4 ms·
i actually do it differently > (1) Let the LLM randomly perturbate the system. instead of this i ask LLM to what's least likely to improve performance and th
by faangguyindia 6mo ago
i actually do it differently
> (1) Let the LLM randomly perturbate the system.
instead of this i ask LLM to what's least likely to improve performance and then measure it.
sometimes big gains come from places you thought are least likely.
- fesens 6mo agoFor sure! The hypothesis generation gotta be improved. Your take on the "least likely" is interesting. In the beginning of the repo I was having problems with "hypothesis convergence", your idea may be a nice way to introduce the much needed variability