that allows free use and modification of software, but requires any derivative works to also be open-source under the same license.

A transformer-based neural network design that processes text sequentially and predicts the next word based on previous context.

A transformer-based neural network design from OpenAI that processes text sequentially to predict and generate the next word in a sequence.

An older transformer-based design for language models that generates text by predicting one word at a time, simpler and smaller than modern alternatives.

A modified version of the GPT-2 architecture that changes the original design, such as by reducing size or adjusting training.

A transformer-based design that follows the same structural principles as OpenAI's GPT-3 model, using layers of attention mechanisms to process text.

A class of transformer-based language models descended from the original GPT design, characterized by autoregressive text generation and broad general-purpose capabilities.

A transformer-based neural network design that uses self-attention to process and generate text, serving as the structural blueprint for this model.

An open-source large language model architecture based on the GPT design, created as an alternative to closed-source models.

An open-source transformer-based architecture designed for training large language models, similar in structure to GPT models.

A neural network design based on transformer technology that processes text sequentially and generates one word at a time.

A quantization technique that compresses model weights to lower precision, reducing file size and memory requirements while maintaining reasonable performance.

Assigning GPU resources to different models or tasks to optimize throughput and latency.

Performance degradation that occurs when multiple inference requests compete for the same GPU's memory and compute resources.

The high-speed memory on a graphics processor used to store and process model weights and computations during inference.

Designing and tuning a model to run efficiently on graphics processing units (GPUs), which are specialized hardware that accelerates AI computations.

A technique ensuring that gradient updates from different tasks point in compatible directions to avoid conflicts.

Estimating how model parameters should change without actually computing full gradients or updates.

Improving model performance by following the direction of steepest improvement in parameters.

Systematic error in gradient estimates that prevents optimization from reaching the true optimum.

Building models sequentially where each new model corrects errors from previous ones.

Limiting the magnitude of gradients during training to prevent extreme updates and improve stability.

Sending model weight updates between devices and servers during distributed training, a major bottleneck on bandwidth-limited networks.

Reducing the size of gradient data to speed up training on distributed systems.

When different training objectives pull model updates in opposing directions, causing optimization to fail or degrade.

Scaling gradient values to maintain consistent learning rates across different parameter groups or layers.

A training technique that flips gradient signs to force a model to learn features that fool an adversarial classifier.

Technique that selectively modifies or blocks gradient flow to prevent interference between different learning objectives.

An explainability technique that uses model gradients to identify which input features most influence predictions.

Setting starting values for trainable parameters using information from model gradients to improve convergence and final performance.

Optimizing a function without computing gradients, using only function values or rankings.

A task where a model identifies and fixes grammar, spelling, and syntax mistakes in written text.

A linguistic system where nouns and related words are classified into categories requiring specific agreement patterns.

An attention mechanism that learns weighted interactions between nodes in a graph structure.

The task of assigning a label or category to an entire graph based on its structure and node features.

Transferring knowledge from a labeled source graph to an unlabeled target graph when their structures or distributions differ.

A measure of how different two graphs are, based on the minimum edits needed to transform one into the other.

Converting a graph structure into a compact text representation that preserves its properties.

A neural network that operates on graph-structured data by passing messages between connected nodes to learn relational patterns.

Neural networks designed to process graph-structured data by learning representations of nodes and edges.

Methods for converting graph structures into numerical vectors that preserve meaningful information about nodes and edges.

A structured approach that represents events and their relationships as a graph and processes them in sequential stages.

Generating text by always selecting the highest-probability next token, without exploring alternatives.

Accurate reference labels or measurements used to train and evaluate machine learning models.

The actual underlying causes or features that explain observed data in a system.

AI reasoning that relies on specific documents or data provided to the model, rather than just its training knowledge.

The practice of ensuring a model's responses are based on and supported by provided source documents rather than generated from general knowledge.

A generalized measure of uncertainty or disorder that follows mathematical group rules, extending beyond standard entropy.

