Rory pitches an AI approach using fewer, smarter neurons instead of massive numbers of simple ones. A hackathon model was 90% smaller on the same data and cut cloud costs 38 times on Google Cloud. The software runs on existing NVIDIA GPUs.
Chapters
is basically making their decision-making teams with 20 million toddlers, giving every little toddler a very simple assignment, and you can build ChatGPT with T with that. But if you have a more intelligently designed neuron, you're building your team with adults. And instead of having 10 million toddlers, you have 100,000 adults, and it's equal intelligence with a significantly more efficient system.
What I'm struggling with is, is this a revolution?
We just ran a hackathon to validate the tech with some external users, and the winner was the CTO of an AI consulting firm who was able to build an AI model to be 90% smaller on the same dataset. And because of the significant compression, the savings in electricity and compute requirements, he actually deployed this model onto Google Cloud and was able to achieve a 38x cost reduction.
And is this all going to be on a CMOS chip? Is it gonna be on a processor or is it gonna be on something else?
What's it gonna look like? It's all software. That's one of the best things about it.
Oh, you build it in software. So what does it run on? What's the chip?
All the same things people are using right now. So yeah, NVIDIA chips, GPUs.
If you're successful, what happens to all the large language model companies today? Is there— is it gonna be a disruption?
Yes, but they also would be my customers, not my competitors. I feel like every single question they asked today are things that if I had a 20-minute presentation, I would've put them in there. That's where all my anxiety was. offer model search space expansion. So if you delete our code from your training repository, you will have a worse model. Whereas with every other MLOps company, you could delete it, but if they already helped you make the right choices, they're not actually enabling something that you couldn't do on your own.
Terrific. Well, thank you, Rory.
Thank you.
Very nice to see you. Thank you for the glasses.
Of course.
Been a little stressed for the last 5 days or so. Been kind of binge-watching Meet the Drapers to prepare, practicing my talk. Definitely gonna blow off some steam today. Dad, what'd you think of Rory?
Well, he was okay. He didn't really knock me over. I wouldn't want to go work for him. Andy, what'd you think?
I think this is the kind of thing early-stage venture capitalists that I do. I take risks on a large market. He mentioned that this will be his first funding round, so I, I'm very open to the idea.
To Andy's point, it's like one of those things where I'm like, I don't understand I don't understand this enough, but I can tell that there is something there, but I haven't been able to put my finger on it. But it gives me the confidence that this is something I should look into more.
Okay, let's move on to the next entrepreneur. But before we do, what do we do? See what's going on behind the scenes. Is this a science project or is this gonna be an entrepreneurial venture that becomes a monster?
Yeah, so we're definitely at a critical moment right now. Like I mentioned, I have just been onboarding my first pilots, I'm not sure how this is gonna turn out, so it's awkward to be on the show right now. But if this does get industry validation and people start paying me, I feel extremely confident that everybody is going to be paying me, that I will really be the person behind the next generation of AI models where everybody is building with my tech.
And how do you get dad to wanna work for you? I don't know, we'll have to grab a beer after the show. And Ali Polta.
