Similarly I too am having issues with a private GPT, it has now all but forgotten to access my custom GPT uploaded ‘knowledge’ and also I note that responses recently have been underwhelming, it all seems a bit off.
1 Like
Mine seems to be working fine now. Maybe it’s a general issue, but I’ll keep this in mind for larger knowledge bases in the future
I’m now at the point where my private GPTs can read neither knowledge files or files uploaded in the chat (unless with the code interpreter).
2 Likes
More perplexing: GPT4 on its own has no trouble with uploaded files.
1 Like
Have you tried code interpreter, or asking it to make an identical copy?
I’m not sure cutting out the knowledge base would be relevant to your use case.
Yeah, that’s a bit all over the place. Must be a general issue.
1 Like
It suggests to me that the issue is specifically with the Builder system. I’ve also tried it with much more generic GPTs in the Builder and it still can’t access files.
1 Like
it works if you activate code interpreter but imo showing your knowledge base is not the way to go.
+1 to anyone who complains about lack of transparency and communication
a moderator simply acknowledging the problem and that they work on a solution, ideally giving some context would be enough.
sad to see.
2 Likes
adding here that it works bad. but at least it works. e.g. when having a 12 step guide of “how to be in the now” in the file
it simply finds the heading and then stops reading lmao
tried to change it all into one paragraph - no change
1 Like
Looks like the browser tool isn’t even in the custom gpt…
there should be a "type": "myfiles_browser". I have dalle and code interpreter enabled
Why did it get removed? afaik it is impossible to add the browser back in via the gpt builder.
2 Likes
I have the same problem my GPT only seems to read the first three documents when I ask them to be listed.
It worked fine last week.
Any ideas, merge the context data into on file?
I’m having this issue too. It’s telling me it can’t access files in Knowledge (.docx files that are 3-4 pages). When I point out it previously accessed them it says I’m correct but it still can’t directly access them and needs me to provide it with clues or context.
Very strange!
1 Like
Exactly the same issue when i import a PDF
It feels like recently there was some change that severely depreciated custom instructions (their importance to the AI). It’s doing weird things, acting lobotomised whenever it has knowledge files available, random refusals stating guidelines for no valid reason whatsoever and general lazyness. It’s worse than its ever been right now.
1 Like
I’m also having this issue since was it the day before yesterday. Very frustrating, as I have several GPT:s with lots of specialized knowledge, and all of them are now just replying general GPT information, so none of them are very good at anything anymore… Hope this is fixed very soon! The possibility to upload knowledge was what made me pay for the subscribtion…
3 Likes
Probably solved in a day if someone from OpenAI WOULD HAVE THE COURTESY TO LOOK AT IT
Thats what makes me mad the most - no feedback, no reaction, no acknowledgement of the issue.
OpenAI, you will lose heavily once your competition will launch a similar program - but with a REAL focus on developers.
4 Likes
browser is “Browse with Bing”, Capabilities: Browsing in the “About” of a GPT.
In “About”, there will be no indication of documents uploaded, even when they are operating successfully.
I did notice a problem with a GPT I just created. The five uploaded documents reported as available to the AI are duplicated in AI context. Example (placeholders for file IDs):
User uploaded file with ID 'file-1111111111111111111111111' to: /mnt/data/2210.07316v3.pdf.
User uploaded file with ID 'file-2222222222222222222222222' to: /mnt/data/2305.10435-testpaper.pdf.
User uploaded file with ID 'file-3333333333333333333333333' to: /mnt/data/2209.05433.pdf.
User uploaded file with ID 'file-4444444444444444444444444' to: /mnt/data/2302.02083.pdf.
User uploaded file with ID 'file-5555555555555555555555555' to: /mnt/data/1001.2072.pdf.
User uploaded file with ID 'file-6666666666666666666666666' to: /mnt/data/2210.07316v3.pdf.
User uploaded file with ID 'file-7777777777777777777777777' to: /mnt/data/2305.10435-testpaper.pdf.
User uploaded file with ID 'file-8888888888888888888888888' to: /mnt/data/2209.05433.pdf.
User uploaded file with ID 'file-9999999999999999999999999' to: /mnt/data/2302.02083.pdf.
User uploaded file with ID 'file-0000000000000000000000000' to: /mnt/data/1001.2072.pdf.
Only the five files with original names are in the Python sandbox.
The GPT AI cannot answer about a specific document, not even finding their titles, only answering about the first injected document.
An additional message seems to appear with the “knowledge search”. Is it failing on file IDs that don’t exist?
Or failing on the supervision injected?
Invoke the search command of the myfiles_browser tool before continuing to respond to the user.
1 Like