ThinkLLM
ModelsCapabilitiesUse CasesBenchmarksPapersGlossary
ModelsCapabilitiesUse CasesBenchmarksPapersGlossary
AboutPrivacyTermsRSS

ThinkLLM

Spot an error in our data? Let us know.

Papers

Recent AI research papers with accessible summaries. Updated daily from arXiv, summarized for developers who don't read papers regularly.

2957 papers11 this month12 topics
AllEvaluation 48Efficiency 37Reasoning 35Training 33Agents 28Applications 26Safety 19Multimodal 11Architecture 10Data 8Alignment 5scaling 2

Nov 2 – Nov 8(11)

SwarmEvolve: Latent Agent Swarms for Evolutionary Discovery and Optimization

Nov 5, 2026

Qile Jiang, George Em Karniadakis

Agent swarms communicating in latent space without text generation can solve complex optimization problems 7-20x faster than traditional LLM-guided search, even with smaller models.

SwarmEvolve uses hundreds of lightweight AI agents working together in swarms to improve programs for math and science problems. Instead of generating long text explanations, agents reason directly in latent space and share information through KV caches, with a novelty filter ensuring diverse ideas. The system is 7-20x faster than existing methods while using smaller models.

agentsreasoningefficiency

A Hidden Markov Preference Model for Online Prediction of Adversary Actions

Nov 5, 2026

Angel Daruna, Michael Cogswell, Tixiao Shan et al.

By modeling attackers as switching between behavioral regimes and constraining predictions to feasible actions, defenders can predict the next attack step 3x faster than recurrent baselines after detecting a strategy change.

This paper develops a hidden Markov model to predict an attacker's next action in cybersecurity scenarios. The model tracks the attacker's behavioral regime (which can switch during an attack) and handles actions with state-dependent arguments that may not always be available.

Oct 26 – Nov 1(44)

Latent Reasoning in Continuous Space for Unified Multimodal Models

Oct 30, 2026

Byungwoo Jeon, Yoonwoo Jeong, Hyunseok Lee et al.

Reasoning in continuous latent space instead of discrete tokens unlocks test-time scaling and better controllability for image generation in multimodal models, showing 7% improvement on GenEval benchmarks.

This paper introduces LARC, a framework that enables unified multimodal models to perform reasoning directly in continuous space rather than discrete tokens, bridging the gap between language-based reasoning and image generation.

reasoningmultimodaltraining

Flide: Accelerating Decision Inference with Exact State Forking

Oct 30, 2026

Bobo Li, Jiajia Song, Hao Fei et al.

For LLM agents making categorical decisions, Flide eliminates redundant computation by executing multiple decision branches simultaneously in one pass, delivering both speed and numerical precision that existing inference engines can't match.

Flide is an inference engine that speeds up decision models used in LLM agents by efficiently handling multiple decision branches in a single forward pass. Instead of recomputing shared parts of the model multiple times, it uses a technique called ForkAttention to track different decision paths through GPU memory, achieving up to 9.5x speedup while maintaining mathematical accuracy.

Oct 19 – Oct 25(32)

Learning Positive Linear Combinations of ReLUs in Fully Polynomial Time

Oct 23, 2026

Ilias Diakonikolas, Jelena Diakonikolas, Puqian Wang et al.

You can now learn ReLU-based models in polynomial time using tensor-based filtering and boosting—a theoretical breakthrough that removes computational barriers for this fundamental learning problem.

This paper presents the first polynomial-time algorithm for learning positive linear combinations of ReLU activations under Gaussian distributions. The key innovation is a boosting procedure combined with implicit tensor methods that efficiently recovers hidden weight vectors, solving a problem that previously required super-polynomial time.

trainingreasoning

SemantEq: Faithful Bidirectional Equivalence of Formal Statements

Oct 23, 2026

Xinrui Li, Pawan Sasanka Ammanamanchi, Siddharth Bhat et al.

When autoformalization translates math to formal proofs, checking logical equivalence isn't enough—SemantEq instead verifies that statements transform into each other only through semantically meaningful steps, achieving 100% precision on detecting non-equivalent pairs while improving recall fr...

This paper introduces SemantEq, a method to verify whether formal mathematical statements in proof assistants like Lean 4 faithfully translate their natural language sources. Rather than checking logical equivalence (which is too permissive), SemantEq requires statements to transform into each other only through specific allowed steps like logical rearrangement and library equations.

Oct 12 – Oct 18(13)

DiscoveryRSI: Recursively Learning to Discover Valuable Hypotheses

Oct 16, 2026

Jianyuan Zhong, Chenwei Wu, Zhaochen Yu et al.

By treating hypothesis generation as an iterative learning problem with explicit feedback loops, AI can progressively improve the quality and testability of scientific claims, achieving better alignment with real published research than existing models.

DiscoveryRSI is a framework that helps AI systems generate and improve scientific hypotheses through recursive learning. It uses multi-agent search to develop hypothesis trees, tests them against criteria, and learns from both successes and failures to refine future research workflows. The system shows measurable improvements in hypothesis quality across life-science domains.

reasoningagentsevaluation

Task-aligned spectral inductive biases in parameterized quantum circuits

Oct 16, 2026

Felix Frohnert, Adrián Pérez-Salinas, Sebastian Lange et al.

Quantum circuit architecture can encode task-specific spectral structure as a fixed parameter that reshapes the circuit's learning space, improving performance on frequency-based tasks without requiring more data or parameters.

This paper shows how quantum circuits can be designed with built-in biases that match specific tasks by encoding a fixed 'spectral query' alongside variable input data. The query acts like a tuning knob that changes which frequency combinations the circuit learns, without adding new information—similar to how radar systems probe signals at specific frequencies.

reasoningagentssafety

Understanding AI from the Inside: The State of Mechanistic Interpretability for AI Safety

Nov 5, 2026

Fazl Barez, Tung-Yu Wu, Hunar Batra et al.

Mechanistic interpretability provides tools to understand AI model internals, but understanding alone doesn't guarantee safety—it must be combined with behavioral evaluation and other safety measures to be effective.

This comprehensive survey examines mechanistic interpretability—the study of how neural networks work internally—and whether understanding a model's inner workings actually makes it safer. It covers foundational concepts like features and circuits, surveys analysis methods, and explores how this understanding can help with safety through techniques like knowledge editing and steering.

safetyevaluationreasoning

Data-driven discovery of intrinsic Lie point symmetries of differential equation systems

Nov 5, 2026

Ilker Gurcan, Kadir Ustun, Ramazan Gokberk Cinbis

You can discover hidden geometric symmetries in differential equations from data alone, enabling model reduction and better understanding of complex systems without needing closed-form equations.

This paper presents a machine learning method to automatically discover continuous symmetries (Lie point symmetries) directly from differential equation data. Rather than requiring symbolic equations, the approach uses operator theory on sampled data to identify geometric structures that can simplify models and improve scientific machine learning applications.

reasoningdata

FlowReconBench: Accuracy, calibration and cost of learned and classical flow reconstruction from sparse sensors

Nov 4, 2026

Nicolas Tricard, Linzheng Wang, Jason Chen et al.

No single flow reconstruction method works best everywhere—classical approaches dominate in well-constrained problems, but learned models become essential when sensors are sparse or poorly placed. Generative models offer better uncertainty quantification at higher computational cost.

FlowReconBench compares 15 methods (10 learned, 5 classical) for reconstructing complete fluid flow fields from sparse sensor data across three test scenarios. The study reveals that the best approach depends on the problem: classical methods excel with dense sensors and simple flows, while neural networks shine with sparse or poorly-positioned sensors.

evaluationreasoning

FrontierPhysics: Evaluating Agents for End-to-End Physics Research

Nov 4, 2026

Bingran You, Wenjun Ke, Ria Rosenauer et al.

