ConceptTraining-free self-reflection protocol
Training-free Self-Reflection Protocol
A method that does not update model parameters and instead enables the model to self-check and correct at inference time through prompt or process design. A typical pipeline is a generate-critique-revise loop, which relies on the instruction-following and self-evaluation capabilities of large language models.
Timeline (last 90 days)
Oct 20
Training-free self-reflection protocols refer to methods that do not update model parameters and instead let the model self-check and revise at inference time through prompting or workflow design, typically represented by the "generate→critique→revise" loop (such as Self-Refine).
Unverified50%