- 3 mesi fa
Categoria
📺
TVTrascrizione
00:00I am joined by Sam Altman, the CEO of OpenAI.
00:02We're both looking around.
00:04You kind of never get over the size and scale of it, Sam, even if you've seen it.
00:08It doesn't, you can't read the numbers and understand what you're kind of looking at.
00:13Yeah, every time I come to one of these sites, I'm struck again.
00:16And you're totally right.
00:17The numbers, say one gigawatt, this, this many jobs or tens of billions of dollars of capital,
00:21it doesn't really get across the feel of watching something like this materialize.
00:24Yeah, I mean, that said, the numbers are pretty staggering too.
00:27$16 billion just to develop the site, build the buildings.
00:32But then in talking to Clay McGurk from Oracle, at least another $30 billion going into it.
00:37And so I guess, you know, when you talk about something like this,
00:40$45 billion, maybe $50 billion, OpenAI being the main customer,
00:45do you feel like that $50 billion will generate the necessary return for both OpenAI
00:51and for all the people financing this deal?
00:53At this point, we're very confident.
00:55So one gigawatt is a huge amount of compute, but we understand what demand looks like
01:00and how much people want to use these models and the degree to which revenue is ramping
01:04at our company and the industry.
01:06And more than that, the degree that value is now ramping, it's not just,
01:10let me try this, but people are really saying, you know, the future of my company,
01:14the future of my scientific research program, it is going to depend on this.
01:17Can you promise me compute long into the future?
01:19Well, can you give me some examples?
01:21I mean, because that conversation has shifted even recently.
01:24Sort of something seemed to have happened.
01:25Maybe it was the end of last year.
01:26I've heard you discuss this as well in terms of just an uptick.
01:29Maybe it was the power of the coding models.
01:32I think that was the single biggest driver of what people realized was the coding models
01:39have so transformed how companies are doing their work and the efficiency
01:43and the speed with which they're able to build products.
01:46The coding models got really good late last year, early this year,
01:50and then another step forward in recent months.
01:52So I think you're right that that's the single biggest driver.
01:55But we are now seeing scientists really use these models
01:58and a much broader application of knowledge work beyond coding.
02:03But I think it's fair to say that coding is the magic right now.
02:06It was a key thing.
02:06I want to come back to that.
02:07But back to that, let's call it $45, $50 billion.
02:10I mean, the price of tokens continues to come down, correct?
02:14Yes, but not as fast as the desire to use those tokens at the lower price goes up.
02:18And so that is what gives you part of the confidence?
02:20Yeah.
02:20That, in fact, I mean, again, the numbers are staggering when you think about it.
02:25This is just one of so many projects, and yet this will consume $45 billion in capital.
02:31We still don't think the world has appreciated how much AI every person in every business is going to want.
02:39You know, right now, you still send a request to an AI, and it does something for you and gives
02:44you an answer back.
02:45But very soon, an AI will just be running for you in the background all the time,
02:48helping you with your job, looking at all your information, aware of all your context,
02:51doing as much as it possibly can to help you out.
02:55And I can already tell in that world, I and everybody else is going to want so much more of
03:00this AI infrastructure than we think.
03:02Yeah, I mean, I was listening to you on a recent podcast on the same theme.
03:05You said, you know, the models are still quite dumb.
03:08I'm quoting you.
03:09But not far from a model that knows you well in the world.
03:13Is that what you're talking about when you say sort of the usage patterns?
03:15Absolutely.
03:16So what does that look like?
03:21Well, for my own job, I could imagine a model.
03:24I can't keep up with all the information inside of OpenAI.
03:26I can't read every document someone writes.
03:28I can't read every Slack thread.
03:30I don't have all of the context.
03:32And so I miss interesting connections or new ideas that could make things better.
03:35and there's no way I'm going to be able to do that.
03:38But a sufficiently smart AI that's running all the time, that understands my goals,
03:44that understands all the information, OpenAI, everything our customers are saying they want,
03:47that could give me really great advice and say, hey, here is the thing to do
03:51in a way that I couldn't, nor could any other person do on their own.
03:55So is that one of the reasons why you prioritize compute?
03:59Yeah.