Current AI agents struggle with frontier physics research not because they lack physics knowledge, but because they reason insufficiently rigorously and lack the expert intuition that guides real researchers—suggesting the bottleneck is in reasoning quality and domain expertise transfer, not fa...

FrontierPhysics is a benchmark that evaluates AI agents on end-to-end physics research tasks created by real physicists. Unlike existing benchmarks that test isolated problems, it requires agents to review literature, plan research, conduct experiments, and write manuscripts across theoretical, computational, and experimental physics.

evaluationreasoningagents

SpecCompact: Accelerating LLM Agent Context Compaction with Slot-Parallel Semantic Speculative Verification

Nov 4, 2026

Yulu An, Yifei Feng, Xiangyang Yin

Parallel drafting with verification can make context compression 2-5x faster for AI agents, enabling them to handle longer interactions without sacrificing speed or information quality.

SpecCompact speeds up how AI agents compress long conversation histories by using a small model to draft multiple summary sections in parallel, which a larger model then verifies and fixes. This reduces the time agents spend summarizing while keeping important information intact, achieving 2-5x faster compression on real agent tasks.

efficiencyagentsreasoning

On Turn-Level Best-of-N Sampling for Scaling LLM Agents

Nov 4, 2026

Jalaj Bhandari, Dhawal Gupta, Hongli Zhan et al.

Turn-level best-of-N sampling with off-the-shelf LLMs can improve agent performance on complex tasks without retraining—just by having a selector model pick the best action at each step.

This paper tests a simple inference-time technique where one LLM generates multiple candidate actions and another LLM selects the best one at each step of a long-horizon task. Testing on complex agent tasks shows 5-10% accuracy improvements and more consistent behavior, with agents making fewer attempts and better following instructions.

agentsreasoning

Reading Why: Born-Measured Attention with Exact Word Evidence on Light and Qubits

Nov 3, 2026

Teerawat Tongloy, Jiraporn Pongsopa, Rangsan Jomtarak

By replacing softmax attention with quantum Born measurements, you get a transformer where word contributions to predictions are mathematically exact and sum perfectly to the output, enabling perfect interpretability while maintaining competitive accuracy.

This paper builds a transformer where attention uses quantum Born rule mechanics (|⟨q|k⟩|⁴) instead of softmax, making word contributions to predictions mathematically exact and verifiable.

reasoningevaluation

Adversarial Speculative Reasoning Improves Long Horizon Agents

Nov 3, 2026

Hongyin Luo, Wei Fang, Lijie Yang et al.

Speculative decoding isn't just a speed trick—it changes how models reason over long horizons. By training draft models to fail at critical moments, you can force the main model to handle important reasoning steps with better precision, improving agent accuracy on complex tasks.

This paper shows that speculative decoding—a technique usually just for speed—actually affects reasoning quality in long-horizon AI agents. The authors discovered that numerical noise accumulates during the verification process, degrading reasoning accuracy over many steps.

reasoningefficiencyagents

A Circuit for In-Context Learning

Nov 3, 2026

Andrew Gao, William Gao

In-context learning works through a mechanistic circuit of identifiable components (memory storage, key matching, gating) that you can trace causally from input to output, and this same circuit appears consistently across model sizes and architectures.

Researchers reverse-engineered how transformers perform in-context learning by identifying the specific circuit of attention heads and signals that enable few-shot classification.

reasoningtraining
efficiencyagentsreasoning

Hierarchical World Models with Implicit Dynamics

Oct 30, 2026

Gaoyue Zhou, Yvonne Wu, Zichen Jeff Cui et al.

Decoupling state reachability from action execution in hierarchical world models dramatically reduces planning errors for long-horizon robotic tasks—achieving 135% improvement over single-level approaches by prioritizing local feasibility over global behavioral patterns.

This paper presents ID-WM, a hierarchical world model for robotic planning that separates high-level state transitions from low-level action execution. A high-level model identifies reachable subgoals across the state space, while a low-level inverse dynamics model converts these subgoals into precise actions.

reasoningagentsarchitecture

RoboArchitect: Automating the Design of Robotic Agents with Persistent Memory

Oct 30, 2026

Shengguang Wu, Suning Huang, Cherie Ho et al.

You can dramatically improve robot performance by automating harness design—the control logic wrapper around a policy—rather than retraining the policy itself, achieving specialist-level performance from general-purpose models.

RoboArchitect automates the design of control harnesses around frozen robot policies using a coding model that observes failures and iteratively revises the harness code. Without updating model weights, it improves success rates 4.5x on unseen scenes and 3.2x on unseen tasks by automatically discovering strategies like obstacle avoidance and persistent memory for learned behaviors.

agentsreasoningapplications

Hyperbolic Inverse Reinforcement Learning

Oct 30, 2026

Iulia Rusu, Anoop Praturu, Sergei Gepshtein et al.

Using hyperbolic geometry to parameterize reward functions in IRL can capture non-separable feature interactions better than standard approaches, especially when training data is limited or environments are dense.

This paper introduces Hyperbolic Inverse Reinforcement Learning, which infers reward functions from observed behavior using hyperbolic geometry instead of traditional linear or neural network approaches.

trainingreasoning

RAMA: Retrieval-Augmented Multi-Agent Routing with Episodic Memory for LEO Satellite Networks

Oct 30, 2026

Xin Ming Li, Peng Hu

Adding episodic memory retrieval to multi-agent reinforcement learning dramatically improves routing decisions in dynamic satellite networks, enabling agents to reuse successful past strategies without retraining.

RAMA is a routing system for satellite networks that combines reinforcement learning with memory retrieval. Instead of just learning from current observations, each satellite remembers successful routing decisions and retrieves similar past experiences when making new routing choices. This approach reduces delivery delays by 13-93% compared to existing methods, especially as networks scale up.

agentsreasoningefficiency

Elucidating the Design Space of LM Data Attribution

Oct 29, 2026

Steven Yu, Sheng-Yu Wang, Kimia Hamidieh et al.

Data attribution for LMs works much better when you focus on the most informative tokens within documents using negative log-odds gradients instead of averaging attribution scores across entire documents—a simple fix that improves performance by 35-72 percentage points.

This paper diagnoses why data attribution methods for language models fail by creating a controlled benchmark with synthetic training data.

evaluationtrainingreasoning

Rethinking Neural Thickets Without Weight Perturbations

Oct 29, 2026

Nicolas Dufour, Elouan Gardès, Amil Dravid et al.

You can get ensemble-like improvements from frozen models by adding noise to predictions and voting, rather than perturbing weights—making it easier to deploy pretrained models without modification.

This paper challenges the Neural Thickets hypothesis by showing that the performance gains from perturbing pretrained model weights don't require the proposed geometric explanation. Instead, simply adding random noise to output logits and selecting/voting on predictions achieves similar or better results—without any weight changes.

reasoningevaluation

From extreme weather to agricultural consequences with agentic AI

Oct 29, 2026

Minghan Yu, Mengyu Dai, Yixin Wen et al.

Agentic AI can bridge the gap between weather forecasting and agricultural impact prediction by combining multiple data sources through causal reasoning, enabling faster disaster response for farmers and policymakers.

WxCrop is an AI agent that predicts crop losses from extreme weather by combining weather forecasts with agricultural data. It uses physically interpretable weather indices and causal analysis to estimate which crops will be damaged and by how much, validated on hurricane, freeze, and drought impacts in Florida with results matching insurance records and production data.

agentsapplicationsreasoning

Probing Spatial State Representations in Visual Foundation Models

Oct 29, 2026

Baiqiao Yin, Zhongang Cai, Ruisi Wang et al.

