Chain of Thought

This is how LLMs train to answer queries using structured token sequences called "thoughts".

These thoughts are used to break a problem into smaller problems and find a path to a given question's solution.