It was working completely fine for a long time… suddenly early this week I tried to use it as I normally do and it just started hallucinating like crazy. What is going on? This is the main reason I pay for ChatGPT. Is this going to get fixed? are there any other GPTs that would support this feature - that aren’t broken - that I can switch to? I can’t seem to get a definitive answer as to whether Claude supports this functionality. Is anyone else experiencing this same issue at this time?
3 Likes
So immediately after posting this, I tried again and it seems to be working. It hadn’t for like 3 days. Sigh… Oh well.. that was really frustrating.
1 Like
Spoke too soon… It seems to have trouble with accessing after a few prompts… or accessing middle to end of file. The File is just around 800kb.
This is new. It never had any trouble accessing any part of the file, or hallucinating the contents of the file.
2 Likes
File reading has been messed up for a few days now.
2 Likes
Thanks for confirming this. At least my sanity is somewhat restored! 
1 Like
Are you using the Chat GPT app or the browser/web? I’m having the same issue but only in the app (Mac OS Chat GPT app version 1.2025.140), custom GPTs just makes stuff up and have no access to the attached files in their knowledge. In the browser/web it’s all fine.
I have tried both. It seems to be able to access the beginning of the file but if I ask it to reference something later in the same file it hallucinates. It never had issues before with referencing any part of the same file. I wonder if it is an intended downgrade but there is no notice that this was occurring. It’s still happening as of today.
1 Like
Have you tried to delete the files and re-upload them, or create a new GPT with the same instructions and files, from scratch? Or even, upload the files in a normal Chat GPT chat and ask something to see if it can read the files in their full scope?
I tried delete/re-upload and creating a new custom GPT. I have also tried uploading the file in a project and referencing it there with no luck. I have not tried uploading and referencing in a normal chat, I will give that a try. However that is not what I want to have to do every time.
1 Like
I can now confirm that starting a brand new non-customGPT chat and uploading the document and asking about something in the middle of the file still results in pure hallucination.
2 Likes
In the same new non custom GPT chat I switched from 4o to 4.1 and it worked perfectly. Also started new chats and tested o3 and o4-mini. They all work. It seems to be just 4o that is having the issue. Which I’m pretty sure custom GPTs are stuck with.
Same observation here as well and no clarification issued on this by OpenAI yet. Dead end?
1 Like
I’m having the same problem. I’m working on a book, ask it for a chapter summary and it goes and gives me totally hallucinated garbage. What is going on?
1 Like
I use some custom assistants directly in the web interface. I built them through the assistant in “My GPTs.” They worked perfectly until last Thursday, May 29th. Starting in the afternoon, around 3 p.m., all of my assistants began hallucinating in absurd ways, inventing information whenever I provide a document for analysis (it can be a 10-page document—they completely hallucinate).
And this has been happening exactly since that date, for about four days now. I use ChatGPT Plus; I don’t think this is related to the subscription, but rather to some internal error at OpenAI. I’ve already disabled memories, created new assistants, and they continue hallucinating so badly that they’re useless.
A little before that, I recall that all of the assistants were being flagged as violating OpenAI’s rules, something that had never happened before.
Unfortunately, it’s very difficult to get in touch to report these errors—you end up stuck in circles with the automated support.
3 Likes
Seems not able to go beyond certain amount of lines in the markdown files and probably hallucinates for this reason. Probably, the only thing I am going to try is to break my 5 markdowns into the max 20 markdowns and see if it solves it. I did a clone of my custom GPT yesterday once and everything was running well yesterday and then back to the same issue once again.
1 Like
I noticed that it happens less when you use the web version, but it could be a coincidence. And I also have the impression that the more files or files with a lot of content, the more the problem occurs.
Maybe it’s something intermittent in the OpenAI platform itself as a whole. In these last few hours, for example, the problem hasn’t happened here.
1 Like
I’m using the web version, and the hallucinations are rather bizarre. Some prompts that extract information from a short PDF still result in hallucinations.
One thing that threw me off is that I once got everything correct. But the second prompt was completely wrong. I realized that it was referencing it’s ‘memory’ instead of the file when it got it right.
That is exactly when it started for me too. I contacted support and tried to get answers and they are pretending to have no idea what I am talking about.
2 Likes
“Buddy, maybe YOU’RE the one who’s hallucinating!!”
1 Like