Home About

Tags

MachineLearning (26) LinearAlgebra (16) GameDesign (12) ComputerScience (11) SoftwareEngineering (11) LargeLanguageModels (9) AudioEngineering (7) DiscreteMath (6) AutomatedTesting (6) Roblox (5) Zettelkasten (5) AgenticReasoning (4) More

Notes by Lex Toumbourou

Self-Refine

Oct 05, 2024 permanent

Self-Refine is an Reflection technique, that allows an agent to iteratively refine outputs and incorporate feedback to improve performance on tasks.

It does not require fine-tuning and is an example of a Single Agent System.

Self-refine example

Image from https://selfrefine.info

See Self-Refine: Iterative Refinement with Self-Feedback.