Visual foundation models excel at recognizing visible scenes but struggle with spatial persistence, predictive updates, and action guidance—capabilities essential for embodied AI. Strong readout accuracy doesn't transfer to maintaining spatial state across time or motion.

This paper investigates whether visual foundation models learn spatial representations that support persistent spatial awareness—tracking objects beyond the current view, predicting future scenes, and guiding actions.

evaluationmultimodalreasoning

Testimony from Timelines: Evidence-Grounded Significance Testing of Temporal Clinical Hypotheses

Oct 29, 2026

Chengyang He, Wenlong Zhang, Koduvayur Subbalakshmi et al.

Language models can sound confident about temporal patterns in data without doing proper statistical testing—this benchmark reveals they fail mainly at selecting relevant evidence, not at running tests, and a structured retrieval-then-test approach significantly improves accuracy.

This paper introduces STRATA, a benchmark for evaluating language models on temporal clinical hypothesis testing. Rather than just grading final answers, it evaluates every step: choosing which patient records matter, selecting the right statistical test, and interpreting results.

evaluationreasoningdata

When to Assist and What to Suggest: Joint Gating and Memory-Grounded Hint Generation in One Compact Decoder

Oct 29, 2026

A. J. A. Jubayer Talukder, Mohammad Ariful Haque

A single small model can effectively decide when to assist and generate helpful hints by leveraging memory grounding, outperforming larger multi-model systems on speed while maintaining quality—memory context matters far more than retrieval or model size.

This paper tackles proactive assistance in conversations by combining gating (deciding when to help) and hint generation (what to suggest) into a single compact model. The authors show that a 1B parameter decoder can match larger two-model pipelines while being 1.31-1.76x faster, with memory-grounded hints being the key driver of quality.

efficiencyapplicationsreasoning

Collaborative AI agents for autonomous experimental research with solid-state quantum spins

Oct 29, 2026

El Mustapha Mansouri, Eikichi Kimura, Masato Koga et al.

AI agents can conduct rigorous science when designed to scrutinize each other's claims, validate theoretical predictions experimentally, and reject results that don't meet acceptance criteria—moving beyond generating plausible-sounding conclusions to actually testing them.

Researchers built an autonomous lab where AI agents collaborate to conduct full research cycles—from formulating questions to running experiments on quantum systems. The agents challenge each other's work, design experiments to test predictions, and crucially, reject inconclusive results and improve protocols.

agentsreasoningevaluation

VeriCUA: Verification as Agentic Exploration for Computer-Using Agents

Oct 29, 2026

Qiushi Sun, Da Yu, Alexander Knop et al.

Instead of asking a model to judge task completion from a single screenshot, VeriCUA treats verification as active exploration—the verifier can interact with the interface and run code checks to gather decisive evidence, improving accuracy especially on hard cases.

VeriCUA is a framework that verifies whether computer-using agents have completed tasks by actively exploring digital environments to gather evidence, rather than relying on static judgments. It works across different platforms (desktop, mobile, web) and combines GUI interactions with code inspection to build a complete picture of task completion.

agentsevaluationreasoning

PIVOT: Progressive Intervention via Online Teacher Correction

Oct 29, 2026

Zhihao Yang, Boyuan Zheng, Puhui Tan et al.

Combining trajectory-level intervention (teacher corrections inserted into rollouts) with token-level supervision (learning from corrected segments) is more effective than either approach alone for training on hard reasoning tasks.

This paper addresses a key limitation in reinforcement learning from verifiable rewards (RLVR): when all attempted solutions fail, the model gets no learning signal. PIVOT combines two strategies—inserting teacher corrections into student attempts and using supervised learning on those corrections—to help models learn from harder problems where pure reward signals are too sparse.

trainingreasoningalignment

Training Foundation Models with a Single Algorithm

Oct 29, 2026

Yuda Song, Alvin Zhang, Fahim Tajwar et al.

Foundation models don't need separate training algorithms for different stages—you can train everything with one likelihood-based method by treating RL objectives as fitting to model-generated distributions, which improves efficiency and capability coverage.

This paper shows that foundation model training stages (pretraining, fine-tuning, reinforcement learning) can be unified under a single likelihood-based algorithm. Instead of switching between different training methods, the researchers reformulate RL objectives as maximum-likelihood fitting, allowing all stages to use the same core algorithm.

trainingefficiencyreasoning

Advancing Clinical Genomic Analysis with a Skill-Centred Agent

Oct 29, 2026

Qiaoyu Zheng, Weike Zhao, Yanjie Fan et al.

Specialized AI agents built from curated domain skills can meaningfully augment expert decision-making in high-stakes clinical work, achieving better results than general LLMs by combining retrieval, orchestration, and executable tools.

ClinGeX is an AI agent that helps clinical genomicists analyze patient DNA sequences by orchestrating 105 specialized skills to interpret genetic variants and guide treatment decisions. In tests with real cases and clinical experts, it achieved 71.7% accuracy compared to 45.7% for competing systems, and significantly improved expert performance when used collaboratively.

agentsapplicationsreasoning

Ramanujan Challenge Results: AI-Assisted Evaluation Pipeline for Multi-Format Proofs

Oct 29, 2026

Alan Sorani, Ashvni Narayanan, Michael Shalyt et al.

Evaluating AI's mathematical reasoning requires specialized pipelines beyond single-pass LLM judgment—combining multiple verification methods dramatically improves accuracy on research-level problems.

Researchers evaluated AI models and formal tools on the Ramanujan Challenge, a benchmark of research-level number theory problems. They developed an AI-assisted pipeline to reliably judge complex proofs in multiple formats (natural language, code, formal proofs), reducing false positives from 23% to 6% compared to simple LLM evaluation.

evaluationreasoningapplications

Retrieval as Traversal: Value-Indexed Memory for Stateful Long-Context Agents

Oct 29, 2026

Daniel Strauß, Zhenshan Bing, Alois Knoll

Instead of searching all memories every forward pass, stateful retrieval navigates locally through memory neighborhoods, potentially making long-context reasoning more efficient by exploiting overlap between consecutive retrieval problems.

This paper proposes a new approach to memory retrieval for long-context AI agents that treats retrieval as persistent navigation through an address space, rather than searching from scratch each time. By separating memory content from its location and charging for access cost, the method encourages related memories to cluster together, reducing redundant retrieval work across similar tasks.

reasoningefficiencyagents

On the Fundamental Limits of Prefill-Based Decoding Length Prediction

Oct 29, 2026

Haowen Shi, Xuzhe Yan, Fahao Chen et al.

Prefill-based decode-length prediction has hard limits set by autoregressive stochasticity, not predictor design—changing the model, data, or sampling breaks existing benchmarks because they measure task difficulty, not predictor quality.

This paper reveals why token-length prediction for LLMs fails when conditions change, showing the problem isn't bad predictors but fundamental randomness in autoregressive decoding. The authors derive mathematical bounds on prediction accuracy and show that decode length follows a normal distribution determined by the model's stopping probabilities, not the predictor architecture.

efficiencyevaluationreasoning

BURGER: Benchmarking and Understanding Robotic Generalization via Embodied Drawing

Oct 29, 2026

Xiangyu Xu, Bangshuai Peng, Ziyi Liu et al.

VLA models show promise in learning scalable, language-conditioned robotic behaviors with multimodal action generation, but their success depends on task diversity, data scale, and how well they ground spatial language into continuous control.

