Grant Harvey

Okay.

Yeah.

Well, OpenAI had this paper not too long ago, right, that was saying that basically like hallucinations are a training problem, or at least that was their thesis, that, you know, the models are effectively being rewarded for guessing as opposed to saying when they don't know.

Do you buy into that?

Do you subscribe to that idea?

And if so, is that something that you could possibly, you know, work on to help reduce that even more?

In a sense, is it kind of a miracle that any of this works in the first place?

Absolutely.

I actually have a question related to this, which is a lot of our readers and viewers are trying to figure out how to implement these systems in their business, in their workflows, right?

As someone who builds and has worked with these types of models for years, how do you recommend that they...

I guess, close the loop in terms of implementing this and actually assess the quality when you're actually in production scale.

Do you have any tips or tricks there of how... Basically, what's your recommendation for how to assess whether they're working or not?

Where should listeners go to give Mercury a try and learn more about diffusion language models?

What if your customer service team could handle 90% of tickets automatically with 99.8% accuracy and still feel human?

Today, we're talking to the guy building exactly that.

Welcome, humans, to the latest episode of The Neuron Podcast.

I'm Grant Harvey, writer of The Neuron Newsletter, and today I'm talking to Matt Price, CEO of Crescendo AI.

Matt spent over 13 years at Zendesk helping build one of the biggest names in customer service software, and now he's betting that AI plus humans together will beat AI alone.

I like that concept, so I'm excited to chat with him about how he's doing exactly that.

Matt, welcome to the show.