John Rush

every cyber security company knows ai can make their job easier so they're implementing this what i'm talking about is doing the flip side building a cyber security firm that's specifically designed to secure the now highest value digital assets in the world maybe next to crypto and bitcoin like if there's a whole bunch of bitcoin that's really valuable but other than that the next most valuable digital asset has to be the weights to these large language models and so

So OpenAI, it's rumored that they spent around $500 million training GPT-4. Now, all of those weights, once the GPUs go off and train the model, it all gets boiled down into these weights. It's just numbers in a file, essentially. It's in a very large database, but you could steal this. And there's a lot of...

So OpenAI, it's rumored that they spent around $500 million training GPT-4. Now, all of those weights, once the GPUs go off and train the model, it all gets boiled down into these weights. It's just numbers in a file, essentially. It's in a very large database, but you could steal this. And there's a lot of...

So OpenAI, it's rumored that they spent around $500 million training GPT-4. Now, all of those weights, once the GPUs go off and train the model, it all gets boiled down into these weights. It's just numbers in a file, essentially. It's in a very large database, but you could steal this. And there's a lot of...

fear-mongering and conspiracy theories in Silicon Valley right now to the effect that, oh, China has already stolen the weights to GPT-4. Maybe they already have stolen the weights to GPT-5. Who knows if it's real? I actually think most of these labs are pretty secure. But it goes without saying that these are going to be increasingly valuable targets for hackers to break into and steal. And so...

fear-mongering and conspiracy theories in Silicon Valley right now to the effect that, oh, China has already stolen the weights to GPT-4. Maybe they already have stolen the weights to GPT-5. Who knows if it's real? I actually think most of these labs are pretty secure. But it goes without saying that these are going to be increasingly valuable targets for hackers to break into and steal. And so...

fear-mongering and conspiracy theories in Silicon Valley right now to the effect that, oh, China has already stolen the weights to GPT-4. Maybe they already have stolen the weights to GPT-5. Who knows if it's real? I actually think most of these labs are pretty secure. But it goes without saying that these are going to be increasingly valuable targets for hackers to break into and steal. And so...

figuring out how to protect a digital asset like it's fort knox seems like something that's going to be a new and ongoing challenge for the artificial intelligence industry broadly now this is different from in past in past generations of tech because if you stole the code base to facebook and you were like okay i have all the code to facebook i'm gonna launch Facebook2.com.

figuring out how to protect a digital asset like it's fort knox seems like something that's going to be a new and ongoing challenge for the artificial intelligence industry broadly now this is different from in past in past generations of tech because if you stole the code base to facebook and you were like okay i have all the code to facebook i'm gonna launch Facebook2.com.

figuring out how to protect a digital asset like it's fort knox seems like something that's going to be a new and ongoing challenge for the artificial intelligence industry broadly now this is different from in past in past generations of tech because if you stole the code base to facebook and you were like okay i have all the code to facebook i'm gonna launch Facebook2.com.

Well, you don't have any of the users. People aren't going to your website. It's not valuable. But with the large language model weights, once you get the model, you can run that anywhere. You can stick that anywhere. You can wrap that into your own product. And all of a sudden, you're not paying OpenAI or Anthropic if you're able to steal the weights. And so...

Well, you don't have any of the users. People aren't going to your website. It's not valuable. But with the large language model weights, once you get the model, you can run that anywhere. You can stick that anywhere. You can wrap that into your own product. And all of a sudden, you're not paying OpenAI or Anthropic if you're able to steal the weights. And so...

Well, you don't have any of the users. People aren't going to your website. It's not valuable. But with the large language model weights, once you get the model, you can run that anywhere. You can stick that anywhere. You can wrap that into your own product. And all of a sudden, you're not paying OpenAI or Anthropic if you're able to steal the weights. And so...

These firms are going to need top-tier cybersecurity. They already do. A lot of these guys, they go to DEF CON. They talk to all the hackers. They have huge, huge resources allocated to this. But it's just going to become an ongoing problem.

These firms are going to need top-tier cybersecurity. They already do. A lot of these guys, they go to DEF CON. They talk to all the hackers. They have huge, huge resources allocated to this. But it's just going to become an ongoing problem.

These firms are going to need top-tier cybersecurity. They already do. A lot of these guys, they go to DEF CON. They talk to all the hackers. They have huge, huge resources allocated to this. But it's just going to become an ongoing problem.

And I think that there might be a new opportunity to figure out, to build a new company around securing these extremely, extremely valuable digital assets, which are the AI models. I think that there's a, in many ways, this is, you know, it's not an entirely novel problem. But if you look at what happened in crypto for a long time, there were kind of two things that were really valuable.

And I think that there might be a new opportunity to figure out, to build a new company around securing these extremely, extremely valuable digital assets, which are the AI models. I think that there's a, in many ways, this is, you know, it's not an entirely novel problem. But if you look at what happened in crypto for a long time, there were kind of two things that were really valuable.

And I think that there might be a new opportunity to figure out, to build a new company around securing these extremely, extremely valuable digital assets, which are the AI models. I think that there's a, in many ways, this is, you know, it's not an entirely novel problem. But if you look at what happened in crypto for a long time, there were kind of two things that were really valuable.

There was like Bitcoin, which was kind of like sticking around forever. And then there's this company called Chainalysis. Did you ever see this? So Chainalysis was a company that they didn't have a token. It wasn't a crypto company. They did the analysis to find like the fraudsters and they were like super downstream of the actual like cryptocurrency.