“Flood The World With Intelligence” – Sam Altman On AI | DRM News | AI1F

Introduction and where AI stands today

00:00Hello again. If I keep doing this, I understand that there's vacancy at 60 Minutes. Maybe I can If this investing thing doesn't work out, maybe I can do that. Another full disclosure, Sam is a friend. I'm also on the board of OpenAI, so I promise you I will not only ask him softball questions. I will channel my inner Brett Bear and and uh and seek Rupert. So, Sam, as I said, CEO of OpenAI, one of the founders. He previously served as the president of Y Combinator, where he helped scale some of the world's most influential startup accelerators, and advised companies that will shape entire industries. At OpenAI, Sam has obviously been at the forefront of advancing artificial intelligence to the benefit of humanity,

00:55helping to bring cutting-edge AI tools to the hands of millions of people. How many people? Eight, 900 million people? >> More, yeah. >> And businesses around the world. So, Sam, let me start with what I think is the question that is on everybody's mind, okay? Which is where are we today in the world of AI? >> I I think at some point in the last few months, we really have crossed a threshold into major economic utility of these models. It may have happened a little bit earlier, but there was such an overhang before we figured out how to use these, and we had to not only continue to make the models get smarter, but figure out sort of the plumbing to make them easy to use. Where we're we're now in a world

01:43where the models are astounding people with the work they can do. Uh and I think this has been most noticeable in coding, yep, but it's also happening in science. It's happening in many fields of knowledge work. Uh at sort of with this orienting speed where people are saying like, "Man, these things that I thought were still years away are happening now and I have my job shifted from doing uh you know, direct technical work or uh you know, legal work to managing a team of agents doing this work." This is going to go much further. Uh I think we're at a very steep part of the curve and right now maybe you can trust say a AI software engineer to do a multi-hour task. Very soon it'll be a multi-day

02:32task and then a multi-week task. And not long after that, I think the paradigm will shift again and it'll feel like these AI systems are just connected to your life, to your company, whatever proactively thinking, working all the time. Uh and having full context on whatever they need to know and just sort of doing stuff like you would trust a senior employee to do. >> And and you think that um companies have real have a real understanding of how these systems can help them and reimagine how they do their businesses? >> Some do, some don't. Uh certainly the new generation of startups thinks differently than any generation of startups before. Uh it used to be that when we would talk to startups, they would talk about how many

AGI definitions and thresholds

03:20employees they needed. Uh now they generally don't want to hire a lot. They think that'll slow them down and they're all focused on how much compute they can get. You know, can I reserve this much capacity? Can I do a cloud deal for that? Uh can I get this many tokens? And I think that is that that is a mental shift that bigger companies are going through more slowly but some are starting to do that. Uh one place I think you can see this happening is engineering orgs and product orgs talking about they're doubling, tripling what they're planning to ship this year. And that has not happened before. >> Yeah. And you've been quite vocal in saying that um artificial general intelligence will come sooner rather than later.

04:07Want to share your views on how close we are and how soon it will come? >> At at this point, I think the definition of AGI really matters. Some people would say we already got there. Some people say it's very close. Some people say we're kind of you know, it's maybe still a year away, but in any case, that word has ceased to have much meaning. >> Mhm. >> Uh there are maybe two thresholds that we could talk about that are interesting. >> Okay. >> Number one, when is there going to be more of the world's cognitive capacity inside of data centers than outside of them? And that to me feels like maybe it could happen. Huge error bars. I could be totally wrong, but maybe that could happen by like late 2028.

04:50>> Mhm. >> And that's an extraordinary shift in the world. Um the other one is when can a CEO of major company, a president of a major country, a Nobel Prize winning scientist, when can they not do their job without making heavy use of AI? This doesn't mean that there will be an AI CEO or an AI president, uh but it does mean that the the role of let's say a human CEO, when I think about my job, it's really quite different. You you still do need a person to stand behind decisions and kind of exercise human judgment and all of the understanding that we expect out of someone running an an important organization to do, but the actual parts of my role that I will increasingly have to rely on an AI to do

05:45because no human can. You know, no human can talk to no human CEO can talk to every employee at a company, every customer, be in every meeting, be an expert in every field. And so, more and more I think of these jobs will be supervising a bunch of AI, um providing oversight, deciding how to like trust the the outputs, how to provide guidance. And that threshold of when you really wouldn't want to be doing your job, running a large organization without heavily heavy reliance on AI, I think that's another sort of interesting threshold. That may take a little bit longer, but probably not a lot longer. >> And and as you do your job, how much are you finding yourself relying on some of the agents and some of the

$110 billion funding round and abundance of intelligence

06:34artificial intelligence that we're developing at OpenAI? >> Uh it's ramping incredibly quickly. Um if I have like a new idea for a business model, a strategy shift, a uh a product offering we should do, the very first thing I do before I even bounce it off somebody else is to ask our tools. And as they get more context, and I think this really is like the next big thing to happen is they can get close to full context of our company, access to all of our internal docs, communication, code, customer data, everything, the quality of the answers, thought, whatever you want to call it, gets better and better. >> Right. Okay. So, let's let's shift um a little bit. 2 weeks ago, um you announced a $110 billion funding