04:01We have seen throughout, from our founding, that the more compute we can provide,
04:06the lower cost of services we can deliver, the smarter models we can make,
04:11this whole stack integrated together, you know, from, if you really think about it,
04:16in some sense, we are transforming electricity into this useful tool for people.
04:22and the better we can do that, the more we can make that smart, cheap, abundant, helpful,
04:28have all your context, the more people want to use it.
04:30And we really do build that entire stack.
04:33And that is, I think, a special thing about us,
04:36but that is what we want to deliver to the world.
04:38It's like, you know, prosperity through abundance of AI,
04:42a lot of it that people can integrate into their lives and work.
04:45On the subject of prosperity, I think you've also said, you know,
04:48there are different futures and one is a floor of up to tenfold of what we have right now,
04:54sort of this abundance, but still with significant inequality
04:57and sort of different variations of that.
04:59Yeah.
04:59You still believe that's kind of what we're headed towards?
05:02I think that's just one possibility.
05:04I think that it seems clear that AI is going to massively increase prosperity.
05:11but getting the questions of equality, fairness, distribution right,
05:17that's going to take work from all of society together.
05:20Yeah.
05:20So I no longer have much concern that AI can deliver on the magic potential
05:24or the, you know, wonderful, all these things we talk about.
05:27How society integrates this and makes sure that it really benefits everyone,
05:30I think that will be the big question in the next few years.
05:33But you, I mean, you were most recently also quoted on another key area,
05:36which is, of course, what it's going to do to jobs.
05:38You'd been fairly sobering in the past when you discussed that.
05:42You seem to have sort of said, maybe I was wrong in a more positive way.
05:46Yeah.
05:46And I'm curious as to what you've seen that's led you to that conclusion.
05:50So, first of all, I should say I'm not sure.
05:52Like, I still...
05:53Well, none of us know the answers here, right?
05:55But a positive update for me has been watching how companies have adopted Codex and other coding tools.
06:02The companies that I know that have adopted AI the most are also the ones hiring the most.
06:06And the companies, as a general rule, that are talking about doing layoffs because of AI are the ones adopting
06:12AI the least.
06:13But, you know, it's a convenient way to explain it.
06:16I think I underestimated how jagged these models are going to be.
06:22They do some things incredibly well, but they don't do kind of the long-term complex task supervision well at
06:29all.
06:29And so, watching people who are really good at using these models, they can do an amazing amount of work,
06:35create way more economic value than people without the models could or certainly the models could on their own.
06:40And I think that's going to go on for much longer.
06:43Then there are all the other things about, you know, people really like other people and want to interact with
06:47other people.
06:48They want to collaborate at work with other people.
06:50When they buy a product, they want to talk to a person at the company.
06:53You know, most people, I think, don't want to watch an AI-generated creator.
06:57They want to know about the person behind it.
06:59So, I think we have under—and this is really good.
07:01I'm really happy about this.
07:02But I think our industry underestimated how much we're going to be able to keep people at the center of
07:08everything in an economy that is—in a world that is based on people.
07:12But to your point, we obviously don't know the answer.
07:14I mean, when you introduced, I think it was 5.2, you said it outperforms professionals across 44 occupations.
07:20You can understand why there may be an AI backlash when people hear things like that.
07:24Totally.
07:25What I wish we had said then is that it outperforms professionals at small tasks in 44 occupations, which is,
07:32I think, a more accurate thing.
07:34And it is the people that are using these that are now seeing, you know, incredible productivity growth, wage growth,
07:39all of the benefits from this.
07:41But I think people are right to be anxious, and I understand it.
07:46You know, this is like a—this is not even a technological shift that happens every generation.
07:51This is one of the big ones.
07:53If not one of—yeah, maybe the biggest.
07:55Yeah.
07:55And so, it would be imprudent not to have some real caution around that.
07:59Well, on that—you know, on the AI backlash, and I have been speaking to a number of the leaders in
08:02the industry who have been saying maybe we haven't done enough to articulate the benefits, which may be difficult to
08:08do.
08:08But there's not just opposition to data centers like this.
08:11There is a more significant, perhaps, opposition to what it's going to do to society.
08:16How do you feel as the leader, one of certainly the key leaders in this, in terms of your ability
08:22to combat that backlash?
08:26Yeah.
08:26So, it's a huge challenge.
08:29And again, as I said, I think this is like—there's something good about this.