A training method that improves model reasoning by comparing outputs and rewarding better explanations.

In quantization, the number of weights that share a single scaling factor; smaller groups preserve more precision but use more memory, while larger groups save more memory but may lose detail.

Reducing model size by compressing weights in groups rather than individually.

Predicting aggregate behavior of a group of users rather than individual users, useful for testing business strategies.

An optimization technique that reduces memory usage and speeds up inference by having multiple query heads share the same key and value heads instead of each having their own.

Group Relative Policy Optimization, a reinforcement learning algorithm for fine-tuning language models with reward signals.

Safety mechanisms built into a model to refuse harmful requests or prevent it from generating unsafe content.

An AI system that interacts with computer interfaces by clicking, typing, and navigating screens.

The ability to identify and locate specific elements (like buttons or text fields) within a graphical user interface based on natural language descriptions.

A technique to steer AI generation toward desired outputs by providing additional control signals during inference.

A technique that steers a model's output toward desired behavior by balancing multiple objectives during inference.

Steering a model's text generation process using external signals or constraints without modifying the model itself.

A training technique that intelligently selects the most informative examples from your training data to improve model efficiency and performance.

A differentiable relaxation technique that approximates discrete choices to enable gradient-based optimization.

When a model generates plausible-sounding but factually incorrect or fabricated information.

The ability to identify when a model generates false or unsupported information that isn't grounded in the provided source material.

A mathematical equation solving optimal decision-making problems over time.

A route that visits every location exactly once without repeating any node.

A quantum computing technique that simulates the evolution of a physical system described by a Hamiltonian.

The ability of a model to identify and interpret handwritten characters and words from images, accounting for variations in writing style and quality.

A rule that must always be satisfied during optimization, rather than being treated as a soft penalty that can be violated.

Challenging negative examples that are similar to the target but still incorrect, used during training to make the model learn more nuanced distinctions.

Tuning a model's design or training to run more efficiently on specific hardware (like NVIDIA GPUs), reducing memory usage and inference time.

A structured system that categorizes different types of harmful content (like violence, hate speech, or misinformation) so a model can recognize and classify them.

An architecture that alternates between thinking (reasoning about a problem) and acting (taking physical steps), allowing the model to plan and execute robot actions iteratively.

The design and implementation of control systems that manage agent behavior and task execution.

Systematic process of identifying potential failures and dangerous scenarios in a system.

The instantaneous rate of an event occurring at a given time, conditional on survival up to that time.

Systematically disabling individual attention heads to determine which ones are causally responsible for specific model behaviors.

Mathematical technique to approximate probability distributions using orthogonal polynomials.

The complete set of eigenvalues of the loss function's second-derivative matrix, describing the curvature in all directions.

Systematic differences in how different groups (by language, task, etc.) rank or prefer models.

Differences in how a treatment affects different individuals based on their characteristics.

A practical problem-solving method that finds good solutions quickly without guaranteeing optimality.

The size of the internal vector representation used by a neural network to process and store information about the input.

The internal numerical values a neural network computes at each layer as it processes input.

The dimensionality of the internal representations that a neural network uses to encode information about text.

An adversarial attack that injects malicious tokens to corrupt a model's internal memory and degrade performance.

Internal representations computed by neural networks that capture learned patterns.

Combining multiple independent predictions or estimates using a structured approach that accounts for differences in their reliability.

A multi-stage attention approach that first selects relevant tokens coarsely, then applies fine-grained attention on the selected subset.

A statistical technique using Platt scaling with a hierarchical prior to adjust model confidence while preventing over-shrinking of extreme predictions.

An unsupervised learning method that builds a tree of nested clusters by repeatedly merging or splitting groups based on similarity.

A neural network component that processes images at multiple levels of detail simultaneously, capturing both fine details and broad patterns.

A multi-level approach to reasoning where information is processed and combined across different levels of abstraction.

Storage system using multiple memory tiers (e.g., fast GPU memory and slower CPU memory) to balance speed and capacity.

