| ▲ | faangguyindia 3 hours 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. | ||