I’ve been creating a few GPTs in the ChatGPT interface, seeing as I had a document full of them even before the feature was rolled out. But I’ve got some confusion around the "knowledge"additions when I create a GPT.
When I click the button, it just brings up my file explorer with no restrictions on what to upload. So what the heck is this supposed to be? A text file? JSON? An embedding? A copy of Doom? How big can it be? How does it get parsed and prioritized?
I have the same question. When I ask the builder, it says that files are only stored in the memory for that chat and will need to be uploaded each time i want it referenced. I’m fairly certain that this knowledge feature is supposed to retain the files I send and remain with the GPT as part of it’s memory. Do I have this wrong?
I uploaded a python file of 700 lines of code to knowledge reference, deployed the gpt.
if asked about a section of the code, it is refering the .py file perfectly and explaining it.
i uploaded a .java file (through chat not through builder) and it refused to parse it says it can’t access the file.
then i tried to upload an ipynb notebook file to knowledge reference files (builder) , it broke the gpt for some reason both gpt builder and gpt itself only reponding “Hmm… Something seems to have gone wrong” .
what are the limits here ? what are the possibilites ? how many files can upload ? what types of files ? if their is a storage limit, is it per gpt or per user ?
I’m also curious about how does it index and retrieve the knowledge.
Does GPT just reach the whole knowledge base ? If so, is there a token limit?
If not, how does it retrieve/index? Embeddings?
I have the same problem with “Hmm… Something seems to have gone wrong”, Did you resolve this problem?
Same here, the functionality is very good, but there is 0 documentation, it is not very serious guys, when you launch something, the users should know how to use it
I guess it is just a database to be queried by ChatGPT. It’s only available for Code Interpreter. ← This ought to be wrong
See Difference between 'analyzing' and 'search my knowledge'
I’m having the same issue. I assumed it could read PDFs or TXT files and hold them in its knowledge base for reference and, ideally, as “learned concepts.” I uploaded 15 books on the subject I was configuring for and the whole thing broke. I started deleting files/books and that helped, until I tested again with just one or two, and, really, the whole thing slows down so much, and the responses (even “summaries of the book”) are so inelegant, that I feel it or I am broken.
Maybe this helps…I found useful to write in the instructions “If you find any error in the knowledge references just skip it and continue with the next one.”
I have uploaded file to Knowledge for particular bot, but says it can’t read the file ??? so what is the point of this !!!
GPT Builder
Understood. The ideal scenario would be for the bot to have direct access to the specific content of the book you uploaded, so it could tailor its responses based on that particular source. However, given the current limitations, the bot cannot read or extract information directly from uploaded files.
Me too, got confused about this feature. I uploaded a zip file to the knowledge and published the GPT. When I chat with it, sometimes it says, It seems there was an error while trying to access the file you provided. But at some times, it works. Most of the time, it failed.
Just test first with a .csv / .xlsx / .pdf / .doc / .htm(l) or plain .txt file.
And do state in the custom instruction how / when and why it should consult those custom knowledge files.
When you have the thing up and running, you can create a custom GPT with custom knowledge in less than a minute and it works like a rocket on steroids.
Sometimes a create a GPT, test it’s functionality and delete it after 5 minutes.
It’s so easy and good working, than when I need it again, I can rebuild it in 30 seconds.
After replacing the zip file with JSON, it worked successfully.
Hey guys, was wondering if anyone could please tell me what is better to upload to the KB, the full transcript of a video or a summary? I uploaded a bunch of transcripts and this GPT is not enjoying life.