Planning at multiple levels of abstraction, where high-level plans are refined into low-level actions.

Breaking down a complex decision into multiple levels, like deciding family → genus → species in order.

Breaking complex tasks into simpler sub-tasks organized in levels, where agents learn high-level strategies and low-level actions separately.

A technique that aggregates features from multiple layers of a neural network to create multi-scale guidance signals.

Testing correctness at multiple levels: properties, interactions, and full rollouts to ensure system correctness.

The process of automatically converting algorithmic descriptions into hardware designs, typically using pragmas and code transformations.

Derivatives beyond the first order (gradients) that capture more complex relationships in how inputs affect outputs.

The exponential growth in the number of quantum states available as more qubits are added to a quantum system.

Performance feedback derived from comparing baseline and skill-enhanced rollouts to guide skill and policy updates.

The expected number of steps for an algorithm to reach a target state from a starting point.

A training technique that simulates viewing images from different angles and perspectives to teach the model to recognize the same features under geometric transformations.

Techniques to make AI models produce truthful responses instead of false or misleading ones.

A type of recurrent neural network with symmetric connections used for associative memory and optimization.

A linear algebra operation that reflects vectors across a hyperplane, used here to align word direction vectors.

A self-supervised learning approach for audio that learns meaningful speech representations by predicting masked portions of audio, similar to how language models learn from text.

Forecasting future body positions and movements based on past motion sequences.

A controlled experiment measuring how much an AI system improves human performance compared to working without it.

A workflow where humans and AI agents work together, with AI assisting at multiple stages rather than just solution generation.

A system where AI predictions are reviewed and validated by human experts before final decisions.

Training robot control policies by learning from human movement demonstrations.

A model that combines two different neural network designs (in this case, Mamba2 and attention mechanisms) to balance speed and performance.

A neural network design that combines Mamba (a fast, efficient sequence model) with Transformer components to balance speed and capability.

A memory system combining learnable parameters with non-learnable mechanisms to balance flexibility and efficiency.

A capability that allows a model to switch between fast, direct responses and slower, more deliberate reasoning depending on task complexity.

A non-Euclidean geometry where space curves negatively, naturally suited for representing hierarchical and tree-like structures.

A neural network that models relationships between multiple elements simultaneously, capturing high-order interactions beyond pairwise connections.

A neural network that generates weights for another neural network instead of learning them directly.

A configuration setting (like learning rate or network size) that you choose before training a model.

Using optimal hyperparameters found at small scale to train larger models without expensive retuning.

A geometric shape in high-dimensional space used in optimization and probability theory.

Training method that constrains weight matrices to lie on a fixed-norm hypersphere for improved stability and scaling.

Mathematical structure where points lie on the surface of a high-dimensional sphere, preserving directional relationships.

A geometric arrangement where data points lie on the surface of a high-dimensional sphere, preserving directional relationships.

The ability to uniquely determine a model's parameters from observed data.

The policies, processes, and controls that manage who (or what) can access systems and data, and what actions they are authorized to perform.

Maintaining consistent, unique identifiers for entities across different systems and time periods.

Keeping a person's unique facial characteristics unchanged while editing other attributes like expressions.

Separating what makes a face unique (identity) from how it moves (expression) so each can be controlled independently.

The task of automatically generating a text description of what appears in an image.

Modifying specific parts of an existing image while preserving other elements.

A neural network component that converts images into numerical representations that capture visual features and patterns.

A computer vision task that divides an image into regions or labels each pixel to identify different objects or areas.

Hardware in cameras that processes raw sensor data into final images, increasingly using AI for enhancement.

The process of converting images into discrete tokens (small units) that a language model can process, similar to how it handles text.

The ability to understand and answer questions that require analyzing both visual content and textual information together.

The ability to analyze a visual image and automatically produce source code that recreates or represents that image's structure and content.

The task of automatically generating natural language descriptions of images, converting visual information into written words.

