GPT-4 and ChatGPT study shows LLMs are getting dumber
I'm inclined to agree in general. GPT has actually gotten worse over time.
For one it's just way too locked down now. But I also have noticed a decline in overall quality of responses, especially for technical stuff. It used to translate between programming languages quite well, and that's been more broken lately.
I'm wondering if they purged a bunch of copyrighted material from it and that's why it got dumber.
https://www.theregister.com/2023/07/20/gpt4_chatgpt_performance/Open linkView original on partizle.com
That would make sense for stuff where you're doing weird writing prompts, but I don't see how it relates to get getting worse at programming tasks.