BURGER is a robotic drawing benchmark that tests how well vision-language-action models learn diverse behaviors from scaled training data.

evaluationmultimodalreasoning

NEOPIC: A Neural Operator Framework for Particle-based Kinetic Plasma Simulations

Oct 29, 2026

Sriramkrishnan Muralikrishnan, Chelsea Maria John, Jeffrey Kelling et al.

Neural operators can efficiently replace expensive field solvers in particle simulations by learning directly from particle data, with proven error bounds and linear computational complexity independent of mesh resolution.

NEOPIC is a neural operator framework that learns to replace the field solver in particle-based plasma simulations. Unlike grid-based approaches, it operates directly on particle positions without requiring a mesh, avoiding noise aliasing issues. The framework is plug-and-play, works with different simulation codes, and comes with theoretical guarantees on accuracy and computational complexity.

architectureefficiencyreasoning

A Low-Rank Space Captures Linear Alignment Between LLM and Human Neural Activity During Naturalistic Language Listening

Oct 29, 2026

Zach Paris, Emily Finn

LLM-brain alignment doesn't require full model representations; a tiny low-dimensional subspace captures most neural predictivity, and these dimensions differ from what PCA identifies as important, revealing that models align with brains through specific, interpretable features rather than their ...

This paper shows that LLMs align with human brain activity during language listening through a surprisingly small subset of their representations—just 1-3% of dimensions.

evaluationreasoning

Algebraic World Model: From Exploration to Proofs

Oct 29, 2026

Guancheng Pan, Qixin Tan, Yang Yuan et al.

By combining environment exploration, hypothesis testing, and formal verification in Lean, AWM can rigorously prove that goals are unreachable—even beyond observed trajectories—enabling certified reasoning about unfamiliar worlds.

This paper presents Algebraic World Model (AWM), a system that explores unknown environments, builds executable Python hypotheses, and formalizes them as mathematical models in Lean 4 to prove what's possible and impossible. AWM discovered that certain game levels are unwinnable by constructing formal proofs, outperforming GPT and Claude on impossible-goal detection and game-solving tasks.

reasoningagentsevaluation

rTaste: Do Research Agents Have Taste? Evaluating Early Stage Research Judgment

Oct 29, 2026

Art Kanke

Current research agents can generate valuable experiments but lack reliable judgment about which ones to prioritize—their rankings don't meaningfully improve experiment selection compared to simpler baselines.

This paper evaluates whether AI research agents can effectively judge which experiments are worth running by having six frontier LLMs propose minimum viable experiments for NLP papers, forecast their success, and rank them.

evaluationagentsreasoning

ThinkWorld: World Imagination Reinforces Chain-of-Dynamics Reasoning

Oct 27, 2026

Chaehyun Kim, Chi-Pin Huang, Jisoo Kim et al.

By grounding action reasoning in world dynamics rather than explicit action labels, robots can learn from unlabeled human videos alongside labeled robot data, dramatically expanding the training data available for robotic manipulation.

ThinkWorld is a framework that teaches robots to learn manipulation skills from both labeled robot videos and unlabeled human videos. It represents actions as sequences of motion tokens (Chain-of-Dynamics) that predict how the world will change, then uses reinforcement learning to strengthen predictions that match real outcomes.

agentsmultimodalreasoning

Attention as Lattice Gauge Theory: Relations Interact Only Through Wilson Loops

Oct 27, 2026

Aakash Sarkar, Ananya Mukherjee, Motoshige Sato

Attention mechanisms aren't arbitrary design choices—they're constrained by symmetry principles from physics.

This paper reveals that attention mechanisms in neural networks follow the mathematical structure of gauge theory when entities can rotate independently. The authors show that relations between entities can only interact through closed loops (Wilson loops), leading to a unique form of attention called holonomy attention.

architecturereasoning

GEPA-Curator: Curated Learning for Self Improving Agents

Oct 27, 2026

Zack Yu, Nicholas Lee, Minjae Lee et al.

Not all agent failures are equally useful for learning—curating which tasks inform prompt optimization, and providing successful examples as context, significantly improves agent performance in complex multi-turn tasks.

GEPA-Curator improves how LLM agents learn from their mistakes by being selective about which failed tasks to learn from. Instead of learning from all failures, it filters out noisy or unreliable learning signals and retrieves successful past examples to guide prompt optimization, achieving up to 20.7% performance gains over existing methods.

trainingagentsreasoning

Can LLMs Uncover Hidden Physical Laws through Free Exploration?

Oct 27, 2026

Zhen Yang, Wen Wang, Xander Xu et al.

Current LLMs rarely discover novel physical laws through autonomous exploration, even when given tools and freedom to experiment, suggesting fundamental gaps in their ability to conduct genuine scientific inquiry without explicit guidance.

Researchers created LongChang, a framework that tests whether LLMs can discover hidden physical laws by freely exploring simulated worlds with altered physics. LLMs act as autonomous scientists, noticing unexpected phenomena (like objects falling at different speeds) and designing experiments to uncover the rules—without being told what to investigate.

reasoningagentsevaluation

Multi-State Linear Attention with Generalized Second-Order Routing

Oct 27, 2026

Hee Seung Hwang, Sanghyuk Chun, Xindi Wu et al.

By incorporating variance information into how segments are aggregated in linear attention models, MERLIN enables better long-context understanding without the quadratic complexity of standard attention—useful for scaling LLMs to longer sequences.

This paper improves Multi-State Linear Attention (MSLA) for long-context language models by tracking both the mean and variance of key alignments across segments, rather than just the mean.

efficiencyarchitecturereasoning

Metacognitive Decoding: Efficient Inference for Diffusion Language Agents

Oct 27, 2026

Mingyu Zhao, Shuhang Lin, Ligong Han et al.

Diffusion models can generate multiple tokens in parallel, but you need a smart controller to decide when to stop and how much to generate; this work shows you can cut inference costs by ~20% by training a lightweight decision module on agent trajectories.

This paper introduces a metacognitive controller that optimizes how diffusion language models generate text during agent interactions. By learning when to stop generating tokens and how long responses should be, the controller reduces computational cost by 20.6% while maintaining task performance—making parallel token generation more practical for real-world agents.

efficiencyagentsreasoning

Rare, Relevant, and Still Wrong: How Memory Fails LLM Group Decisions

Oct 27, 2026

Anouar Imel, Imad Charradi, Abderrahman Sghir et al.

When LLMs make group decisions from compressed discussion summaries, the memory strategy matters less than the actual evidence balance it preserves—and both common heuristics (rarity and relevance) fail to preserve it correctly.

This paper reveals how LLM group decisions fail when discussions are compressed into memory. Testing two common strategies—keeping rare claims or relevant claims—the authors show both fail predictably on hidden-profile tasks.

reasoningevaluationagents

VideoLVR: Latent Visual Reasoning over Grounded Video Evidence

Oct 27, 2026

Shilong Li, Xingyuan Bu, Hao Shi et al.

Instead of describing or reinserting video regions during reasoning, VideoLVR reconstructs visual tokens from relevant moments—letting models "see" evidence again while staying efficient.

VideoLVR enables video understanding models to reason over video content by reconstructing visual evidence as latent tokens during reasoning steps, rather than relying on text descriptions or expensive visual crops. This approach keeps visual information accessible without bloating the context, improving performance on fine-grained video reasoning tasks.

multimodalreasoning

GAS-PFN: Graph-Aware Spatiotemporal Prior-Data Fitted Networks for Zero-Shot Traffic Forecasting

Oct 27, 2026

Joseph Natter, Long Vu, Yifan Zhang et al.

