Scaling Competence, Shrinking Reasoning: Cognitive Signatures in Language Model Learning
This provides insights for diagnosing training stages and optimizing reasoning model training, though it is incremental in applying cognitive science concepts to AI.
The paper analyzes how reasoning tokens in language models evolve during fine-tuning, showing that reasoning length first increases as performance improves, peaks at a stage of conscious competence, and then declines as the model internalizes the task, with models retaining performance even after reasoning is removed.
We analyze reasoning in language models during task-specific fine-tuning and draws parallel between reasoning tokens--intermediate steps generated while solving problem and the human working memory. Drawing from cognitive science, we align training dynamics with the Four Stages of Competence: models initially produce incorrect outputs without reasoning, then begin reasoning (but still fail), eventually reason effectively, and finally solve tasks without explicit reasoning. We find that reasoning token length expands as performance improves, peaks at the stage of conscious competence, then declines as the model internalizes the task. Notably, after training, models retain performance even when reasoning is removed--suggesting it scaffolded learning but is no longer needed. This progression offers actionable insights: reasoning token dynamics can serve as a signal for diagnosing training stage, identifying convergence, and guiding early stopping. We propose metrics to track this trajectory and argue that reasoning behavior is valuable for understanding and optimizing reasoning model training.