3 ms·
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 th
by weinzierl 4y ago
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.
- neodypsis 4y agoI wonder if someone has tried to chain ChatGPT's output with static analyzers/compilers/linters in a loop to improve the final response.
- blatant303 4y agohttps://github.com/newhouseb/clownfish https://github.com/newhouseb/clownfish Structural Alignment: Modifying Transformers (like GPT) to Follow a JSON Schema