Chain of Thoughts
CoT allows models to leverage asymmetry of verification.
A class of problems has “asymmetry of verification”, which means it’s easier to verify a solution than to generate one.
For example, a crossword puzzle, sudoku, or writing a poem that fits constraints.
The Sudoku Solver with Human Reasoning project explores explicit reasoning steps in one of these settings. This distinction between producing an answer and checking it also relates to my questions about evaluating AI agents.