colindoyle0000/llms-as-method-actors

Implements role-based prompting strategies where LLMs adopt specific personas or "method actor" personas to improve reasoning on constraint-satisfaction tasks, with support for OpenAI, Anthropic, and Gemini APIs. The approach uses multi-stage prompting techniques (vanilla, chain-of-thought, and actor-based variants) with intermediate reasoning tracking and comparative evaluation. Includes Jupyter tutorials and documented sample outputs demonstrating performance on word-puzzle tasks, showing the actor-based approach achieving zero incorrect guesses versus baseline methods requiring multiple correction attempts.

No commits in the last 6 months.