#Document access problem.

1 messages · Page 1 of 1 (latest)

glossy bronze
#

Hi everyone, maybe somebody could help. I built a GPT, but the model constantly refuses to access my uploaded documents. They are all perfectly within the size limit of the GPT builder and have an appropiate format (docx). Anybody has an idea what the specific problem could be. Thanks a lot for your time and sharing any advise in advance.

deft wagon
#

Have you tried putting it in the "Instructions" in the Config tab to reference the uploaded documents?

toxic oriole
#

Custom Instructions could have something worded wrong - a directive to “always use all available knowledge” resulted in an inability to web search, try turning them off first.

Other strategy is using trigger keywords in the document, directly invoking those for specific parts seems to work if that keyword is tied to a directive ‘activating’ that part of the document.

glossy bronze
#

@deft wagon I tried it in my last GPT, didnt work somehow. With "reference" you mean upload it in the Config section, or put the whole content in the tab?

#

@toxic oriole I will check them. Thanks you both for the advise.

plush chasm
#

There's currently a technical issue causing this. I and some others have been seeing the issue for several days now, but the custom GPT finally acknowledged the issue for the first time today:
GPT
I can provide feedback and insights based on your manuscript, including summarizing the final chapter. However, due to a technical issue I'm currently unable to access the manuscript directly.

glossy bronze
#

@plush chasm Exactly, that was the message. I follwed also the advise of DocWobble. Now at the moment, it is working. Unfortunately I gave up the prototype of my current GPT, cause of this issue and started anew. Thanks to all of you again.

#

Uh, now it hallucinates a bit.

deft wagon
deft wagon
toxic oriole
#

Yeah I saw this at least a week ago, at which point the GPT was saying it was due to ‘privacy and copyright protection’ - except having it disable verbatim access to all document uploads maybe wasn’t the most subtle approach if that’s the case. It’s basically made every model that needs precisely worded Knowledge Base info nonfunctional, the trigger word tactic is a workaround for having it actually run through pseudocode/etc correctly in documents.

glossy bronze
#

@toxic oriole Thank you. Does anybody know how many token context the GPT has?

noble pawn
#

Could you elaborate more on this? Just for clarification. Is it that it won't reference the documents directly, or is it more of, it's like you didn't load anything at all? Like the AI is unaware there are even documents there, and when you tell it to look at the files, it tells you it can't? Can you walk us through how the interaction goes? I'm not sure I will have any insight into this, but I'd like to know more about what you're experiencing spcifically.

glossy bronze
#

@deft wagon Thank you.