Joel Hron

analyze the brief against case law and say, like, is somebody extending the law?

Are they like, you know, using sound logic here in terms of what the law actually says in making this statement?

And so if you think about that, that's actually a tool for a research agent to use.

Like, hey, am I am I making up BS right now?

And and it can use these kind of things along the way to help, you know, control itself from from a hallucination standpoint.

So we actually build, I think, tools and logic into the process.

And then the third thing is really just training.

Like I said, I guess, you know, if you do that one time as a lawyer, you remember like, OK, I need to check this next time.

Right.

And so, you know, we hope that that doesn't happen to any of our users and we try to train them in a way that.

to, to indicate to them, here's how you should use these visual clues.

Here's how you can read like what the agent is doing.

And, and here's where you should spend your time validating or invalidating certain parts of this answer.

So I think this, this element of training and change management for attorneys and, you know, to the extent we can, we, we try to lean in and really help them with that process.

But I think it's really all three of those things in combination that

Yeah, no, 100%.

I would say we have a pretty agnostic model philosophy internally, and we're constantly benchmarking the latest models as they come out.

Smart.

And hallucinations are one part of that rubric.

I mean, obviously, they have much more, you know, I think, breadth to them than just hallucinations.