Hacker News Reader

You would think so, but not a single person complaining has provided convincing proof of the degradation.

I went through a bunch of old prompts and the response was definitely worse now than it was then.

Since it's not determinstic though, it's hard to draw conclusions. Especially since the sample size was 10ish.

GPT isn’t deterministic

You need to have used the API and set temperature zero, but if you have that historical data you can reasonably test and see.