Changes to ChatGPT (GPT4)

[lusty void]

Is it me, or has ChatGPT changed over the last day? It no longer recalls information posted several times in the last conversation, I have to keep sending the same data. Also the level of detail in answers has completely diminished. Lastly the answers seem much more political, in that it just ignores aspects of the question, or when asked a direct question, you get the AI equivalent of "I'm sorry you feel that way...." I am neck deep into a very sensitive and detailed piece of work with CG and cant continue with CG in its current state.

[true isle]

It is capable of forgetting about stuff from a previous message or 2 even if under 3K

I am curious about the context of what you have been talking about when you say he is getting political (politically correct?)

[marble pecan]

hallo

[marble pecan]

w

[lusty void]

hi @true isle it is strange because I have been in chats for many hours over 12 hours straight, in fact over many days, in the same conversation and CG4 did not forget a thing. Then two days ago it started forgetting things (algorithms I had messaged, acronyms defined, lists sent in the previous 2 and 3 messages). It started interjecting its own meanings to acronyms defined. I then asked specific question, do you recall my XXX algorithm, where did you get the acronym xxx from, why did you not respond to my previous question.....it continued with responses that I call political (in that CG4 did not actually answer it) like, and I paraphrase, Apologies for any confusion caused, i am an AI model and your question is beyond my capability.

@marble pecan Hi

[silver niche]

Experienced similar degradation of ability with GPT4 in the past week or so.
Prior it was perfectly capable of even generating a full software solution (with some serious prompting), but now not only is the message length many times shorter than before, the quality is abysmal with 3.5 being on par with 4 while 3.5 has experienced only minor degradation of ability.

[lusty void]

Exactly. It is very disappointing. Let me know if you find any alternatives or ways to improve the quality of responses.

[true isle]

It is something that if you are able to get 😅 a pretty good conversation with your cg4 and don't want to stop because it is going better than usual it will happen at one point that it would be better to make a new conversation especially if you are going closer to the 100 message mark...

If you have a version than can browse you can tell him to follow those instructions gist.githubusercontent-dot-com/Luxcium/a1c206b4b05c368e4298404f2a481161/raw/c4f714f761df138da58712c5f0e42585f3de7151/ChatGPT-export_by-luxcium.txt to get a message to copy paste in a next session and it will start mostly in the same contexte and state more or less...

[lusty void]

thanks🙏 , yes I have a version that can browse. Let me try this. Have you noticed any degradation in the quality of CG4 responses?

[slim glacier]

agree. It seems like gpt4 got worse than 3,5.

[silver niche]

What confuses me most is the radically reduced input length. I can use literally 2x longer input for 3.5 vs 4 and part of 4's claim to fame was specifically the longer input. Given 4 is effectively behind a paywall then this kind of behaviour is bordering on false advertisement/fraud.
Sure, it's a "research product" with heavy quotes, but morally there is absolutely no reason for why they shouldn't be communicating these things to their users and community.