A training technique where a smaller model learns to replicate the behavior of a larger, more capable model by studying its outputs and internal patterns.
Multi-step reasoning, logic puzzles, mathematical problem-solving