Illia Polosukhin: AI thesis, crypto x AI, and announcing the Delphi Labs x Near accelerator
Episode publishedPodcast: The Delphi PodcastThe Delphi Podcast
Transcript preview
Preview only
You're now plugged in to the Delphi podcast, my guy. Hey,
guys. So Josemarie Masaidu here hosting a special episode of the Delphi podcast with special guest, Ilya from NEAR. We're actually announcing our crypto XAI accelerator with NEAR, Delphi Labs and NEAR crypto XAI accelerator. And so to celebrate this, we thought we'd bring Ilya on the podcast, obviously, for for those who who don't know, he's one of the authors of the original transformers paper that that's pretty much the predecessor of today's LLMs. So we wanna cover a bunch of stuff. We wanna cover Ilya's background, which I'm actually really interested in, his views on AI, the crypto XAI th
Yeah. So I was actually interested in kind of AI machine learning from pretty early on. I think, like, from 10 or or 11 years old. And Classic. And I was trying to build, like, my first neural networks when I was, like, 14 or something. And so I went to study applied math, but I'm also, like, from, you know, not very well off family. So I was looking for a job right away and kinda through a set of circum circumstances, I got a job for an American company that had an office in Kharkiv in Ukraine that was doing machine learning. And so that that company actually had been doing machine learning f
And so it was This was the software systems?
Yeah. Software systems. Yeah. And so it was a huge learning opportunity because I was, like, during the day in school, pretty much learning, you know, algebra, like, discrete math, control systems, etcetera. And then during the evening, pretty much working and doing, you know, decision trees, pretty much regression, like and the company did a lot of consulting projects. So I would work on, you know, finding oil in oil fields based on data or, like, detecting cancer based on DNA sequencing. So a lot of just like a lot of project natural language understanding, etcetera. So it was really good. B
so this was back when I was still sort of, like, rules based, more like econometrics and and
I mean, this is, like, 2008. So, yeah, most of the machine learning back then and, you know, like, even Google Google Ads, everything until, I think, like, 2013, 14 was powered by these algorithms. So gradient decision trees, boosted decision trees. And so, yeah, so I worked on that. They moved me to US, so I moved to San Diego, and worked there full time, after finishing school. But I ended up going to Silicon Valley pretty much every month because there's conferences, friends, kind of more exciting stuff going on there. And so I'm like, well, I should just move there. And I applied. I kind o
Anymore.
And so so for me, it was always, yeah, like, how do we how do we take all of this knowledge we have on Internet and how do we kind of be able to answer questions and be able to, like, teach machines to answer questions. And so that's why I worked on a team doing question answering and, obviously, like, there's a lot of auxiliary tasks like machine translation others. Interesting.
Keep exploring with a free account.
A free account preserves the short preview. Upgrade to Pro for the full transcript and episode chat.