protologue

ProTeGi

Also called Prompt Optimization with Textual Gradients, textual gradients.

ProTeGi (Prompt Optimization with Textual Gradients) improves a prompt by having a language model describe, in natural language, what is wrong with it based on the errors it produces, then edit the prompt to fix those problems, keeping the best candidates with beam search.

Description

The natural-language critique plays the role of a gradient, and editing the prompt plays the role of a step in the opposite direction. A bandit-style selection procedure limits how many evaluations are needed to choose among candidates.

Sources

  1. Pryzant et al. (2023). Automatic Prompt Optimization with "Gradient Descent" and Beam Search.

Cite this entry

Protologue. (2026). ProTeGi. In Protologue: A Taxonomy of Prompting and LLM Techniques (v1.4.0, PTL-0151). https://protologue.com/t/protegi/

BibTeX
@misc{protologue_protegi,
  title = {ProTeGi},
  author = {{Protologue}},
  year = {2026},
  howpublished = {Protologue: A Taxonomy of Prompting and LLM Techniques, v1.4.0},
  note = {Entry PTL-0151},
  url = {https://protologue.com/t/protegi/}
}

Markdown JSON