Sam Altman

Altman said on the Cheeky Pint podcast that it is quite terrible for individuals to hold a large share of the world's limited DRAM supply in their personal computers.

It’s quite terrible to have everybody have a bunch of the world’s limited supply of DRAM sitting in their own computer.

Altman clarified on X that he wants people to have as much memory as they like in their computers, saying his earlier podcast remarks on the memory shortage were taken out of context.

I actually do want the world to have tons of memory and people to put as much as they want in their computers

Altman says people must always remain in control of the future and OpenAI rejects relegating humans to roles that AI systems decide to reward.

People should always be in control of the future. OpenAI is against the notion that it would be acceptable for people to be relegated to a role of doing things that ‘AI systems think makes sense to reward in humans.’

Altman argues the public should not cede control of the future to superintelligence, and that this should ultimately constrain AI progress.

should not cede control of the future to superintelligence, and that requirement should ultimately constrain AI progress

Altman told staff that if AI visibly goes badly wrong this year, biology is a plausible area for it to happen.

“If something goes visibly really wrong for AI this year, I think bio would be a reasonable bet.”

Sam Altman said he wants people to have plenty of RAM and called for building more capacity, amid severe DRAM shortages.

i want you to have so much ram lets build more

Altman says he wants the world to have abundant memory, that his earlier RAM comments were taken out of context, and that he has long argued for ramping up memory production.

i actually do want the world to have tons of memory and people to put as much as they want in their computers, and it is fucking annoying that this was taken out of context. i have been arguing for the world to ramp up memory production for a long time.

Altman named energy as the single biggest challenge facing OpenAI and the AI industry.

Asked to identify his biggest challenge, Altman answered with a single word: "Energy."

Altman says patching every vulnerability in the whole supply chain one by one may be too hard, so the industry needs new remediation approaches as AI speeds up vulnerability discovery.

It may be too difficult to patch every vulnerability across the entire supply chain one by one.

Original (zh)

要对整个供应链中的所有漏洞逐个打补丁，可能过于困难

Altman cautions that many people wrongly assume the challenges of local inference have already been solved.

A lot of people want local inference. And somehow, they assume that the problem is solved now.

Sam Altman said that as AI speeds up bug discovery, patching every vulnerability one by one across the supply chain may be too hard, so new ways of fixing them will be needed.

It’s maybe too hard to patch every bug one by one across the entire supply chain, I think we’ll need new ways to fix them.

Sam Altman argues that, given the DRAM shortage, it is inefficient for individuals to hold scarce memory in local machines used about 5% of the time, and that cloud data centers offer better utilization.

But from a global allocation perspective, actually, it’s quite terrible to have everybody have a bunch of the world’s limited supply of DRAM sitting in their own computer and being used five percent of the time. You want maximum utilization in the cloud if you’re trying to deal with this very supply constrained environment.

Altman argues that local AI inference wastes the world's scarce DRAM, since memory in personal computers sits idle most of the time compared with cloud data centers.

A lot of people want local inference. And somehow, they assume that the problem is solved now because we are going to have open-source models. We’re going to be able to run them on our computers. And so we’re not going to need all the stuff in the data center. But from a global allocation perspective, actually, it’s quite terrible to have everybody have a bunch of the world’s limited supply of DRAM sitting in their own computer. And be used 5% of the time.

Altman says that in a supply-constrained environment, cloud utilization should be maximized, contrary to many people's intuition.

You want maximum utilization in the cloud if you’re trying to deal with this very supply-constrained environment. And somehow, many people’s intuition is the opposite of that.

Altman says market forces are failing to resolve the memory shortage, with the solutions front going badly.

I think the invisible hand is failing right now. It has a checkered track record at best. But, I think it’s not quite working here.

Altman says memory makers are cautious about expanding capacity because they remember past busts from overbuilding fabs.

I think they all remember their history. How many times they almost went bankrupt by making too many fabs and being promised this time it was different

Altman argues that, amid DRAM scarcity, it is wasteful for individuals to hold scarce memory in personal computers used only about 5% of the time, and that cloud data centers offer far better utilization.

From a global allocation perspective, actually, it's quite terrible to have everybody have a bunch of the world's limited supply of DRAM sitting in their own computer, and being used maybe 5% of the time. Like, you want maximum utilization in the cloud if you're trying to deal with this very supply constrained environment. And somehow many people's intuition is the opposite of that.

