undefined | Better HN

0 pointsjacquesm2mo ago0 comments

I'm not sure if that will hold up. The LLM is not going to do anything random and that is actually a powerful component that makes original output possible.

0 comments

kyralis2mo ago

I wonder if a combination would be useful. Use an actual GA to do the mutation, and then let an LLM "fix" each mutated child.

jacquesmOP2mo ago

Could be. But the interesting thing is that all you can do here is optimize. Random chance is - like attention ;) - all you need.

j / k navigate · click thread line to collapse

0 pointsjacquesm2mo ago0 comments

I'm not sure if that will hold up. The LLM is not going to do anything random and that is actually a powerful component that makes original output possible.

0 comments

kyralis2mo ago

I wonder if a combination would be useful. Use an actual GA to do the mutation, and then let an LLM "fix" each mutated child.

jacquesmOP2mo ago

Could be. But the interesting thing is that all you can do here is optimize. Random chance is - like attention ;) - all you need.

j / k navigate · click thread line to collapse