Reply to thread

Message

[QUOTE="B1GEast, post: 5332894, member: 39"]
It's programmed to provide a result. It returns a token. GPT has tens of billions of parameters if not a lot more.

There probably nothing that it hasn't seen since it uses not only the LLM that it's trained on, which is exhaustive, but it also uses RAG.

Mark is a smart guy, but his statement is off by a mile.

It's not about returning a response with I don't know as much as it is I don't understand contextually how to provide an answer.

I've seen the transformer respond with "here's where it gets tricky" and indicate that the response may not be conclusive and then say if you provide more detail so that I can better understand.

That logic that I described is incredible.

And yes, it does hallucinate at times, like really fail, and there can be many reasons for that, like poor prompting or overwhelming it with too much data.

GPT is at a PhD level and for most of what I use it for, I consider it to be an incredible force multiplier.

We have only scratched the surface, it's scary.
[/QUOTE]

Verification

First name of men's bb coach