Altman says MBA students keep chasing what was cool five years ago, and argues now is a great time to build something new.

My chronic disappointment with MBA students is that they always want to go do the thing that was cool five years ago rather than try something new. I think this is a great time to go do the new thing.

Altman rejects the idea that AI will cause a job apocalypse as many people claim.

I don’t think there’s going to be like a job apocalypse in the way that many people talk about.

Altman says huge numbers of tasks and wide swaths of current jobs will be done entirely differently and better by AI, even though he believes the future won't change as quickly as people think.

huge numbers of tasks that are done today by people will be done better, cheaper, faster. You know, completely differently by AI,” and that “wide swaths of current jobs will just — the way they’re done today, at least — will be entirely done by AI.

Altman says it's not a good time to clone OpenAI, but many companies built on OpenAI's technology could become bigger and more important than OpenAI.

I don’t think this is like a good time to start another copy of OpenAI. But I think there are many companies that can be built on top of technology that OpenAI has put out into the world that will be much, I hope, bigger and more important than OpenAI.

Altman says huge numbers of tasks people do today will be done better, cheaper and faster by AI, in completely different ways.

It's totally true that huge numbers of tasks that people do today will be done better, cheaper, and faster by AI - and completely differently,

Altman argues that young people should start their careers amid the greatest change, when the field is most open and high-potential people tend to thrive.

If you're heading into the world right now to work on something, I think you want to start your career at a time of the greatest change, when the playing field is most open." "I think people with high potential, drive, ideas, a new way of looking at things, and new skills tend to do relatively well when the ground is shaking the most.

Altman says many companies built on OpenAI's technology could become much bigger and more important than OpenAI itself.

There are many companies that can be built on top of the technology OpenAI has put out into the world that will be much bigger and, I hope, more important than OpenAI,

Altman says the world deserves to understand how the universe works and he sympathizes with mathematicians, but OpenAI will keep releasing math results.

I think, on net, the world deserves to understand how the universe works,” Altman said during an episode of the Cheeky Pint podcast released Thursday. “But I do have empathy for the method.

Altman argues humanity's right to fundamental knowledge outweighs the disruption, and OpenAI should not withhold major scientific discoveries.

I feel unequivocal on this point: Humanity deserves fundamental knowledge, and we should not be sitting on a major fundamental science discovery,” he said. “The world deserves that.

Altman confirms OpenAI holds more mathematical discoveries than the hundreds of proofs already released.

We do. Of course.

Altman says OpenAI's math push briefly got out of control, deliberately, to probe model limits, and cost hundreds of millions including about $10 million on Navier-Stokes.

Model Name-Hold My Beer,” where employees “were just throwing that at the hardest problems they could”

Sam Altman warns that AI systems are uncovering security vulnerabilities at a staggering pace.

AI is uncovering security vulnerabilities at a staggering pace

Altman says AI is increasingly helping discover new science and invent technology, with strong examples in maths and other sciences.

Many things here, but one of them that I can’t resist talking about is how much AI is now helping us discover new science, invent new technology, actually do this work of innovating and creating better things. “We have seen tremendous examples with maths problems. We’re seeing that in lots of science.

Altman says GPT-6 was a step-change in models' ability to discover knowledge and in people's ability to get substantial work done.

GPT-6 was one of those moments where I think the models took a step forward all at once, both in terms of their ability to discover new knowledge, to do very hard things, but also for people’s ability to work with the models and get substantial pieces of work done at a rate that they never could before.

Altman urges companies to prioritize understanding and mitigating AI-driven security threats above even Codex productivity.

If I could recommend only one thing to a company right now, even more than become super productive with Codex, it would be to really understand the security threat landscape from AI and mitigate against that.

Altman says vulnerabilities must be found and fixed, and new remediation methods are needed since one-by-one patching across the supply chain is too hard.

I think the only way through this is to go find all these issues and fix them. In fact, I think we’ll need new ways to fix them. It’s maybe too hard to patch every bug one by one across the entire supply chain that we’re all dependent on.

Altman says AI has moved from chatbots to coding agents to converting compute into innovation, promising better, cheaper products and services.

We have seen in a small number of years AI move from a chatbot to coding agents to the ability to directly convert AI into innovation, like AI compute into innovation. Whether that’s make a chip or cure disease, it means we’re all going to get much better stuff, higher-quality services, lower-cost services.