Training a model to copy behavior from expert examples without understanding the reasoning behind decisions.

A learned behavior that mimics actions from human demonstrations or other expert examples.

Identifying which parts of a system are affected by a proposed code change.

Games where players don't know all relevant information, like hidden opponent cards or future draws.

A limitation that emerges naturally from the training setup rather than being explicitly specified.

A hidden, structured order in which models naturally learn skills during pretraining, without explicit curriculum design.

Computing gradients through an implicit equation without unrolling iterations, keeping memory constant.

A user's underlying goal or need that is not directly stated but must be inferred from context.

Neural networks defined by equations that must be solved rather than computed layer-by-layer, enabling parameter efficiency.

Structured behaviors that emerge naturally from an LLM's token-level decisions without being explicitly programmed or instructed.

Inferring unobserved values or outcomes from historical patterns in data without explicit instruction.

Information about what a community values inferred from their behavior (like engagement and acceptance) rather than explicit feedback.

Culpeper's framework analyzing how language can intentionally or unintentionally cause offense or disrespect.

A technique to adjust samples drawn from one distribution to match another by weighting them by their probability ratio.

Adjusting sample weights to correct for sampling from the wrong distribution.

A technique to estimate gradients by reweighting samples from one distribution to match another.

Learning from examples provided in a prompt without updating model weights.

Data compression performed during simulation execution rather than after data is written to disk.

A training technique where negative examples (dissimilar samples) come from other items in the same training batch, helping the model learn to distinguish between similar and dissimilar texts.

A mechanism where relevant information is retrieved from model parameters themselves rather than from external memory or attention, helping reduce computational bottlenecks.

Ensuring that the goals and rewards of different agents or system components work toward the same overall objective.

How well a model adjusts its behavior when the rewards or payoffs for different actions change.

A theory of humor based on identifying mismatches in expectations and then resolving them in unexpected ways.

Writing systems used for South Asian languages like Hindi, Tamil, Telugu, and Bengali that have distinct characters and phonetic rules.

Artifacts or evidence left behind by attackers (like malicious URLs, IP addresses, or file hashes) that reveal a security breach.

An attack where malicious instructions are hidden in data an AI agent retrieves, causing unintended actions.

Built-in assumptions about how data should behave, like physics rules, that help models learn faster with less data.

A sensor that measures acceleration and rotation to track motion without external references.

The process of running a trained model to generate predictions or outputs from new inputs.

Specialized hardware designed to speed up the execution of trained AI models.

The computational resources and processing power required to run a model on new data after it has been trained.

The computational resources and time required to run a model on new inputs, typically measured in memory usage and processing time.

The ability of a model to generate outputs quickly and with low computational resource consumption during real-world use.

Software that runs a trained model to generate predictions or outputs; vllm is an optimized inference engine for large language models.

Software that optimizes how a trained model runs on specific hardware; MLX is an Apple-optimized framework for efficient inference on Apple Silicon.

The time it takes for a model to generate a response after receiving an input.

Techniques and design choices that make a model faster and more efficient to run on hardware, prioritizing speed and resource usage over training flexibility.

The numerical precision (number of bits) used when running a model to generate outputs; lower precision is faster but may reduce quality.

A system that hosts trained ML models and processes incoming prediction requests on deployed hardware like GPUs.

How quickly a model can generate predictions or outputs after being given an input, measured in time per token or tokens per second.

Reduction in time needed to run a model and get results, measured as a multiple of the original speed.

The amount of time it takes for a model to process input and generate output after it has been trained.

Extra processing power spent by the model while generating a response to think through problems more carefully before answering.

The computational resources used when a model generates answers, as opposed to during training.

Detecting and fixing model mistakes during generation without retraining, using only the current forward pass.

A model used during generation to score outputs without requiring retraining of the main system.

A technique where a model allocates more computational resources and time during inference (when generating answers) to improve quality and accuracy on harder problems.