By explicitly encoding road network structure into the training process and using a two-stage inference design, you can build traffic forecasters that work on new networks without retraining—a major practical advantage for deploying to evolving city infrastructure.

This paper introduces GAS-PFN, a neural network that forecasts traffic flow without needing to be retrained for new road networks. It learns from synthetic traffic patterns during training, using graph structure (road topology) to understand how traffic moves between connected sensors.

reasoningarchitectureapplications

Understanding the limits of latent world models

Oct 27, 2026

Rodrigo Gonzalez Laiz, Steffen Schneider

World models can solve tasks without learning accurate representations of environment dynamics—task performance alone doesn't guarantee the model understands how the world actually works.

This paper challenges the assumption that world models learn accurate internal representations of environment dynamics. The authors introduce a formal definition of 'faithful' world models based on equivariance and propose a metric called latent rollout consistency to measure whether learned latent dynamics actually match the environment's true dynamics.

evaluationreasoning

REGER: Modeling inter-item Relations with Semantic IDs for Generative Recommendation

Oct 27, 2026

Yuxi Huang, Silin Zhou, Chenhao Wang et al.

By pretraining on semantic item relationships and using hierarchical preference decomposition, generative recommendation models can better capture both what items are similar and what users actually want to buy or interact with.

REGER improves generative recommendation systems by explicitly modeling relationships between items through semantic IDs. It pretrains models to learn item similarities, uses hierarchical attention to capture user history patterns, and applies preference distillation to better predict which items users will interact with next.

trainingreasoning

TeamGym: A Gym for Team Decision-Making by AI Agents with Distributed Information

Oct 27, 2026

Chanwoo Park, Jaywoong Jeong, Zachary Wojtowicz et al.

Current language models struggle with distributed team decision-making—they fail to effectively communicate and retrieve decision-relevant information. Training with reinforcement learning on final outcomes can improve coordination without explicit communication rewards.

TeamGym is a benchmark environment for training and evaluating AI agents on team decision-making tasks where information is distributed across participants. Using real-world data from legal, finance, medical, and software engineering domains, it measures how well language models share and integrate information.

agentsreasoningevaluation

Gym-Anything-RT: Scaling Real-Time Environments for Vision-Language Agents

Oct 27, 2026

Pranjal Aggarwal, Sean Welleck

Vision-language agents struggle dramatically with real-time interaction: GPT-4o's performance drops from 75% to 9% when environments run in real time instead of paused, exposing a major limitation in current models that excel at deliberative reasoning.

This paper introduces Gym-Anything-RT, a benchmark of 100 auto-generated minigames designed to test vision-language agents' ability to interact with environments in real time. The key insight is that real-time interaction—where the environment state changes while the agent reasons—is fundamentally different from deliberative tasks.

evaluationagentsreasoning

Collective Dynamics of Self-Reinforcement: How Cross-Task Coupling Drives Policy Collapse in Unsupervised RLVR

Oct 27, 2026

Peize Ding

Policy collapse in self-reinforced language models isn't a training bug but a phase transition caused by cross-problem coupling—it becomes inevitable beyond a critical dataset size unless the model's shared representations are explicitly protected.

This paper explains why unsupervised reinforcement learning from a model's own majority votes causes catastrophic collapse—where the model gives identical answers to all problems despite initially improving.

trainingsafetyreasoning

Kinematic Constraint-Referenced Evaluation for Articulated Object Manipulation

Oct 27, 2026

Baptiste Bonin, Caro Strickland, Audrey Durand

Binary success metrics hide whether manipulation failures come from poor strategy or from commanding movements the object's physics don't allow—CRS separates these by measuring constraint-aligned vs. wasted motion, enabling better policy diagnosis without needing demonstrations or learned judges.

This paper introduces Constraint-Referenced Strategy (CRS), a new way to evaluate robot manipulation policies that goes beyond simple success/failure. Instead of just checking if a task succeeded, CRS measures how well a robot's commanded movements align with an object's physical constraints—like how a door hinge only allows rotation around one axis.

evaluationreasoning

Retriever-Free Retrieval: Navigating Documents with Decision Models

Oct 26, 2026

Chao Wan, Paul Jünger, Beixi Du et al.

Language models can retrieve documents more effectively by making comparative judgments across candidates in shared context, rather than relying on fixed embeddings or lexical matching—eliminating the retriever bottleneck that even strong generators cannot overcome.

DecisionRAG replaces traditional retrieval with a decision model that navigates document hierarchies like browsing a library, evaluating multiple candidate passages simultaneously without generating tokens. It outperforms dense retrieval systems by 13-17 points on standard benchmarks while remaining fast (0.7-1.8 seconds per query on millions of documents).

reasoningefficiency

Understanding Jev-Style System One Models: Does Structure Reduce Selection Bias?

Oct 26, 2026

Mingyang Song, Mao Zheng, Zhiyuan Ma et al.

Jev-style models maintain repeatable decisions but fail to resist selection bias better than generative LLMs—stable predictions don't guarantee correct answers across equivalent option orderings, challenging assumptions about structured decision-making reducing bias.

This paper investigates whether Jev-style models—which predict probabilities directly over candidate options—reduce selection bias compared to generative LLMs.

evaluationsafetyreasoning

Pretraining Knows What, Simulation Teaches How: Compositional Generalization in a Sim-Trained Robot Agent

Oct 26, 2026

Benjamin Pikus, Daniel Saunders, Mircea Mironenco et al.

Pretrained language models can be efficiently adapted for robot control through simulation-based training alone, enabling strong real-world performance and compositional generalization without collecting real robot data.

ZAP is a robot control system that combines a pretrained vision-language model with simulation-based training to enable robots to understand tasks and execute them without real-world training data. The approach achieves 96.7% success on real-world manipulation tasks while being 3x faster and 17x cheaper than GPT-4 Astra, and generalizes to unseen objects, goals, and robot embodiments.

trainingapplicationsreasoning

Learning When to Commit: Schedule Distillation for Fast Diffusion Language Models

Oct 26, 2026

Aidar Myrzakhan, Zhiqiang Shen

Diffusion LLMs can generate much faster by strategically choosing which tokens to commit to first—prioritizing tokens that unlock predictions for other positions, not just the easiest ones to predict.

This paper introduces dSpark, a method for speeding up diffusion language models by learning which tokens to reveal first during generation. Instead of just predicting easy tokens, the approach identifies tokens whose revelation helps other positions become easier to predict, reducing the total decoding rounds needed.

efficiencytrainingreasoning
evaluationreasoningsafety

Direct Pass@$k$ Optimization for Thought Pruning: Balancing Quality and Diversity in Test-Time Reasoning

Oct 22, 2026

Mahdi Salmani, Adel Javanmard, Devansh Gupta et al.

Directly optimizing pass@k for pruning outperforms separate quality and diversity strategies by naturally balancing both objectives, achieving better results especially when keeping moderate numbers of candidate solutions.

This paper tackles test-time reasoning efficiency by proposing a pruning method that directly optimizes pass@k—the probability of finding at least one correct solution. Rather than separately balancing solution quality and diversity, the method unifies both objectives through a single metric that naturally rewards good partial solutions while penalizing redundancy.

reasoningefficiencyevaluation

METTA: Memory-Enhanced Test-Time Adaptation for Meta-Reinforcement Learning

Oct 22, 2026

Ali Najar, Hosein Hasani, Amir Kooshan Fattah et al.

By splitting task adaptation into fast within-episode memory updates and slower cross-episode summaries, METTA enables meta-RL agents to handle long interaction histories efficiently without the quadratic cost of transformer attention.

