Large Language Models as Optimizers

• https://arxiv.org/abs/2309.03409
• Chengrun Yang, Xuezhi Wang, Yifeng Lu, Hanxiao Liu, Quoc V. Le, Denny Zhou, Xinyun Chen

Prompt tuning, Regression

LLMs can be used as an optimizer. For instance, for a linear regression, we can prompt an LLM with (parameters, error) tuples and then ask it to find parameters that lower the error. The results from this can then be fed again with the same metaprompt.

This procedure can be applied to the prompts. Given a set of (prompt, score), ask LLMs to find a better prompt to improve the score.