LLMs

“These are also not deterministic models, and they're not deterministic outputs, meaning that it's not a discrete And specific answer that's going to be repeated every time the model is run. These are statistical models. So they infer what the right answer coul…”

“If you limit How they can tokenize to just being all entire sentences. The product will not be that good. Like the whole idea of these LLMs is that you're running, you know, so many iterations to literally figure out what is the next most best word that comes …”

“Over time, the ability to run LLMs and the ability to scan, to scrape data, to generate a novel, you know alternative to the ones that you guys are describing here is gonna emerge faster than we realize.”

“I think that the point of view is just that this is going to be a near impossible task to try and track and approve LLMS and audit servers that are running LLMS and audit apps and audit what's behind the tools that everyday people are using.”

“Almost all the efforts I have seen are very limited. They're all trying to scale LLMs. And, you know, I spend my time looking at what besides LLM will play An important role. I haven't seen one effort among the majors that isn't following the same model.”

“LLMs that can run on small desktop machines that are open source, that outperform all models that were in existence, even a few months prior.”

“I think the New York Times will make more money from licenses from LLMs than they will make from advertising or subscriptions eventually.”

“And they were able to achieve planning accuracy of up to 94% on some standardized benchmarks that are used for chain of thought reasoning and planning Using LLMs. This is a 66% absolute improvement over baseline models.”

“I do do it myself. I also do it just to see what the different models are producing about our drugs.”

“And so the amount of differentiation that's possible in video AI is I would say this point today so much wider than the differentiation and text based AI like what we see with LLMs.”

“With the advancements of LLMs, it actually allows this interesting period of time where you can now communicate with a machine in a way you couldn't really before, and so there's this probably middle ground that happens where there's pilots talking to machines…”

“You're minting new adversaries because you don't have to have all the knowledge that you had to have in the past. You can ask any number of LLMs and you can get answers back. So what's happened is the attack timeline has been compressed. You can automate all o…”

“A year ago people talked about the commoditization of the LLMs. And frankly, it's gone the opposite because as you start building an agentic layer, and we've all started to use this word harness, but the harness is what brings the context, the memory, right?”

“I like to say, like, all the cool stuff that we're doing on the, with LLMs now, is really based on just the text of the internet being absorbed into these models, which is incredibly powerful already, but they don't know shit about the real world. I call them …”

“What these LLMs actually do really well with is taking a massive stream of consciousness where you just keep talking and talking and talking, so I'll give it a one to two minute prompt, then I let go, and it has changed everything.”