METTA proposes a memory-enhanced approach for meta-reinforcement learning that adapts quickly to new tasks by combining a causal neural memory for fine-grained within-episode learning with compact episode summaries for long-range task inference.

reasoningefficiency

Causal Chain of Thought: Mitigating Attention Leakage via Context Isolation

Oct 22, 2026

Gabriele Dominici, Teague McMillan, Fazl Barez et al.

Chain-of-thought explanations can be misleading because models can look up answers directly from context while generating plausible justifications. Isolating reasoning steps to only their causal dependencies makes explanations more faithful and reasoning more transparent.

This paper introduces CausalCoT, a method that improves chain-of-thought reasoning by breaking problems into isolated sub-questions that only see their direct dependencies.

reasoningevaluationsafety

Beyond Best-of-$k$: Prior-Consistent DAPS for Ambiguous Nonlinear Inverse Problems

Oct 22, 2026

Merve Kocabas, Benjamin A. Burns, Andreas Geiger et al.

When using diffusion models to solve inverse problems with multiple valid solutions, tracking prior consistency during sampling—not just measurement fit—prevents the solver from converging to measurement-consistent but prior-inconsistent modes.

This paper addresses a failure mode in diffusion-based solvers for inverse problems: when measurements have multiple valid solutions, existing methods can get stuck in solutions that fit the measurements but don't match the learned prior.

reasoningevaluationsafety

Action Learning for Scientific Simulations

Oct 22, 2026

Sarah Liaw, Yuanqi Du, Yilun Du

Action learning provides a unified, reusable objective for reasoning about dynamical systems from trajectory data—enabling path ranking, control, and inference without retraining for each new task.

This paper introduces LEAP, a method for learning stochastic actions from trajectory observations in dynamical systems. By anchoring unknown actions to a reference process and learning corrections through trajectory comparison, LEAP creates a differentiable objective that captures both local dynamics and long-term dependencies.

reasoningtrainingapplications

On the Token Collapse Phenomenon in Textual Graph Soft Prompting

Oct 22, 2026

Andrea Giuseppe Di Francesco, Rishabh Jain, Franco Maria Nardini et al.

Graph soft prompting methods suffer from token collapse, where learned graph tokens become indistinguishable across inputs and fail to convey graph structure; task accuracy alone is insufficient to validate that models actually use graph information.

This paper reveals a critical failure mode in graph soft prompting methods like G-Retriever, where graph-specific tokens learned by neural networks become nearly identical across different inputs—a phenomenon called token collapse.

evaluationreasoning

When to Follow the Teacher: Advantage-Based On-Policy Distillation

Oct 22, 2026

Mahdi Sabbaghi, George Pappas, Hamed Hassani et al.

You can make distillation more efficient by learning which teacher actions actually improve student performance, rather than blindly copying everything the teacher does—and you can estimate this using only the data you already collect.

This paper proposes Advantage-Based On-Policy Distillation (AOPD), which improves how students learn from teachers by selectively following teacher actions based on whether they actually help the student succeed.

trainingreasoning

Fine-Grained Analysis of Catastrophic Forgetting in Continual Learning via Algorithmic Stability

Oct 22, 2026

Guodong Zheng, Yan Sun, Peng Wang et al.

Catastrophic forgetting can be analyzed without assuming specific model architectures by tracking how sample perturbations propagate through tasks, revealing that learning-rate schedules and gradient alignment are key factors controlling how much previous knowledge is lost.

This paper develops a theoretical framework to understand catastrophic forgetting in continual learning—when learning new tasks hurts performance on old ones. Using algorithmic stability analysis, the authors decompose forgetting into measurable components and derive bounds showing how learning rates, task structure, and optimization dynamics affect memory retention across sequential tasks.

trainingevaluationreasoning

LongZero: Enhancing Long-Context Reasoning in LLMs from Zero Data

Oct 21, 2026

Zhennan Wan, Jianfei Chen

You can train LLMs to handle longer contexts better using only self-generated data—no external corpora or teacher models needed—by having the model create its own reasoning tasks and learning from them.

LongZero enables LLMs to improve their long-context reasoning abilities without external data by having models generate their own training tasks. The method creates multi-hop reasoning and retrieval tasks by first synthesizing compact question-answer pairs with evidence, then expanding that evidence into long multi-document contexts, before training with reinforcement learning.

trainingreasoningefficiency

Think Harder Where It Matters: Item-Level Adaptive Recursion for Sequential Recommendation

Oct 21, 2026

Zhanshuo Cao, Jiakai Tang, Yuqi Zhou et al.

Adaptive computation depth per item, learned through gating decisions at each recursion step, outperforms fixed-depth processing in sequential recommendation by matching computational effort to actual item complexity.

This paper identifies that sequential recommendation systems waste computation by treating all items equally. The authors propose LoopRec, which uses adaptive recursion to allocate deeper processing only to items that need it—like recent interactions or rare items—while letting simpler items exit early. This improves recommendation quality without requiring manual depth labels.

efficiencyarchitecturereasoning

PhysiCalWorld: In-Context Physics Calibration for World Models

Oct 21, 2026

Yunhao Luo, Wentao He, Yinpei Dai et al.

World models can learn to calibrate themselves to scene-specific physics by encoding exploratory interaction data into reusable descriptor tokens, improving prediction accuracy without requiring physics labels or fine-tuning.

PhysiCalWorld improves world models by learning to identify and adapt to the unique physics properties of a scene (like object mass or friction) through brief exploratory interactions. Instead of fine-tuning or using physics equations, it encodes these properties into a compact token that guides predictions, making visual planning more accurate in novel environments.

reasoningefficiency

Environment Duels: A Microcosm of LLM Co-Evolution

Oct 21, 2026

Shibo Jin, Zhiqiu Xu, Yiling Zou et al.

LLMs have specialized capabilities as task creators versus solvers, and measuring these distinct skills through head-to-head interactions reveals stable patterns that reflect how models influence each other in practice.

This paper introduces Environment Duels, a benchmark where LLMs both create interactive tasks (with helpful hints) and solve tasks created by others. Testing nine frontier models reveals that solving and authoring are distinct skills, smaller models write harder tasks, and the interactions mirror real-world LLM co-evolution patterns.

evaluationagentsreasoning

The End of Consensus Research?

Oct 21, 2026

Clément Burgelin, Rachid Guerraoui, Zicong Hong et al.

AI can accelerate discovery of practical distributed systems protocols by automating protocol design, implementation, and verification, but human experts remain essential for guiding the search and validating results.

Researchers used AI to discover A-greement, a new consensus protocol for distributed systems that outperforms existing solutions by 12-33% on geo-replicated workloads.

agentsreasoningarchitecture

Multimodal Embodied Agents: A Unified Capability-centric Perspective from Computer-Use to Robot-Use

Oct 21, 2026

Yanzhe Chen, Ziyi Yang, Jifeng Zhu et al.

Building reliable robots requires evaluating all five capabilities—especially Anticipate and Verify, which are largely invisible in current benchmarks—not just task success.

This paper proposes PAPAV, a unified framework for understanding multimodal embodied agents (systems that combine AI reasoning with physical action) by breaking them into five core capabilities: Perceive, Anticipate, Plan, Act, and Verify.

agentsreasoningevaluation

Did the Problem Actually Get Harder? Testing LLMs on Paired P and NP-Complete Problems Hardened by Adversarial Multi-Agent Rewriting

Oct 21, 2026

Nurul Labib Sayeedi, Md. Faiyaz Abdullah Sayeedi, Zabir Al Nazi et al.

LLMs fail to recognize computational complexity boundaries even when problems differ only structurally—they score 28 points lower on NP-complete problems, and this gap persists even with more computation time, indicating reasoning failures rather than speed issues.

