Beth Lyons

I'm going to run it through perplexities computer.

I'm going to come back over here, but this is the way I, I sort of natural language talk to my computer.

This is my process.

But at the end of the day, I'm like, but I use four different tools because I'm bouncing ideas off of this, this builder, that builder, you know, Jeff in our comments, uh, really likes, um, uh,

Yeah, I love it.

And I look to see more of this.

We've talked about this many times in the show over the years.

As latency drops, as costs drop towards zero, it only gets easier to say, well, have a couple models check that answer.

I've always said that I think, just my opinion, that the true answer to hallucinations, that idea, that concept of hallucinations, is simply to...

ping it against so many different runs, you know, in parallel that the chances of it hallucinating would be extremely low.

Cause sure.

One hallucination.

Okay.

I ran it through 10 times.

Okay.

Two of them are hallucinations, but even at 10 times, eight of them were agreeing with each other, you know, and that, that gets to that counselor level that you were just saying, Andy, like, okay, but now you have the counselor or we would call it the AI agent orchestrator level.

right where who's taking in all the advice from all the different agents and saying like okay man you know it's like the old uh the old shows when i was a little kid one of these things isn't like the other uh now it's time to play our game anybody remember that what was that romper bumper stomper boo anybody remember that anybody yes they never said bath ever in the in the magic thing never that's rough

That's like a baby back memory that just came down anywhere.

One of these things is not the same, right?

That may be even Sesame street.