Altman urges companies to prioritize understanding and mitigating AI-driven security threats over productivity gains in the short term.

If I could recommend only one thing to a company right now to do with AI in the very short term, even more than become super productive with Codex (OpenAI's AI coding tool), it would be to really understand the security threat landscape from AI and mitigate against that

Altman says the only way forward is to find and fix all the flaws, requiring new methods and services that continuously monitor traffic, since one-by-one patching across the supply chain is too hard.

too hard to patch every bug one by one across the entire supply chain that we all depend on.

Altman is shocked by the number of vulnerabilities AI systems are finding and says getting ahead of the problem will take a herculean effort.

I find myself shocked at just how many vulnerabilities these AI systems are finding

Altman says AI progress has been among the year's biggest surprises, with models going from failing grade-school math to tackling the hardest open problems, and GPT-6 seeing the strongest adoption yet.

couldn't solve grade school math, and now they're like knocking down the hardest, unproven math problems we know of,

Altman said GPT-6 saw OpenAI's strongest adoption of any new model, citing coding, enterprise software, security and science.

At the keynote, Altman said GPT-6 drew OpenAI's strongest adoption of any new model. He cited coding, enterprise software, security, and science.

Altman says GPT-6 was a step-change in models' ability to discover new knowledge and in people's ability to get substantial work done, with the strongest adoption of any OpenAI model.

GPT6 uh was one of those moments um where I think the models took a step forward all at once both in terms of their ability to discover new knowledge to do very hard things but also for people's ability to work with the models and get you know substantial pieces of work done at a rate that they never could before.

Altman says the single most important short-term step for companies is to understand and mitigate AI-driven security threats, even above adopting coding tools.

If I could recommend like only one thing to a company right now to do with AI um in the very short term, it even more than like become super productive with codecs, it would be to really understand the security threat landscape from um from AI and and and mitigate against that.

Altman argues that patching bugs one by one across the supply chain is infeasible and new continuously monitoring security services will be needed as AI finds vast numbers of vulnerabilities.

I think the only way through this is to go find all these issues and fix them. In fact, I think we'll need new ways to fix them. It it's maybe too hard to patch every bug one by one across the entire supply chain that we're all dependent on. Um and I think we'll have to like look at new kinds of services that are constantly monitoring traffic and not letting these things be exploited.

Altman says AI is excellent at chip design and that OpenAI built the Jalapeno inference chip, deployed on OCI, far faster than possible without AI.

AI is really good at chip design. In fact, we made a chip called Jalapeno that we're deploying in OCI and we're thrilled to be doing um and we did this chip which I think is the best AI chip in the world for inference um faster than would have ever been imaginable without AI to do a lot of this work.

Altman says AI has progressed from chatbots to coding agents to converting compute directly into innovation, promising better, cheaper services.

we have seen in a small number of years AI move from a chatbot to coding agents to the ability to you know directly convert AI into innovation like AI compute into innovation and whether that's make a chip or cure a disease it means we're all going to get much better stuff higher quality services lower cost services

Sam Altman warned that no single company can mitigate the risks of advanced AI, such as novel bio threats and rapid economic change, and that a society-wide response is needed.

No company can sufficiently mitigate these on their own; we will need a society-wide response to things like novel bio threats, a massive and fast change to the economy, extremely capable models causing complex emergent effects across society, and more

Altman argues AI makes patching every bug unfeasible, so defense must shift to using the same AI technology as a new kind of shield.

I no longer think we’re going to be able to just patch all the bugs. I think we’re going to have to completely switch the defense model to sort of using the same technology for sort of a new kind of defense shield.

Altman calls AI-driven security risk a huge issue and is surprised at how little attention it receives.

I think it’s a huge safety and security issue, and I am somewhat astonished at how relatively little attention it’s getting.

Altman says the industry will need to rethink how it paces AI progress going forward.

We’re going to have to make some changes to how we think about pacing progress from here.

Altman said ChatGPT should reflect a weighted average of humanity's moral views.

“What I think ChatGPT should do is reflect that weighted average, or whatever, of humanity’s moral view,”

Altman says he no longer believes software vulnerabilities can all be fixed by patching, reflecting the dual-use challenge of AI in cybersecurity.

“I no longer think we’re going to be able to just patch all the bugs.”