Jeff Dean
π€ PersonAppearances Over Time
Podcast Appearances
And there, I think it was just all in memory on one machine.
Yeah, I think it was one machine.
Yeah, yeah.
And it just nailed it every time.
Yeah, and I guess that was language modeling.
Yeah.
Not really.
I mean, like I don't think I ever felt like, okay, Ngram models are going to, you know, are going to⦠Sweep the world.
Yeah, the artificial intelligence.
I think at the time, a lot of people were excited about the Bayesian networks.
That seemed exciting.
Definitely seeing those early neural language models.
Both the magic in that, okay, this is doing something extremely cool, and also it's just
struck me as the best problem in the world, in that, for one, it is very, very simple to state.
Give me a probability distribution over the next word.
Also, there's roughly infinite training data out there.
There's the text of the web.
You have trillions of training examples of unsupervised data.
Or self-supervised.
Self-supervised.