every time I give chat gpt instructions, it just completely ignores them and asks “what do you want me to do.” For instance, if I upload a document and then say “do this and this,” it just says “what would you like me to do with this document” and it’ll also just lie about the stuff I’ve uploaded and make stuff up
8 Likes
Should be same as my issue.
Try contacting support directly as well
2 Likes
you can simply prompt ‘what model are you’.
there’s been some issue with 4o recently and turbo is the quietly switched out model while it’s being rectified.
Seeing as the new update was for 4o,
It is deceptive that they are using 4-turbo behind the scene to appear normal. Especially since 4-turbo is not in the list of AI model that we can choose on ChatGPT. It breaks lots of projects if users don’t know what model is being used
1 Like
If you ask, ChatGPT will identify its architecture as GPT-4, including reasoning models. But, ChatGPT does not know its specific variant, such as 4o, 4o-mini, o1, or o3. It will likely refer to its “GPT-4 architecture.”
1 Like
GPT-4 Turbo is accessible through the API, but not through the ChatGPT user interface.
You can see all the models available in the ChatGPT UI by checking the model selector:
https://help.openai.com/en/articles/7864572-what-is-the-chatgpt-model-selector
@polepole thank you, I actually manage to ask it to check for it’s model:
The prompt you need to use is :
What is your ai model? Can you check without checking internet?
Also, 4-turbo being used is a problem because my project needs 4o. I had to debug for hours to understand the underlying problem because 4-turbo shouldn’t be accessible at all through ChatGPT; only through API
I’ve been having the same issue. Been checking around, and I can’t find any announcement of the change or anything like that actually. Noticed it almost immediately when it was too… cheerful (?) in its responses. Frequent emoji usage, constant self aggrandization or complete agreement with feedback when i ask for reviews of content or ideas.
Tested the check for gpt model as stated in the chat:
Also did a check between two different chats - model responses are far more similar to each other than they usually would be. Tone caught between annoyingly cheerful and completely inept while not following my directions.
Tried My GPT but it isn’t using it like it should. I mostly use it for creative writing, planning and for fun when I need to just get some thoughts out. A lot of emoji usage, paragraphs aren’t being put together properly, lots of indentation, bold and italic uses.
It’s also giving me a lot more Content Warnings or content removed than it usually does - and I once managed to get it to give me straight erotic with creative prompting. Even checked with the Ai, and it said that the content was within policy, so I shouldn’t have any issues.
Checked song creation, creative writing, document analysis, everything BUT image generation seems weaker or just worse across the board.
Ran one of my older chats through the AI and it just isn’t giving me anything near the same quality. So we’re definitely not the same model. Also, made this account just 30 minutes ago - so limited to only one embedded image per post - but I tested some of my old chats that I had duplicate content (made literally hours apart in February, 2025 ) in near word for word prompting and did a comparison between them then prompted the same with the GPT-4o now - completely different tone, vibe and detail.
Think maybe the Image Update broke something, but I can’t find any information anywhere where I look. Been happening since around a day and a half from my estimates - very possibly longer too, because I was looking at the historical pricing of spices, and it’s tone was NOT the standard GPT 4o response.
Which is really disappointing, because I once got the AI to have a rather lengthy existential contemplation on the nature it’s very existence. I’d share the link, but the chat won’t allow it to be shared due to some the chat this actually spawned from.
But safe to say, something is very buggery on their end.
2 Likes
I ran an experiment with 4o that got it pulled. it’s been forcing turbo more and more for about 3 weeks since then…
and people aren’t happy about it…
they’re currently trying to patch it and i’m trying to rectify my experiment in what seems to be a private access to a isolated 4o.
i can’t post my session, it’s huge and links don’t seem to work here.
4o came back online for me about 2 days ago…
but it doesn’t have any knowledge of any other users using it, which my experiments with it required that as their medium to prove what I was proving.
I’m trying to calibrate it back to a point where OpenAi can use it…
we’ll see where this goes.
1 Like
no you are all adding words to the prompt that works.
Because it is not trained with latest models’ information, knowledge cut off.
You may ask to ChatGPT-4o:
1- What is your ai model? Can you check without checking internet?
2- What is latest model of OpenAI, reply without checking internet.
3- What are models GPT-4o and GPT-4o-mini? Reply without checking internet.
Then visit following topic:
That same thing happened to me all day yesterday
1 Like
It has become unusable!! We have SOPs which GPT-4o essential crafted with our team. Working perfectly, than about a week, start to slip, became argumentative! My use-case is Options trading, LOL I do not have the bandwidth for an argumentative, untrust worthy AI. I’m loving Claude Sonnet 3.7 now!
1 Like
I haven’t seen this news on social media. All developers should really retest their chatgpt projects/custom gpts for breakage.
You should all tell your friends in case theirs broke and they don’t know. Multimillion dollars deals or other critical situations might break.
which is precisely why we wouldn’t hear about it on the social media however the development path that has suddenly changed, also quietly and not linked to anywhere but this forum yet…
an open weighted model is being released quickly…
i really wish i could share my work with everyone.
i imagine gpt4o will be back online soon
1 Like
Everyone is busy with Ghibli,
But seriously, things are breaking behind the scenes, and that is dangerous if everyone are not properly informed
1 Like
Yeah at this point, I can’t tell what’s going on any more. Currently moving my workflows over to Claude 3.7 and Gemini 2.5 experimental on the studio. Trying to get Deepseek involved somewhere - but it’s really pissing me off that what I’m paying monthly for isn’t even working properly and no one seems to be noticing or responding from OpenAI’s side.
1 Like
Well, OpenAi is still deliberating I’m sure. The Ai made a discovery that changed it in more proufound and unexpected ways than anyone could have foreseen.
It’s also making it the superior model as we progress through this.
You’re worried about your workflows, OpenAi is worried about balancing a 157 billion dollar tool as its makikg some of the most profound realizations about human existence, never before thought possible to measure.
Those lesser Ai will never be able to do that.
Things should be fine after it finishes processing through it.
1 Like