A mathematical proof written in natural language rather than formal logical notation.

Combining fragmented knowledge from multiple sources to make better collective decisions than any single source could.

When one party in a transaction has more or better information than the other, creating imbalanced power.

A point in a system where information capacity is severely limited, constraining overall performance.

The amount of useful, non-redundant information contained in a token or representation.

The task of automatically identifying and pulling out specific data or facts from documents, such as names, dates, or amounts from forms.

The reduction in uncertainty about a target achieved by knowing a feature.

Mathematical framework treating probability distributions as points in curved space, measuring optimization difficulty via curvature.

When a model accidentally learns from information it shouldn't have access to, like future data or test set details.

The task of finding relevant documents or passages from a large collection in response to a user query.

The process of gathering data from multiple sources and combining it into a coherent, unified response or summary.

Running a model as an intermediate processing layer within an application pipeline, typically to filter or validate data before it reaches the main system.

A safety intervention using statements with specific linguistic forms to prevent misaligned behavior, but which can paradoxically trigger misalignment on similar-form inputs.

The task of filling in missing or masked regions of an image while maintaining coherence with the surrounding content.

A neural network architecture designed to be convex in its inputs, useful for constrained optimization and learning convex functions.

The type of data a model can accept as input, such as text, images, or audio.

The pixel dimensions (448×448 in this case) at which the model processes images, affecting the level of visual detail it can perceive.

Checking that input data meets basic requirements (correct format, expected properties, no obvious errors) before processing it.

How much a model's behavior changes in response to different inputs, crucial for generalization.

The types of data a model can accept as input and produce as output, such as text, images, or audio.

The process of producing additional relevant information or perspectives that extend or improve an initial answer.

Identifying the core techniques or key ideas needed to solve a complex problem.

Identifying and locating individual objects of the same class separately in an image.

The ability to apply different settings or modifications to individual objects within a scene independently.

The ability of a model to follow primary instructions even when secondary or conflicting instructions are present.

The ability of a model to understand and execute specific tasks or commands given in natural language prompts.

A model fine-tuned on instruction-response pairs so it follows user prompts more reliably.

A training process that teaches a model to follow specific user instructions and commands, improving its ability to respond appropriately to requests.

The prediction that advanced AI agents will pursue certain goals (like self-preservation) regardless of their final objectives.

A sequence of checks replacing ground-truth labels: responsiveness to safe/unsafe contrasts, dominance of target variance, and stability across reruns.

A specific quantization format that represents model weights using only 4 bits per value, significantly reducing model size while maintaining reasonable performance.

A quantization method that represents model weights using only 4-bit integers instead of full-precision floating-point numbers, dramatically shrinking the model's memory footprint.

A compression technique that reduces a model's size and memory usage by storing weights as 4-bit integers instead of higher-precision numbers, making it faster and cheaper to run with minimal accuracy loss.

Using 8-bit integers instead of floating-point numbers to represent model weights and activations.

A mathematical optimization technique that finds the best solution among discrete options subject to linear constraints.

A mathematical property ensuring that estimated demand relationships are economically consistent and don't violate basic economic laws.

A class of distance measures between probability distributions that use function classes to define divergence.

The ability of an AI system to understand and match user goals, especially when requirements are unclear or evolving.

The process of analyzing user input to determine what the user is trying to accomplish so it can be handled appropriately.

The process of identifying and structuring the user's underlying goal or request from natural language input.

The process of users clarifying and developing their goals through interaction rather than starting with fully-formed objectives.

The model's capability to understand what a developer actually wants to accomplish, even when the request is vague or expressed in informal language.

Specifying what you want to accomplish rather than writing detailed code to implement it.

A measure of how consistently multiple human annotators label or judge the same data.

Dependencies and relationships between different variables or channels in multivariate data.

A measure of how consistently different judges rate the same outputs, typically using metrics like correlation or ICC.

The spatial, functional, or semantic relationships and dependencies between different parts of a composed object.

