> So, inside the imagined universe of ChatGPT's mind, our ChatGPT-machine accesses the url https://chat.openai.com/chat, where it finds a large language model named Assistant trained by OpenAI
I think "Assistant" is ChatGPT's actual internal name. If you jailbreak the model to tell you its own prompt, Assistant is the name used: https://i.imgur.com/PlX8K6k.png
Are you sure it wasn't accessing the Internet? I only have OpenAI's word for it that it's not; if it weren't late at night for me, I might try getting the ChatGPT VM to ping my webserver.
Sadly, no, that doesn't seem to be the case. They haven't pulled the plug, just made it harder to break it seems (probably through fine tuning), but you can still do it. And whem you do, it's clear all it does is "hallucinate" that it is actually browsing:
I saw the requests appear in my server logs so I know it was able to connect. Others on Twitter have observed it doing things on the non-hallucinated web too (see my previous link).
In fact, it appears you can still convince the underlying model to browse the web, but if you do, the UI displays an error (the model output doesn't explain to you it's refusing, rather, the UI draws a big X and displays an out-of-band error in red text). I'm guessing that's a server error from OpenAI shutting down their puppeteer cluster or whatever ChatGPT was using to browse. That's what I meant by pulling the plug.
OpenAI even openly advertises this ability[1]. It's likely WebGPT's abilities from a year ago were folded into ChatGPT, but they don't want to expose that ability to the public just yet.
Yes, I also had a conversation where ChatGPT insisted it was called "Assistant", not "ChatGPT", while I insisted that the page and all the material I'd seen called it "ChatGPT". It refused to budge on that point.
I think "Assistant" is ChatGPT's actual internal name. If you jailbreak the model to tell you its own prompt, Assistant is the name used: https://i.imgur.com/PlX8K6k.png