This paper introduces P2NPCBench, a benchmark that tests whether LLMs can distinguish between polynomial-time solvable problems and NP-complete problems by pairing them with identical surface structure. Using an adversarial rewriting framework called AgentGAN, researchers created 2,114 hardened problem statements that remove textbook cues.

evaluationreasoning

GBClaw: An agentic AI framework for atomistic grain boundary research and mechanistic discovery

Oct 20, 2026

Yazhuo Liu, Kunqing Ding, Xiaoguo Gong et al.

Agentic AI can tackle domain-specific materials science by combining LLMs with specialized simulation tools and memory systems, enabling discovery of mechanistic insights that require both high-throughput screening and deep physical reasoning.

GBClaw is an AI framework that combines large language models with specialized tools to automatically investigate grain boundaries in materials. The system orchestrates complex simulations and analysis to understand how grain boundaries deform, discovering that defect structures significantly affect material strength—work that would be difficult for humans to do manually.

agentsapplicationsreasoning

Planning under Noise with Neural Operators

Oct 20, 2026

Michael Chen, Hong Chul Nam, Xi Deng et al.

Neural operators can efficiently learn noise-aware planning by amortizing expensive PDE solutions during training, enabling fast, reliable navigation in high-dimensional spaces with execution uncertainty.

This paper introduces PUN, a neural operator that learns to plan under control noise by predicting value fields for navigation in unseen environments.

reasoningefficiencyarchitecture

MindView: Benchmarking Generative Multi-View Spatial Intelligence

Oct 20, 2026

Yanbo Wang, Siqi Luo, Siyu Zhang et al.

Generative multi-view spatial understanding is a measurable capability that improves when models are fine-tuned on spatial generation tasks, and this improvement transfers to other spatial reasoning benchmarks without additional training.

MindView is a benchmark with 1,500 samples testing how well AI models can generate spatially consistent views from partial observations using language guidance. The benchmark includes seven tasks split between geometry-focused (camera position/orientation) and semantics-focused (object grounding, spatial relations) challenges.

evaluationmultimodalreasoning

Semantic Observability for Autonomous Systems Diagnostics

Oct 20, 2026

Jitendra Solanki, Anand Todkar, Ziran Min

By representing logs as a connected graph of infrastructure and events rather than flat text, autonomous agents can efficiently isolate failure causes and explain them in code-level detail—cutting through noise that overwhelms traditional log analysis.

This paper presents Semantic Observability, a multi-agent system that automatically diagnoses failures in distributed systems by converting unstructured logs into a queryable graph database, then using causal reasoning and LLMs to pinpoint root causes and suggest fixes. It reduces manual work for engineers investigating system outages.

agentsreasoningapplications

Value-Rich, Differently Organized: Measuring Value Expression and Organization in AI-Generated Debates

Oct 20, 2026

Maoyuan Li, Haoyuan Li, Zhongsheng Wang et al.

AI debate quality shouldn't be judged only on whether it mentions important values, but on how clearly and consistently it develops those values throughout the conversation—a dimension current evaluations miss.

This paper introduces a framework for evaluating how AI-generated debates express and organize values compared to human debates. Using Schwartz's value theory, the researchers measure not just which values appear in debates, but how they're distributed and sustained across turns.

evaluationreasoning

What Do RTL Repair Benchmarks Actually Measure?

Oct 20, 2026

Yeyin Eva Zhu, Rishabh Ranawat

RTL repair benchmarks need careful evaluation design—simulation and formal methods can disagree, and you must verify models actually use the specification, not just get lucky with fixes.

This paper creates RTLRepairBench, a dataset and evaluation framework for measuring whether AI models can fix broken hardware designs (RTL code). The authors identify critical measurement problems: simulation and formal verification give contradictory results, and models might appear to fix code without actually using the specification.

evaluationreasoning

OpenVLHarness: Real World Visual Agentic Tool-Use with Multimodal Memory

Oct 19, 2026

Yu Zhou, Cheng-Fu Yang, Rui Sun et al.

Adding tool access and persistent multimodal memory to vision-language models dramatically improves performance on real-world tasks that require grounded perception and numerical reasoning, while actually reducing computational cost compared to scaling inference alone.

OpenVLHarness is a system that helps multimodal AI models solve complex visual reasoning tasks by combining external tools, persistent memory, and code execution.

agentsmultimodalreasoning

Can It Actually Work? Synthesizing Functional CAD Mechanisms

Oct 19, 2026

Wenqian Zhang, Yangyi Huang, Zigeng Xu et al.

LLM agents can generate visually plausible CAD designs, but most fail functionally—this work shows how to verify and iteratively fix mechanisms using physics simulation rather than just visual checks.

This paper tackles functional CAD design by proposing FunCAD-Harness, a framework that helps AI agents create working mechanical designs by translating abstract requirements into physical tests and using simulation feedback to fix failures.

agentsevaluationreasoning

Reinforcement Learning with Visually Attributable On-Policy Self-Distillation

Oct 19, 2026

Shiao Xie, Siyu Chen, Jianwei Lv et al.

When training vision-language models with self-distillation, you need to verify that token-level corrections actually come from visual reasoning, not linguistic artifacts—VAG does this by checking if corrections persist when visual evidence is degraded.

This paper addresses a key problem in multimodal reinforcement learning: when a teacher model provides token-level corrections to improve visual reasoning, these corrections may be contaminated by non-visual factors like language patterns or the teacher's perspective.

trainingmultimodalreasoning

CoopUQ: Robust Multi-Robot Social Navigation via Uncertainty-Aware Cooperative Forecasting

Oct 19, 2026

Zehao Wang, Jianpeng Yao, Yuping Wang et al.

Multi-robot navigation becomes safer and more robust by having robots cooperatively predict future occupancy with uncertainty estimates, then using those predictions to constrain navigation decisions rather than relying on individual robot perception alone.

CoopUQ enables multiple robots to navigate safely around humans by combining uncertainty-aware predictions of future occupancy with cooperative decision-making.

agentssafetyreasoning

Where is Physical RSI capability from? An Observational Scaling Law of LLM-as-policy

Oct 19, 2026

Zekai Li, Jiayi Yu, Jinlu Zhang et al.

LLM-based robot policies improve through physical experience in predictable ways; you can forecast how far and how fast a robot will improve by measuring early performance across five dimensions: first-attempt success, efficiency, overall success rate, interaction budget, and learning from failures.

This paper investigates how large language models used as robotic policies improve through physical interaction. The authors introduce PhysRSI-eval, a five-metric framework measuring robot learning across 18 manipulation tasks, and discover that different capabilities limit improvement in different ways—like a barrel constrained by its weakest stave.

agentsreasoningevaluation

Train An Imaginative Agent

Oct 19, 2026

Jiaxin Ge, Junyi Zhang, Tianning Chai et al.

Agents can learn to imagine and reason about action consequences before acting by combining world modeling with planning in a single model, dramatically improving performance in complex environments with irreversible actions.

This paper introduces Agentic Imagination (Agima), a training framework that teaches agents to simulate action consequences before executing them.

reasoningagentstraining

AnyGoal: On World Model Generalization Beyond Expert Rollouts

Oct 19, 2026

Heejeong Nam, Divyat Mahajan, Quentin Le Lidec et al.

World models trained on expert trajectories struggle to generalize to new combinations of familiar states and actions—a critical limitation for real-world planning that requires handling novel situations built from known components.

This paper introduces AnyGoal, a benchmark for testing whether world models can generalize beyond their training data by recombining familiar state and action factors in new ways.

evaluationreasoningdata

Same Simulation, Different Evidence: Where Retrieved Tool Output Costs LLM Agents Accuracy

Oct 19, 2026

Ke Zhang, Sahchit Chundur, Jerry Peprah Owusu et al.

