AI model

“Because I do think models in general will become commoditized over time, in particular because of the open source growth and adoption.”

“Fundamentally, the harvesting of publicly available data has, for the most part, run its course, and Ilya Sutskiver said as much I love, you know, it's an interesting quote by saying, ah, AI data is fossil fuel for AI, and we've run out of it. And I generally …”

“They're using it, like, it's, like, it's commodity business, we all know that. Like, we've known that from the beginning, so I don't know why we go through these circular phases of, like, thinking it's gonna change the world, and it's, like, a power law thing,…”

“The race to HGI is on, and the data needs of these models have shifted. These models need incredibly complex data for training them to get better at coding, advanced reasoning, STEM, etc.”

“We are creating data for coding now, for agentic coding, that would stump all of today's models and agents built on top of those models.”

“I would give, give that maybe six months before the models climb that hill, and then we'll generate data that's even harder for the models.”

“In the AI era, there's really only one place where actual human-generated content exists, and the value of that content is super valuable, like, really valuable, because it helps train the models.”

“And I think that's the next big shift that we're going to have to see is that these models, which are doing wonders for, you know, consumers and for knowledge workers, they need to be put in the hands of the people who really need them when you have to make sp…”

“All explainable techniques for AI today are kind of not, are, they're empirical. So they're looking at the results, they're evaluating it. That, that's not fundamentally explaining why a model does what it does.”

“It's not like you have to do some fancy JSON thing. Like, that's what people like to popularize it online with, but I mean, you can just talk to them and ask them questions.”

“I actually think that there needs to be a kind of different posture towards trying to get people to break jailbreak models in order to battle test them, the same way that we do with, like, HackerOne. There are programs that are being run by the labs, of course…”

“In this era it's AI optimization. And so these products are now the distribution mechanism for all the downstream products and services that you can use. And they tend to prioritize the tools that have the best developer and user experience.”

“We're kind of getting to the point where you can solve basically any benchmark, right? Because what does it mean to have a benchmark? It means you've already defined the task. You've clarified what success or failure looks like. You've given a bunch of example…”

“It's really, probably, like, 75% of it is, like, the data that you're training on. Like, I would say for any AI model, it's like, there's always, like, you know, like, these, like, step functions and, like, algorithms and architectures and stuff, but the data …”

“You've seen that it can find cyber vulnerabilities. It can daisy chain vulnerabilities, attack infrastructure. So it's clear these models are going to be extremely powerful. They already are.”

“I think that's kind of mainstream because the capabilities of AI models have sort of, it's fair to say, have surpassed humans in certain cases from a coding perspective.”