[code_review] Investigate automatic prompt engineering to improve prompts

[marco-c]

https://www.deeplearning.ai/the-batch/research-summary-automatic-prompt-engineer-ape/

Also related and interesting idea: https://www.reddit.com/r/AutoGPT/comments/12t7bud/comment/jh4oelk/ (given expected input-outputs, improve prompt with evolutionary algorithms to get as many as possible right outputs from inputs).

We can use the field data to improve the generation part or the filtering part.

[marco-c]

There are now tools such as https://github.com/hinthornw/promptimizer that we could use for this.

As optimization objective, we could use:

• field data (maximizing accepted and minimizing rejected);
• human comments (but we'd need to curate a dataset of valuable comments);
• synthetic comments from past bugs (see also [code_review] Add generated synthetic examples from past bugs for RAG #4586).

[marco-c]

https://github.com/codelion/optillm is also pretty interesting.

[marco-c]

And https://github.com/zou-group/textgrad.