Chain of Thought (CoT)

Chain of Thought (CoT) prompting significantly improves an LLM's performance on complex logic, math, and reasoning tasks. By explicitly instructing the model to "think step by step" or allocating a `<scratchpad>` for it to write out its logic, CoT allows the autoregressive model to allocate more compute (tokens) to the problem before arriving at a conclusion, effectively simulating a deeper cognitive process.