07:26round. I asked ChatGPT, how does that compare compare to any other fundraising that's been done in the public markets? >> I actually don't know. >> Uh four times as large, okay? The largest public offering ever done was roughly 25 billion that Aramco did several years ago. And the public markets are supposed to be the broadest and deepest sources of capital, okay? Three strategic partners, Amazon, Nvidia, and oh and SoftBank. Tell us a little bit about this. How is this an inflection point of the company? And one of the questions somebody asked me is, what are we spending all that money on? The >> There is many hard parts of this business, but one of the hardest ones is the infrastructure is so expensive. You

08:15need so much of it, and you have to commit so far in advance. I have never seen any other industry quite like this. I mean, there have been clearly many capital intensive industries throughout history, but as I look at the what's to come in front of us in the future in future years, if the ramp stays as steep as it looks like it is right now, the demand is growing as fast as it's growing, you have to do some pretty unusual things. Uh OpenAI does a lot of things that look weird. We spend a ton of money on infrastructure in advance of revenue. We do new business models like ads that seem like, you know, maybe not the most profitable thing we could do. Um a long list of other things, but we have

09:06this fundamental belief in abundance of intelligence. And that one of the most important things in the future is that we make intelligence, you know, to borrow an old phrase from the energy industry that didn't quite work, too cheap to meter. We want to flood the world with intelligence. We want people to just use it for everything. We want this to just be something that the future generation doesn't think about. They expect everywhere, and everybody has access to like geniuses, as many as they need in any area that they need. And this principle, which is one of our kind of like top guiding principles, does lead to a lot of behavior that would look less natural for other companies. And one of those is

09:49we really want to get out of this world that we have been in that we still think we're on a trajectory to stay on without changing what we do of always being capacity constrained. >> Right. And capacity constrained you mean compute? Yeah. And I've often heard you say a lot that compute is revenues. You want to talk a little bit about how you think how you think about that? >> Fundamentally, our business and I think the business of every other model provider is going to look like selling tokens. You know, they may come from bigger or smaller models which makes them more or less expensive. They may use more or less reasoning which also makes them more or less expensive. They may be running all the time in the background

Compute as revenue and intelligence as a utility

10:30trying to help you out. Uh they may run only when you need them if you want to pay less. They may work super hard. You know, spend tens of millions, hundreds of millions, someday billions of dollars on a single problem. >> Right. >> That's really valuable. But we see a future where intelligence is a utility like electricity or water and people buy it from us um on a meter and use it for whatever they want to use it for. The demand that we see for that seems like it's going to continue to just go like this. And if we don't have enough, we either can't sell it or the price gets really high and it you know, it kind of goes to rich people or society makes a bunch of sort of central planning decisions that I

11:20think almost always go badly about, you know, we're going to use our limited compute supply for this and not that. So, the best thing to me throughout all the history of capitalism, innovation, whatever you want is to just flood the market. >> Yeah. And obviously a critical part of addressing that compute demand is Stargate. And this is an infrastructure conference and you know, you announced that several months ago, or almost a year ago, I guess it was, yeah. How's it going in the US? And then how is it going, because I also started in in Abu Dhabi. >> Yeah. Yeah. Um Honestly, there's many cool parts of the job. One of the coolest is getting to go visit these mega data centers under construction and operation.

Stargate and gigawatt data centers

12:07>> Yeah. >> Just the scale is, you know, of like these gigawatt campuses. It's it's it's really hard to explain and you see these photos and it's like, okay, it does look big. And then you go there and you're walking through it and building to building and it's just, you know, 10,000 people there, all these different skilled trades doing all these different things and you know, it looks like a spaceship inside. It's really quite incredible. Um we are training right now on the first site in Abilene, what I think will be the best model in the world, hopefully by a lot. And it is it's so amazing to have gone from, you know, many visits while it was under construction, and just really internalizing the scale and incredible

12:47complexity to the day that one researcher at OpenAI types in like one command and it pushes enter, unbelievable number of GPUs spin up and start doing this one huge computation all together. >> Yeah. It's very And and what have been the most pleasant surprises and what have been the hardest things about getting it up and running and scaling it going forward? >> Uh I mean, there have been all the expected challenges and then the unknown unknowns. Like there was a crazy weather event in Abilene that we hadn't, you know, was outside what we had planned for and that brought things down for a little while. >> Mhm. >> Um there's all of supply chain challenges. Anything at this scale is just like so much stuff goes wrong. Lots

13:32goes right, but the just try trying to build something around the clock with such complexity, there's all the stuff that goes wrong. One of the biggest surprises on the upside is how many different organizations had to come together to do this in an incredibly short period of time. Uh, and how much we all like ended up working together as one team uh, under a lot of pressure. >> And of course, the power demand part of it is the one that a lot of people are focused on. Um, are you optimistic that we will solve that challenge first in the US and then we can talk about in in other places? >> Uh, in the long term, I am. >> Okay. >> Uh, I have no doubt we will figure out how to build huge amounts of power