When building LLM agents with large tool outputs, retrieval-based access can cause accuracy loss not because agents fail to compute correctly, but because they ignore retrieved evidence and use their own outputs instead—a failure mode invisible in final accuracy metrics.

This paper investigates how LLM agents lose accuracy when tool outputs are retrieved from files rather than kept in context. Using PHREEQC, a geochemistry simulator, the authors show that GPT models suffer 7.9-11.3 point accuracy drops under retrieval, primarily because they substitute their own outputs for the authoritative tool results. Claude models show no detectable loss.

agentsevaluationreasoning

Kronecker Delta Rule for Long Context Linear Attention

Oct 19, 2026

Anamaria-Roberta Hartl, Levente Zólyomi, Michael List et al.

The Kronecker Delta Rule offers a theoretically grounded way to update associative memories in attention models that's both stable and efficient for long-context tasks, outperforming existing linear attention methods.

KronLSTM introduces a new memory update mechanism for linear attention models that modifies memories along a single direction, reducing interference with stored information. This approach provides mathematically stable memory dynamics without extra normalization, making it effective for processing long documents and sequences.

architectureefficiencyreasoning
architecturereasoningefficiency

Beyond the Answer: Student-Conditioned Trace Construction for Reasoning Distillation

Oct 16, 2026

Shaoang Li, Yanhang Shi, Yufei Li et al.

Reasoning traces are more effective for training when customized to the student model's capabilities and selected based on how well they help recover answers, not just whether they're correct.

This paper presents ReDiST, a method for creating better reasoning traces to teach AI models. Instead of just showing correct answers, it constructs traces tailored to what a specific student model can learn—revising correct attempts and fixing wrong ones using evidence.

trainingreasoningevaluation

Self-Building Labs

Oct 16, 2026

Jiaru Bai, Yuchi Zhao, Han Hao et al.

Instead of manually rebuilding labs for new experiments, AI can now autonomously design lab infrastructure and workflows from high-level scientific goals, making experimental automation dramatically more flexible and scalable.

Self-building laboratories (SBLs) are AI systems that automatically design and configure their own lab setups to achieve scientific goals, starting from scratch or existing equipment.

agentsapplicationsreasoning

ENFR: Evidence Networks with Flow-Matched References for Amortized Simulation-Based Model Inference

Oct 16, 2026

Kai Lehman, Jameson Dong, Carolina Cuesta-Lazaro et al.

ENFR enables fast, amortized estimation of absolute model evidence by learning Bayes factors against a flow-matched reference—solving a key bottleneck in model validation for simulation-based inference without requiring likelihood evaluations.

This paper presents ENFR, a method for computing the evidence (marginal likelihood) of statistical models, which is crucial for model comparison and goodness-of-fit testing.

evaluationreasoningefficiency

Learning without relearning physical symmetry: an SO(3)-equivariant convolutional network (CNN) on the sphere

Oct 16, 2026

Joshua Williamson, Niall Jeffrey, Lorne Whiteway

Encoding physical symmetries directly into neural network architecture—rather than letting the model learn them—dramatically improves data efficiency and accuracy for scientific problems with inherent geometric structure.

This paper presents a spherical convolutional neural network designed for data naturally living on spheres (like cosmic microwave background maps). By building in rotational symmetry mathematically rather than learning it, the network uses training data more efficiently and achieves better results on cosmology tasks with fewer simulations needed.

architectureefficiencyreasoning

CocoRNA: Collective RNA Design with Cooperative Multi-agent Reinforcement Learning

Oct 16, 2026

Tianmeng Hu, Biao Luo, Ke Li

Decomposing sequence design into cooperative multi-agent sub-problems with centralized guidance improves both the quality and speed of RNA structure prediction compared to single-agent approaches.

CocoRNA uses multiple AI agents working together to design RNA sequences that fold into specific shapes. By breaking the complex design problem into smaller parts and having agents collaborate with guidance from a central critic, the system learns to create better RNA designs faster than existing methods.

trainingagentsreasoning

Learning Personalized Memory Gating from Delayed One-Sided Feedback

Oct 16, 2026

Kaixuan Ji, Zixiang Chen, Wenting Zhao et al.

Personalized memory gating—deciding which facts to remember for each user—can be learned efficiently by treating it as a bandit problem with delayed feedback, outperforming one-size-fits-all memory policies.

This paper tackles personalized memory management for AI agents by learning which facts to keep in a limited memory bank. The authors propose a method that learns user-specific patterns of what information gets reused later, combining offline learning from past data with online adaptation to individual users.

trainingagentsreasoning

Discovering state-of-the-art quantum circuits with large language models

Oct 15, 2026

Andi Gu, Alek H. Kemeny, Susanne F. Yelin et al.

LLMs can effectively co-design quantum algorithms by proposing circuit structures that an evaluator scores, discovering practical improvements over hand-optimized quantum code without requiring the LLM to understand quantum mechanics deeply.

Researchers used large language models working with an independent evaluator to discover optimized quantum circuits for various computational tasks.

reasoningarchitectureapplications

PathFM: Fast Generative Forecasting of Multivariate Trajectories

Oct 15, 2026

Jiachen Yao, Xiyuan Zhang, Annan Yu et al.

Generative trajectory forecasting is now practical: PathFM generates full multivariate paths in near-quantile-model speed, enabling better predictions of trajectory-dependent quantities like cumulative demand and peak timing.

PathFM is a time series forecasting model that generates complete future trajectories rather than just point predictions. Using flow matching, it can produce multiple realistic paths for a single series almost as fast as traditional quantile models, while better capturing dependencies between variables that matter for real-world applications like peak load prediction.

reasoningevaluationefficiency

OmniVerifier-C1: Decomposed Graded Verifier for Rule-Governed Generative World Modeling

Oct 15, 2026

Bowei Liu, Xinchen Zhang, Xuhuan Li et al.

Graded, component-level verification with pairwise comparison significantly outperforms binary verification for evaluating complex multimodal outputs, especially in structured reasoning tasks where compositional correctness matters.

This paper introduces OmniVerifier-C1, a visual verifier that moves beyond simple True/False judgments to provide detailed, component-level scores for evaluating generated images.

evaluationreasoningmultimodal

MLExpert: Towards a Generalist Agentic Model for Autonomous Machine Learning Engineering

Oct 15, 2026

Shangheng Du, Jinxin Shi, Shuaiyu Zhang et al.

Training AI agents on long experimental workflows requires learning from intermediate outcomes, not just final results—this paper shows how to extract useful learning signals from expensive, multi-step ML engineering tasks.

MLExpert is a 35B AI model trained to autonomously manage machine learning experiments and optimization. Using a novel training approach combining supervised learning from expert trajectories and reinforcement learning from experiment outcomes, it learns to make better decisions across long sequences of costly ML tasks, improving performance from 38% to 73% on benchmarks.

agentstrainingreasoning

GeoState: Learning View-Decoupled Geometry States for 3D World Modeling

Oct 15, 2026

Hongchang Liu, Jian Liu, Yujuan Ding et al.

By decoupling scene state from camera viewpoint, GeoState makes 3D world predictions more accurate and generalizable—the same predicted scene can be rendered from different camera angles without retraining.

GeoState is a 3D world modeling approach that separates scene geometry from camera viewpoint by learning view-decoupled state representations. It uses a scene tokenizer to aggregate multi-view geometry features into a unified state, then predicts future states with an autoregressive model while handling camera motion separately.

architecturereasoningmultimodal