Hacker Newsnew | past | comments | ask | show | jobs | submitlogin

There is also a huge difference between code generation (be it Copilot or ChatGPT). With Code you still have the compiler that rejects the utter nonsense and then you hopefully have tests to check if what was generated makes sense.

With prose it's all your manual work again to check. If your text is mundane it might be trivial but for anything a pg would write I guess an AI is pretty useless in its current state.



I wonder if someone has tried to chain ChatGPT's output with static analyzers/compilers/linters in a loop to improve the final response.


https://github.com/newhouseb/clownfish

Structural Alignment: Modifying Transformers (like GPT) to Follow a JSON Schema




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: