Paper summary
Paper summary: START: Self-taught Reasoner with Tools
START teaches a long chain-of-thought reasoning model to call a Python interpreter mid-reasoning, using inserted hints ("Wait, maybe using Python here is a good idea") and rejection-sampling fine-tuning.
- A simple inserted hint is enough to get a reasoning model to start using code execution, with no demonstration data.
- The tool lets the model compute, self-check and debug instead of relying only on its internal reasoning.