Glossary term
Glossary term
Foundations
A form of prompt chaining that divides complex problems into an ordered set of simpler problems. For example, here's a least-to-most prompting strategy for a certain problem:
Divide a complex problem into an ordered list of simpler sub-problems. For this example, assume it is three sub-problems.
Prompt 1: Ask the LLM to solve the first sub-problem. The LLM returns Response 1.
Prompt 2: Integrate all or part of Response 1 into the prompt to solve the second sub-problem. The LLM returns Response 2.
Prompt 3: Integrate all or part of Response 2 into the prompt to solve the third sub-problem. The LLM's response to Prompt 3 is the "final" answer to the initial complex problem.
Note that each step depends on the solution to the preceding step.
Contrast with tree-of-thought prompting.
Created for this library
A math tutoring tool uses least-to-most prompting to decompose a word problem into simpler subproblems before solving the original.
A coding assistant uses least-to-most prompting to ask the model to first plan, then implement, when refactoring complex code.
A research assistant uses least-to-most prompting to break a long research question into a sequence of grounded sub-queries.
Definition source: Google for Developers Machine Learning Glossary | Creative Commons Attribution 4.0 License