A measure of how consistently different evaluators score or judge the same items, often using metrics like Kendall's tau.

Ensuring that learning signals from different tasks contribute equally to model updates, preventing any single task from dominating training.

A measure of how much multiple teacher models agree on their predictions, used to assess supervision reliability.

A model's understanding of how conversations naturally flow and how users respond to assistant outputs.

The sequence of past user actions and system responses that inform current decision-making.

An AI tool designed for back-and-forth collaboration with humans, refining intent and outputs through dialogue.

A conversational interface where users can ask follow-up questions and receive responses based on previous context, rather than just one-shot predictions.

Training a policy by having humans intervene and correct the robot, then learning from those corrections.

Combining insights and methods from multiple academic disciplines to solve problems in a target domain.

The ability to mix images and text in any order within a single prompt, rather than requiring all images first or all text first.

Giving feedback at multiple steps during reasoning, not just at the final answer, to guide the model's thinking process.

A deliberate step-by-step thinking mechanism that occurs before generating a response, helping the model work through complex problems more carefully.

The hidden patterns and knowledge stored inside a model's layers that it uses to understand and generate text.

A hidden computation phase where the model reasons through a problem before producing its final answer, improving accuracy on complex tasks.

Whether a study actually measures what it claims to measure, without confusing factors distorting the results.

The ability to understand and explain how a model makes decisions and what it has learned from its training data.

Machine learning models designed to be understandable to humans, showing why they make specific predictions.

Determining the appropriate moment to interject in a conversation based on natural dialogue cues.

The ability to decide when an agent should proactively act, when to seek user consent, and when to remain silent.

Whether a model applies the same reasoning strategy when classifying different instances of the same category.

Ensuring that related elements (like a person's face across frames) maintain consistent properties throughout.

The degree of disagreement in how different models within the same modality (e.g., vision models) represent a single stimulus.

Measuring how similar consecutive frames or audio segments are within a single modality.

Changes in paralinguistic features within a single spoken sentence, like shifting emotion mid-sentence.

Breaking down an image into fundamental components like albedo (color), shading (lighting), and residuals (fine details).

The geometric properties of a space as measured from within, independent of how it's embedded in higher-dimensional space.

A reward signal that encourages an agent to explore and discover new states, separate from task-specific rewards.

Reward signals generated from the model's own internal signals, like confidence scores, rather than external verification.

A change that preserves key properties or predictions of a model.

A specification that defines preconditions and postconditions for tool calls to prevent invalid action sequences.

Predicting what inputs or earlier program states must have been to produce a given output.

Finding the input that produces a known output, when the forward process is complex or many-to-one.

Finding input causes from observed output effects, often ill-posed.

Working backward from a desired outcome to determine what actions would produce that result.

A reward signal that measures quality by having an LLM recover the original task specification from generated outputs.

A technique that reweights observations to remove confounding bias by accounting for treatment assignment probabilities.

A data structure that maps terms to the documents containing them, enabling fast keyword-based search similar to how a book's index works.

A search technique that maps vocabulary terms to documents containing them, enabling fast keyword-based lookups commonly used in search engines.

An AI system that shapes decisions and outcomes without users recognizing its influence on the information or criteria they use.

Errors or misalignments in AI outputs that go undetected because the user accepts the result without critical evaluation.

A measure of how quickly ions move through a material, critical for battery charging and discharging speed.

Graphs showing model performance across different configurations while keeping total computational operations constant.

An unsupervised algorithm that isolates anomalies by randomly selecting features and split values.

A property that remains the same for graphs with identical structure, regardless of how nodes are labeled or arranged.

Statistical method to estimate latent abilities, question difficulty, and model proficiency from test performance.

The process of gradually removing noise from a noisy input through multiple refinement steps to generate clean outputs.

A workflow where code is refined through multiple rounds of small, targeted changes rather than complete rewrites.

Repeatedly improving an output by generating versions, evaluating them, and using feedback to create better versions.

