Lately, I've noticed issues with ChatGPT generating code that is nearly unusable, often leaving massive gaps in the code and prompting me to fill them up. Have you experienced similar outputs?
Looks like it! Even code quality has dropped 'significantly.' I used it once in a while to clean up the code after writing a POC. It used to do decent work before, but now I almost always can't use the code generated at first shot.
This is never reflected in the benchmarks and everyone uses them as imperial universal truths. ChatGPT once reached a certain quality but is scaling down on first requests to save compute power.
