I think mine when when I got one model to generate a spec for SDL 1.2, and another to generate a reimplementation of the whole thing in rust. It was pretty amazing to see the implementation getting fleshed out and fixed as I fed back results from a real project that still compiled with SDL 1.2. Kept pushing it. Added fully functional audio in 2 prompts. Applied optimizations with another.
Terrible code, but to think you can just rewrite whole libs with AI with relatively lazy prompting + feedback and get decent enough results is kind of mind boggling for me.
Terrible code, but to think you can just rewrite whole libs with AI with relatively lazy prompting + feedback and get decent enough results is kind of mind boggling for me.