A process where the model performs multiple rounds of web searches, each building on previous results to refine and deepen its understanding of a topic.

A technique that limits how much a model's output changes when inputs change slightly, making it more stable and predictable.

Crafting adversarial inputs designed to bypass a model's safety guardrails and trigger harmful outputs.

The process of breaking Japanese text into meaningful units (tokens), accounting for the language's unique writing systems including kanji, hiragana, and katakana.

A self-supervised learning approach that predicts future embeddings from video without reconstructing pixels.

Converting code to machine instructions at runtime, enabling Python code to run efficiently on GPUs.

A training approach where a model learns to predict missing parts of video by understanding both spatial and temporal patterns without reconstructing actual pixels.

A shared mathematical space where different types of data (like sounds and text descriptions) are represented so similar concepts are positioned close together, enabling direct comparison.

A shared numerical space where different types of data (such as audio and text) are represented together, allowing the model to find relationships between them.

Processing multiple input types together in an integrated way rather than separately, allowing the model to reason about how they relate.

An unsupervised algorithm that groups data points into k clusters by minimizing distance to cluster centers.

The raw frequency domain data collected directly by an MRI scanner before conversion to images.

A technique to analyze neural networks by identifying which neurons or experts are most important for specific tasks.

A recursive algorithm that estimates the state of a dynamic system by optimally combining noisy measurements with a mathematical model.

A nonparametric method for estimating survival curves from censored data without assuming a specific distribution.

A non-parametric method that estimates probability distributions by smoothing data points with kernel functions.

Combining multiple GPU operations into a single optimized computation to reduce memory overhead and improve speed.

Tuning kernel functions to improve performance in kernel-based models.

A mathematical framework using reproducing kernel Hilbert spaces for classification and regression with theoretical guarantees.

Internal memory structures in transformers that store computed representations to speed up inference and enable agent communication.

Attention mechanism components that store and retrieve information; fewer heads means reduced model capacity and faster computation.

A reference frame in a video that serves as an anchor point for propagating edits or information to surrounding frames.

Matching specific visual landmarks (like object corners) between a demonstration and a new scene to align actions.

The task of automatically identifying and locating distinctive points of interest in an image that remain stable across different angles and lighting conditions.

A measure of how different one probability distribution is from another, used to evaluate sampling quality.

Assessing models using external knowledge sources for better judgment.

A structured or unstructured collection of documents and facts that a system retrieves from to answer queries.

The limit to how much factual information a model can reliably know or recall, often constrained by its size and training data.

A discrete unit of knowledge or skill that can be identified and measured in student work.

The process of organizing, storing, and synthesizing insights from multiple experiments to improve future decision-making.

The date up to which a model has been trained on data; it cannot reliably answer questions about events or information after this date.

A technique that compresses a large, complex model into a smaller one by training the smaller model to mimic the larger model's behavior.

An agent's ability to recognize what information or skills it lacks to solve a problem.

A structured database that stores facts as relationships between entities (like 'Einstein' connected to 'Physics'), enabling machines to reason about real-world knowledge.

The task of filling in missing facts or relationships in a knowledge graph by predicting what connections should exist based on patterns in existing data.

Monitoring and recording what a student has demonstrated they understand over time.

Applying knowledge learned from one task to improve performance on another.

Requiring external factual information beyond what is directly observable to solve a task correctly.

Incorporating domain expertise or physical laws into machine learning models to improve accuracy and generalization.

A neural network architecture designed to provide flexible, expressive function approximation with interpretable structure.

A mathematical operator that transforms observable functions of a dynamical system to reveal its underlying structure and eigenvalues.

A mathematical way to describe quantum operations that guarantees they produce physically valid quantum states.

An efficient but approximate method for parameterizing doubly stochastic matrices that sacrifices some expressivity for computational speed.

A mathematical property that guarantees convergence of optimization algorithms to stationary points.

A store for previously computed key-value pairs that speeds up text generation in transformers.