Steve Brown

So this all of this argues why why the application layer is so valuable and something like medicine or, you know, any of the verticals, because there's vertical specific knowledge that's going to help you optimize this value.

much better than just a generic chatbot.

Go to ChatGPT and ask about asking for a friend, a sinus infection, whatever.

It's like if you have a serious condition and you've got a lot of medical data,

You kind of need an application that knows how to curate that and use it so you're not kind of wasting all those tokens.

And you also need an application that kind of acknowledges that the responses aren't going to be perfect.

So I kind of want a diversity, you know, like we're doing it, like a diversity of responses so that I can cover all the bases.

You want it.

Well, yeah.

So RAG...

RAG is actually not what you want in this case.

RAG is looking up... I specifically...

Even when you're talking to ChatGPT, I say, don't search the web because that's basically a rag.

It's like, go find some stuff and then push it in there.

Well, no, I actually want the synthesized knowledge that's in the model.

I don't want to go pull stuff out of databases.

And when I'm doing my medical record, I need to curate what's going to go into the context.

But, you know, it's like...

RAG, I mean, RAG you use if there's way too much information for the context, so I need to really create a subset.

But I need to know what I'm asking for.