OpenAI API GPT

[wet bison]

An unfinished GPT with a limited set of documents comprising the OpenAI API Documentation and API Reference pages.

Please feel free to share documents here to add to it's context.
https://chat.openai.com/g/g-y70GVFsqM-myapi-gpt

[faint lotus]

I thought OpenAI was doing this with base ChatGPT

[maiden kite]

really? source? cause chatGPT never seems to have up to date knownledge of openAI APIs which is annoying

@wet bison how did you download the documentation from their website?

[wet bison]

Just did a PDF print of selected text on the page. But I plan to do more of a scrape for raw text when I have time. Not sure if the PDF is in anyway worse for retrieval of text data than a text file.

There are 8 pdfs in total, I can share those here when I'm back at my laptop

[maiden kite]

yes pls I would like to experiment with that as well

[wet bison]

The API Refence page is one doc. The Documents page is all separate pages that you have to navigate to, so I just grabbed the some I was interested in and a couple of their internal links to more info

I plan to fill out the GPT with as much docs as I can get from the page and it will take in. Hopefully tonight.

[maiden kite]

good luck this would be very useful, thanks for the effort!

[faint lotus]

rip I guess you're right

[wet bison]

If we analyze why it might not be included, it's both possible and not. They would have, in a way, had to have the documentation written to inject it into the flow. A simpler approach, I don't see why it couldn't be added as file context in the same way as adding it to an assistant (which I am sure there are many in the "all-tools" model, master takes in request decides which assistant needs to run the request, activates that tool assistant, returns it's response, other stuff). But if they don't do it in that way, they would have to have the API and how to use it predefined, coded, and documented, prior to training the model so that it could be used in training.

[leaden escarp]

Hello neurart! Did you just go to the website and use an extension to create pdf of the entire website page by page?

[wet bison]

I selected the page content I wanted in a PDF, then used print and chose pdf, then under extra setting chose "no headers and footers" and "selection only". This gave me a PDF without the extra fluff.

I plan to try scraping the pages as plain text data, but will have to test it to make sure the extra layout and formatting offered in a PDF is not lost in translation and a plain text file will simply offer the same results. For now I haven't tested it much.

[leaden escarp]

Thank you for the reply ☘️
I will do this too, does the custom GPT work well??

[wet bison]

It worked better after scolded the builder and told it to do a better job of referencing its documents and promote vision, assistants, and threads over all other code suggestions. I have to put more data than the 8 docs into it and test it out more. If I have time tomorrow.

[leaden escarp]

Haha, okie.
I hear if you ask it nicely, and tell it that your career depends on it, it will do better XD.
There was a paper on a subreddit regarding this!

Thank you though friend!

[meager flame]

nice GPT and nice tool

[desert juniper]

@wet bison it seems that it isn’t referencing its knowledge base unless it is told to. I think this is a common issue. It automatically defaults to the last training update. I’ve been having this happen with my GPTs too. Tried tweaking the instructions but not much luck. Anyone know how to fix?

[steady sage]

I did a non-knowledge base implementation of this and it works very well. I go into details of how it's implemented in the thread. Maybe it'll help#1172289801098109009

[desert juniper]

Oh, ok. I’ll keep playing with it. It just didn’t get the Assistants api usage right when I asked the first time. Good to know, I’ll go back and read.

[calm bloom]

How did you upload the whole api documentation? Is there a PDF or something on their website regarding their documentation?

[wet bison]

I threw this up in the first hour I had access, and have since been working on other things. I see your's and another user's on twitter. I will likely be abandoning this for now and suggest others use your's.

[wet bison]

I had coincidentally manually saved a couple docs to PDF for personal use the night before and just reused them and uploaded them. Outside of the first hour or two that I had access and set this up, I have been working on other things. @steady sage has linked what appears to be a superior implementation, and I suggest users head over to that GPT as I will likely not be making any updates to this right now.

[steady sage]

oh someone posted it on Twitter? Can I have the link?

[wet bison]

Someone post another one on twitter, let me see if I saved anything from it

[steady sage]

thank you!

[steady sage]

ah ok