Daniel Jeffries (Unknown)

The Brave Search API brings affordable developer access to an independent index of over 20 billion web pages.

While it's built from scratch without big tech biases, powered by real anonymized human page visits to filter out data and refreshed daily with tens of millions of new pages.

Perfect for AI model training and retrieval augmented generation,

Whether you're working on language models or information retrieval systems, Brave offers representative data sets and up-to-date information affordably.

We'll get started with 2,000 free queries monthly at brave.com forward slash API.

Cool.

Andrew, it's an honor to have you on MLST.

I've been a fan for years now.

What's your bio?

What's your background?

Yeah, and that's really interesting because I think a lot of people, they put machine learning models into production and then they see all sorts of problems coming up.

So we need to have a holistic approach rather than just looking at the individual components.

And what are your broader interests?

I mean, how did you get to where you are today?

So we're going to come back to adversarial examples, but just to give the audience a taster, what are adversarial examples?

Very cool.

And when we look at the entire predictive architecture, how do you kind of sketch out that structure?

Yeah, and reliability engineering in general, it's really interesting because it's a moving target, right?

It's not something that you can ever solve because the system is non-stationary.

People are adversarially attacking your system in a myriad of different ways.