AI model

“This entire generation of AI models, from my perspective, has basically all been illegally trained, right? It's like, there's all sorts of rumors that are flying on the tech community about how a lot of the kind of more image-based you know, large language, la…”

“I tend to agree with you, Sam. Like, I think if, or, you know, the best models are going to be months ahead of the next best models. I mean, they may, that gap may already be almost that slow now.”

“I mean, it's gonna be tens of billions of dollars to train these models going forward, and that's not, you know, a research lab at Stanford kind of stuff.”

“Yes to crystals, yes to all the fun stuff you can do there, and then, like, you're gonna have these wars over who makes better show notes for podcasts and like, the marginal value that goes to zero, because they'll all be kind of fine.”

“The future is going to be almost like, this is what happened after the printing press is the news got super scummy and shitty, and all the elites retreated to super high price Italian private newsletters to know what was actually going on in the world. That's …”

“I'm an asymptote guy, and I think we're asymptoting faster than people want to admit in this generation of technology.”

“We've also seen that the models haven't been getting the efficiency gains and the training gains, they thought.”

“The answer has to be yes.”

“In fact, there's a bunch of science people are, I think there's, I'll have to go find it, but there's been a few papers produced about the sort of cognitive overload that these models are creating because they create so much content so quickly and sort of lead…”

“A harness is the software built on top of a model to make the model useful. The models are not useful by themselves. They need to be given software that helps them think in certain patterns, repeatedly run loops to accomplish tasks”

“Your hands can only type so fast and it's just more context per second, right? That you can input into the computer. And I think that's the big change here is that the models want more context to do a better job for you. And so we are shifting our way that we'…”

“The future is Multi-model. You mix them together. You think about routing to the right model in the right circumstance. Like, there's no one god model. Like, that's clearly where we're going, right?”

“That the only thing holding you back now is attention and, like, time and creativity. Like, you don't need, like, it's already so powerful that it doesn't really matter.”

“Like, I honestly don't care, which means these companies are all fucked.”

“There's no hardware that you can't hack with one of these things.”

“When you get down to a lower level, because you can no longer have on-off APIs, because everything is scriptable, because these things can script any interface and interact anyway, as long as you have data sovereignty, meaning like it's your data, they're all …”

“This does great damage to the discourse and to society, anthropomorphizing these things.”