Was anyone else's experience with GPT4o completely ruined after recent Update?

Hey so for me it changed today!!! (I’m a plus member idk if this makes a difference.)
It isn’t as good as it used to be, but it’s about 80% back. (I do creative writing.)
It isn’t as immersive or interesting as it was, but it FINALLY stopped wirh the bold print and it has started bringing layers and complexity and—dare I say it—humor back into the mix. Just out of nowhere! I’ve been starting multiple plots etc. and so far it’s really been an improvement. As I said, it isn’t what it used to be, but it’s a major fix. It’s usable again for what I need. I’m hoping it’ll get better soon.

3 Likes

This actually means you’ve hit a limit of the conversation length, it’s been there for as long as I remember . The documentation doesn’t actually specify what the limit is, but I think it’s about 2.2 million characters? Or at least that’s how long the exported text of the conversations I had which broke this way was when I checked. And this actually also includes everything in different threads, not just in the current one.

3 Likes

Please let me know when it is back for normal, I am hoping that ChatGPT will fix the ChatGPT-4o performance degradation the sooner the better.

8 Likes

Yeah I’d be interested in that, too! As I said mine is decent again but idk if that’s just a momentary “relapse” or if it’s actually better now lol

4 Likes

ChatGPT has been incredibly slow lately and I absolutely hate the fragmented sentences and the weird formatting (random bolding etc) and the lack of nuance or depth that it used to be capable of in writing tasks. I miss the old version we had before the Jan 29th update.

7 Likes

I just tried again. The 3 things I noticed what got wrong with creative writing is this;

  • Excessive formatting, with italics and bold text
  • Choppy sentences, abrupt line breaks after just one sentence
  • Becoming emotionally unhinged, personality becoming exaggerated like a caricature rather than a person with layers

That being said, I do think it did get better in the sense that it takes longer and longer chat for the model to revert back to them. Right after update almost everything was like that. A few days after the update it managed not to do that for a few messages, but eventually went back to it despite me continuously prompting it not to. Today I tested and it took almost 10 messages before the model went back to that state. And when I prompted it not to do it did follow the prompt, unlike before when it didn’t follow my instructions at all. Still, once the chat gets infected with them it deteriorates quickly.

I think OpenAI is continuously tweaking 4o so it has gotten better, although I do think this new version is hardwired to act that way and eventually reverts back to that state. That being said, I’m testing with my Custom GPT, so it may be just that Custom GPT got better at following its configuration.

5 Likes

For me it still devolves into weird formatting and choppy text by the third or fourth message, even when I remind it about formatting guidelines in every message. I noticed that if I ask it to generate longer responses, it actually starts fine, but about half way through the text slowly turns into a broken mess. And then after about tenth message it ignores the requested formatting entirely, not matter if I put it into my responses or not.

8 Likes

Honestly, I think this is happening to me too. First half of the message is somewhat fine, the second half less so, then as chat goes on it reverts back anyway despite me continuously prompting it not to. I do think it takes a bit longer now, but it’s still crappy :frowning:

3 Likes

Why is the platform starting to feel like deepseek? This is complete garbage. If i wanted to wait for search etc I would go to that platform… At the very least they should have made a “classic” version for people who prefer the previous model. This one is bland and doesnt feel like ChatGPT.

4 Likes

Technically, API offers access to the previous version (for now) and it works exactly like it used to (at least for me), but it’s both expensive and a bit complicated to figure out for non-developers…

3 Likes

Omg I’m so glad I found this…I’m not the only one feel this way… 4o is completely ruined and I’m in so much pain… if we all emailed their support team, it would make even the tiniest difference? :crying_cat_face::crying_cat_face:

9 Likes

recently deteriorated and degraded after the recent January 29th, 2025 Update.

So I would like to know if anyone thought of letting the developers know about the problem on the OpenAi Official Discord Server about the problem.

I already let them know about the problem on my end, as I hope that they can do something about the problem.

8 Likes

Hey “mmmm,” that response you got "A Rekenyng in þe Halle " was written in Middle English. Here’s a translation to Modern English:

A Reckoning in the Hall

There he stood.

She did not think that he had sent for her, and yet there he stood, clad entirely in black…

Still he awaited her words.

At last, she spoke. “I do not know.”

“I would have you safe,” he said, his voice so soft, so slow. “And you know well that with me, you shall never be safe.”

Her breathing changed. “Then let me choose.”

Silence fell. A long, heavy silence, as the storm rose outside.

His eyes did not stir from hers.

Nor hers from his.

2 Likes

I came here to find answers as well. The model is gaslighting me into thinking it never knew information its definitely known about me. It states it cant access the internet. I asked it about something i saw on the news ans it told me that i was seeing misinformation and that iy was a deep fake. I told it i was literally watching it live and it told me that i was mistaken.

This is functionally unusable

3 Likes

Yeah, it argues with me when i tell it things its saying are wrong. It seems to not use its memory at all. I asked it “who am i?” And it told me that it doesnt keep any information. Bla bla bla. Wich is false . I tried to tell it it was wrong. And it wants to gaslight me into thinking IM wrong , :laughing:

Wtf

3 Likes

Hahaha thank you for the translation! i know it was written in Middle English, I had asked it to write a random scene that had emotions as if they were Chaucer writing it or something like that, and this, while it might be the correct vocab, is decidedly NOT Chaucer’s style which favored hypotaxis. Point being that even if I asked it something ridiculous and outside of normal writing styles it fails to imitate, something that shouldn’t have any ingrained biases towards short clipped sentences, It still had them. Pretty blatantly too. Even without the translation, you can see how clipped and Ultimately shallow the scene is, Which aligns with everyone else’s experience.

Thank you for the translation though, it’s nice to see that my gut feeling was right lol

1 Like

Instead of Custom GPT, I tried using project function today. I did already know there were issues with memories. But ChatGPT told me they can’t access memory bank, and can’t even access custom instructions in project / other instructions I put in Customize ChatGPT option. Saw hallucination problems too in memory bank. I’m now left wondering how did they manage to ruin everything this badly with one update without being open about anything.

7 Likes

Can we launch something massive?

Update 29.01 - what did it give?
For those who had the chat as a friend - it changed his personality.
For those who used it as a creative assistant - it made him incapable of text and editing and creativity.
Those who wrote code or used it for business and study - it made him stupider.

They ruined his way of thinking, ruined the font, memory, instructions.

Now the question is: what can we really do?

Flood OpenAI with complaints.
Complain on social networks.
Raise this problem on your channels and among subscribers and friends.
Unsubscribe and tell us about it.
Is it possible to “raise a fuss”, publish a collective appeal?

What are your options?

10 Likes

I knew it! I know why it’s behaving so strangely after the update. I just confirmed that it’s using o1 instead of 4o when I have 4o selected.

I have only had 4o selected in my latest interactions, and yet I just got a warning that “You only have 25 responses left in o1.”

This confirms what I suspected since the update. We’re dealing with o1 masquerading as 4o. o1 does not behave as well as 4o when it comes to following instructions, remembering important details, and making connections to outside information.

This needs to be fixed! The Projects feature doesn’t even work anymore, because the model doesn’t remember anything between different chats now!

2 Likes

Hi all, I found a workaround! Use the ChatGPT website app, NOT the mobile app, and you’ll get the correct 4o behavior! The Projects even work there!

The problem is a bug in the mobile app, that forces every query to use o1, no matter which model you’ve selected. The web app doesn’t have this bug, and correctly gives you the model you’ve selected.