Power demand and efficiency gains

14:25generation. AI will help, of course. Uh, but I think the portfolio the world has in front of it gas, solar, nuclear fission, nuclear fusion, more like we're I feel good about what what we will be capable of and what we will eventually do. >> Mhm. >> Um, given the demand growth we are seeing, I am sort of hoping for a miracle in terms of figuring out how we can get way more efficient per watt >> Uh, okay. >> models to give us time to build out all of this infrastructure. Now, the track record there has been incredible. Um, people cite whatever amazing statistic they like about how much more efficient our models have gotten over our industries models have gotten over time, but one that I think is incredible,

15:14our first reasoning model was called 01, came out like 16 months ago. Uh, and latest model where we're now integrated reasoning is 5.4. To get the same answer to a hard problem from that first model to 5.4 has been a reduction in cost of about 1,000 x. Maybe I was a little bit wrong on the timeline. Maybe it was a little bit longer. But in any case, since 01 till now. >> Yeah. >> 1,000 x. Uh, that is unbelievable in a relatively short period of time. And the thing that that the two things that I think that points to one, we are still so early and in this paradigm and we're still so we have so much more to gain about our understanding of how to develop these models and train them and run them efficiently

16:09>> Mhm. >> that we have were then and still are doing things in dumb ways and will get better and better. And two, is that human ingenuity and the ability to operate in constraints and to find ways to solve problems almost always surprise you on the upside. So, it's not just that the models have gotten better, it's that we have figured out, you know, kernel engineers came to help figure out how to write more efficient kernels and like power engineers, the people that design data centers found more efficient ways to do that. So, people are answering the call well beyond just the model side >> Right. >> to make this more efficient. >> Right. And then of course, um, obviously OpenAI is a big customer of

OpenAI's inference chip

16:51the Nvidias and AMDs of the world, but um, you know, I think we signed a big contract with Amazon. But we're building our own chip. >> Yeah. >> What's what's the thinking behind that? >> Uh, so the chip we're doing is inference only. >> Okay. >> And the thinking behind it is that on this, you know, come rise up to solve problems in front of us, um, we think that a specialized chip to be not necessarily the fastest inference chip, but the cheapest inference chip, the most efficient per watt given the constraints we see in front of us, is going to be important for all the agent demand we see in the future. So, it's a it's an opinionated bet, it's a limited chip, but the thing that it does, in a world where we're energy

17:32constrained, I think it'll be very important. >> Maybe you can explain because I am not sure everybody in the audience is AI proficient with the difference between an inference chip and a training chip. >> Sorry, I should have done that first. So, there there are two main phases to uh AI workloads. Now, they're going to blend together eventually and it'll all be one continuous thing. But right now, first we train a model, you know, a gigantic number of GPUs crunch for weeks or months on a bunch of data and you can think of that like maybe it takes you 22 years in life to get your education and you learn a bunch of stuff starting when you're a baby and you drop things and see they fall and then eventually like in college physics

18:16understanding like at a very detailed level what's going on there. Um And then after that, if you ask a model to solve a physics question, it that's called inference uh and that's quite efficient. But this is true for people as well and when people talk about all these AI models are so efficient, they're usually comparing the 22 years that a human takes to train to the 1 second that an adult takes to solve a physics problem. If you compare the you know, the model solving the physics problem to the human solving physics problem, actually the models already probably are more energy efficient. Um but training is this hugely massive amount of processing power that then produces really just like a file of numbers that

19:01you can then pose a question to and get a response. >> And are you optimistic about the progress we're making on the chip? >> Yeah, um we should have the first chips deployed at scale uh by the end of this year. We should have the first chips back just in a few months now. Um and looks like it'll be really good. >> Fantastic. So, you announced a new partnership this morning, actually, with the North American Building Trades Unions to expand training pathways for skilled construction, which has been a subject of my many of the discussions today. Can you tell us a little bit about what you and uh Sean McGarvey >> Yes. >> agreed and What's he And he's going to be on stage later on. So, let's hear

Partnership with building trades unions

19:40your version of it and then see if his version is the same. >> So, we talked earlier and we've talked The world has talked many times in the past about the need for AI infrastructure, the need for the world to have the physical infrastructure, power plants, transmission lines, data center halls, chillers, obviously the racks and the GPUs and everything that goes inside of those. Um and I sort of wish everyone would go visit at some point one of these mega scale data centers because it it's hard, you know, when you ask ChatGPT a question, you get your answer back and it's really hard to visualize the scale of what it took to make that or what it takes to make that happen. Uh people talk about all of these different

20:34places that were limited. >> Mhm. >> We are limited in the, you know, number of turbines or now it's the voltage transformers or it's the memory fabs or it's building the data centers, whatever. All of these things have one thing in common, which is they are massively complex physical infrastructure that require skilled tradespeople uh and a lot of them to do. And no matter where the choke point in the supply chain is at any one time, when I talk to people about what it would take to accelerate it, um it is more skilled trades workers to build out all of this infrastructure that we all depend on. I think this will be incredible jobs. Uh more than that, I will I think they will lay the foundation for the next

21:20generation of American infrastructure and economic prosperity. And we are thrilled to get to work uh together to to drive that faster. >> Uh just so you know, we at BlackRock share that