Chris Sacca

Put away the phone, Jason.

How?

How's it been three years in?

This is a three year anniversary, like a couple of years, three years.

Yeah.

Did you seed it with Wikipedia?

Actually, take a step back.

How did you guys, how did you do this?

But meaning totally unsupervised, just a complete training run on its own, totally synthetic data, no seeded set, nothing?

This is what I was going to ask you.

Do you think that you can take this corpus of pages now and get Google to de-boost Wikipedia or boost Grokopedia in traditional search?

Because a lot of people still find this and they believe that it's authoritative because it comes up number one.

So how do we do that?

How do you flip Google?

Wikipedia, yeah.

When you have this version that's maximally truth-seeking as a model, do you think that there needs to be a better eval or a benchmark that people can point to that shows

how off of the truth things are, so that if you're going to start a training run with Common Crawl, or if you're going to use Reddit, or if you're going to use... Is it important to be able to say, hey, hold on a second.

This eval just... You guys suck on this eval.

This is crappy data.

Looking good.