Memory

“And you train these agents the same way you train a new hire. You start with small tasks, you watch at work, you correct the mistakes, You add the correction to memory, because now we have memory, and then you expand the scope as you increase your comfort leve…”

“I think they're gonna make everything else more expensive. I think that iPhones will cut their memory, iPhones are gonna go up in price, and we're just gonna deal with it.”

“The way you choose to record or inject memory lives in the executor, but the actual memories themselves go down in the exo harness as artifacts for that conversation or for that agent. And so it's kind of like the, how you do things is defined in the executor.…”

“I think in the short term it'll probably keep looking like the way it is, but it seems inconceivable that it'll stay that form factor. Like, it's pretty clearly going to be voice. I think there's a lot around memory.”

“Memory is, the more memory you put, With flop for a given unit of compute, the more tokens you get out. It's the single most important thing you could do to increase kind of token output per unit of compute.”

“Today we're all talking about compute, compute, compute. I think the primacy of memory will become even more important.”

“Because, you know, the price of memory isn't getting cheaper. And the price of tokens aren't getting cheaper.”

“Everybody know right now, memory is the biggest shortage. And everybody tried to scramble for memory. And then even though you want to build a fab to capacity increase, it would take a couple of years to do that.”

“So we think ultimately that in this era, the amount of memory per user could quintuple just based on the demand that these AI systems are requiring to provide their services.”

“And now, you know, in the case of memory, for example, you're now hearing Supply agreements, like, guaranteed commitments through 29, twenty-thirty.”

“And so I think that that is just creating a massive need for more CPUs and more memory. And we're seeing that we're actually just at the beginning of that. I think agents is really what is driving that.”

“I mean, it's, I've never seen so many bottlenecks at the same time, and I think that is what makes it so hard, because if it's just one vertical slice that's the bottleneck, you can say, okay, well, if they put enough dollars in, they'll be able to solve the b…”

“Actually this, it's going to be energy and memory. Energy and memory is where the money is made.”

“Even though we're working on machines, not biological brains, and so potentially you could have, you know, millions or tens of millions size context window or memory, and it can be perfect. There's still a cost to looking it up and finding the right thing that…”

“I think personalization and memory is probably like the single most important job of the OS.”

“Well, it's not only anxiety, Jess, but like there's actual now data coming out that people have been measuring people's, not only their memory, but their problem solving, like their constructive thinking and people, the more people are using all these AI tools…”

“Every time you think of a memory, you are actually revising it in your brain.”

“I guess evals and prompt optimization are pretty closely tied, but like evals and memory are actually not at all tied.”

“The way we see the future is you will separate memory from intelligence. So the memory will be captured by infinite layers and the intelligence will be captured by finite layers of, you know, intelligence and that small model that has those finite layers that …”

“There's a lot of consumer hardware right now where people are trying to buy memory, and it's a huge problem.”

“I think memory is turning into an actual moat for like that consumer experience.”

“Memory is a feature and a stickiness, not network effect.”

“AI is definitely gonna be able to do that. And we actually talk a lot about this. Like right now, memory is still very crude, very early. We're in like the, you know, the GPT-II era of memory, but what it's gonna be like when it really does remember every deta…”

“We're in like the, you know, the GPT-II era of memory”

“So when we talk about how time flies, it doesn't actually fly. That's just nonsense. What happens is it just gets forgotten. So we look at a full year, 365 days, and we reduce it to 50 days or 50 moments or 20 moments. That's why time feels like it flies.”

“I would say that memory is definitely like a feature that has been under, under-invested in by the field. But I would say that it is kind of difficult to invest in memory in this very, like, task-centric regime. Because if you have, like, A bunch of these, lik…”

“I think it's the most defensible part of an LLM.”

“The real crazy unlocks come from the ability to take that memory that builds over time and intuit it forward.”

“I think with the stateful agents thing in particular, I think one interesting thing about scaling sleep time compute is you need memory for it. You just fundamentally cannot do this sort of scaling without memory. Memory is like, you know, kind of table stakes…”

“There's a whole lot of memory sitting, sitting in warehouses and in China that they'll be good through for at least the next two years.”

“Like I think in GPT-V, it's not going to be the final state, but I think you're going to see the beginnings of memory and the beginnings of actions. Right. And this is, you know, months away and you and I have a bet on this.”

“And I think next year we're also going to start losing our memory. And I think we're going to lose our memory more because AI is going to start handling so much stuff for us and remembering stuff for us.”

“Some of the things we want to work on the most are like memory cause you know, our users definitely want their virtual friends to remember them.”

“The amount of memory that would hold a song Haas. You know, we were back in the days of tape. It cost about a million dollars, a good fraction of a million dollars.”

“The best games create strong emotions, because emotions are the foundation of our memory.”

“The fact of the matter is, is that when we experience in any kind of pain or travail, your brain immediately Gets to work on softening that memory. Literally right after it happened. And that's a really good reason, because if we actually remembered pain to it…”

“The cost is decreasing very fast. It's about every eight months, the cost decreased by 50%.”

“Now Spark, if it was created six, five, six years before its time would have completely failed because memory was so much more expensive.”

“When people talk about how the internet and Google is ruining our brains or our memories I am actually much more positively inclined towards those technologies simply because it, if we don't necessarily have as good memories anymore, It means that we now actua…”

“Narrative is like one of the fundamental principles of memory. You know, it turns something into a story. It's five times more, more memorable.”

“Just putting data from disk into memory, you'll see, you know, on the order of a 50 x improvement.”