LLMs are good at optimizing towards local goals. Getting types right at compile time is a local goal. Entry and exit assertions are local goals. Unit tests are local goals. So those constructs all help AI-generated code.
Matching a desired output is a global goal, but even that sometimes works now.
Someone sent me a LLM-generated JPEG 2000 decoder. They got Fable to generate a decoder that uses a GPU to get the same answer as the reference implementation gets on the GPU.
I’ve reimplemented some python science code into rust using LLMs and it absolutely works well if you setup the criteria for done. In my case it was GIS-ish code so a key requirement was identical in/out pairs at 10k points across CONUS. It worked, and when it didn’t it showed up bugs that the agent would go and research, read up on other implementations, then try again.
Net result faster (10-70x) and massively lower memory footprint. Now we’ve got a service that can do something essentially instantly that most people wouldn’t have even bothered to try before.
Animats · · focus · HN ↗
Matching a desired output is a global goal, but even that sometimes works now. Someone sent me a LLM-generated JPEG 2000 decoder. They got Fable to generate a decoder that uses a GPU to get the same answer as the reference implementation gets on the GPU.
te_chris · · focus · HN ↗
Net result faster (10-70x) and massively lower memory footprint. Now we’ve got a service that can do something essentially instantly that most people wouldn’t have even bothered to try before.