GPT-4 has undergone a severe downgrade

This forum is not for ChatGPT support, if you need help with your account, payment, etc, please visit

Please DO NOT SHARE personal information (email, credit card, etc) or account related details publicly.

People always say that GPT hasn’t been downgraded, but when you encounter such a situation, you tell GPT-4 what needs to be done, and it merely provides you with… but doesn’t help you address the problem. This is nonsense! It can’t solve my problem at all.

I want him to follow certain rules, and I have informed him about the rules, so that he can help me accomplish a certain task based on those rules. But now he is simply repeating the rules and asking me to complete the task according to the rules. This is utter nonsense!

There’s not nearly enough information presented here for anyone to give you any useful feedback.

Do you have a specific example of prompt which is failing?

My requirement was for him to analyze the purpose of this code snippet and add comments to my code based on the rules I specified. However, he only executed a portion of it, and even in that portion, he omitted my original code.

Clearly, this code is not complex but rather lengthy. I was hoping he could save me time and effort, but he didn’t.

Again, not enough information given to be able to figure out what your issue is.

Both of those links 404 for me, so I’m still not able to see what you’re issue is.

Why do you do this? Go from post to post belittling the problem?

Dont mind this guy. He goes from post to post asking for “proof” as if he was able to do something with it.
There are lots of people experiencing your pain.

I’m not sure what you mean. I’m trying to help identify and isolate the problem so I might be able to help other users achieve the results they are after.

In all of this, of all the people I’ve interacted with, I have yet to ever be shown a concrete and verifiable example of decreased performance.

Hi @caoyujie1213

Just wanted to say generally thank you to everyone reporting their experience with GPT-4 model performance. :heart:
Everyone @OpenAI wants the best models that help people do more of what they are excited about. We are actively looking into the reports people shared.

If you have time and a use case you care about, it would be awesome to see more Evals being made: It really does help us a lot to have specific metrics to fall back on. Qualitative feedback is helpful but often hard to take action on.
Logan’s tweet

Just give it a little more time, I think you’ll be served. You seem to think that there is only smoke when the fire has been well lit for several weeks now.

There is no need for people to show evidence and explain something that is undeniably very well known to openai themselves. To claim otherwise is totally naive.

As this topic is not going anywhere productive closing it.

1 Like