08:32Like, society should have antibodies against too rapid of change.
08:36And there should—like, part of the reason that we believe in this strategy of iterative deployment is we want society
08:42to see the technology.
08:43We want society to really understand what's happening and have a chance to debate, react, sort of say, hey, this
08:51doesn't make sense or this isn't going to work for me.
08:53Like, this has got to be—I have no interest in, like, building a, you know, super smart AI that accomplishes
09:00some non-human goals.
09:01Like, this has got to be about something that is working for people and that people are at the center
09:05of it and human values are what we drive forward.
09:09So, people should react.
09:11People should say, hey, this is what I want and not this.
09:13I don't think it's about not explaining the benefits.
09:17Because, you know, we say, hey, AI is going to cure a bunch of diseases.
09:20And people say, okay, that's great.
09:21But, like, that's not really my question.
09:23No.
09:24My question is, you know, what is my role in the future?
09:26What is my economic future?
09:27What is my agency?
09:28Like, how do I know that my kids, my family will still be able to have a fulfilling, you know,
09:37creative expression, struggle to drive the world forward, to grow, to kind of do this thing together in a way
09:43that has worked for a long time?
09:45And when you have people in AI say, well, yeah, sure, there's going to be no jobs or 50%
09:51of jobs are going to go away or 90% of jobs are going to go away.
09:53And, you know, AI is kind of going to be smarter than you at everything.
09:56And, you know, we'll give you some basic income.
09:59But there's, like, you're not really going to have a role.
10:03That's horrible.
10:04And by the way, you know, this AI company, maybe we're going to destroy all the jobs.
10:08We'll be the most valuable company in the world.
10:09People just look at you and you're like.
10:11Yeah.
10:12So, I think it's a terrible message.
10:13And I don't think it's that we haven't articulated the upsides.
10:16I think people actually believe us.
10:17Like, you know what?
10:18Go cure cancer.
10:18That sounds great.
10:19I think we have failed to articulate as an industry how people stay in control of determining the future at
10:27every step and have a really meaningful life in all the ways we care about.
10:32Another part of it may be the pace of change itself.
10:34I mean, you're releasing a major model, what is it, every six weeks or something like that.
10:38It's very hard.
10:39Things seem to be changing so quickly.
10:42You know, I maybe, maybe you're right.
10:46I don't think that's, I think at this point people believe us that the models are getting smarter.
10:51And, you know, there was a time when the first iPhones came out that every iPhone release was a huge
10:55deal and people lined up overnight and got very excited.
10:57And now I couldn't even tell you the number of the latest iPhone.
10:59It's great.
11:00It's amazing.
11:01It's my favorite piece of technology.
11:03But I expect it to continue to get better.
11:04And I know they make a new one every year.
11:08I kind of think the same thing for this technology, which is people expect the models to keep getting better.
11:15What they really want to know is, like, what's going to happen with society.
11:19Yeah.
11:19And none of us, I mean, you don't really have the answer.
11:23You can guess.
11:27Of course I don't have the answer entirely.
11:29But what I can say is our whole effort, our whole company is about giving this new kind of infrastructure
11:39at massive scale to people and trusting that the democratization of power, of wealth, of opportunity, of agency will continue
11:48to do this incredible story of civilization going forward.
11:53I'm sorry.
11:54No problem.
11:54We're trying to figure out how much time you have left with us.
11:57But we're in this race with China.
11:58At least that's the way it's described.
12:00And that is one reason why we're just full speed ahead.
12:04Don't stop.
12:04Build as many data centers as you can.
12:06Move as quickly as you can.
12:07I think in some ways that's OK.
12:09And in some ways that's really dangerous.
12:11Like, I think it's fine to say, hey, we're going to win this.
12:16We're going to have, you know, most of the economic benefit.
12:18There will be some global scale safety issues.
12:23And we have had time in the past, times in the past, like with the IAEA for atomic energy and
12:29weapons, where the world comes together and says, you know, none of us should be taking global risk.
12:33Different countries, different systems, they can sort of say, you know, I'm going to treat economics this way.
12:38You're going to treat it that way.
12:39I'm going to think about using AI in health care this way.
12:41You're going to think about using it that way.
12:42But on the really big things, making sure we don't ever lose control of AI systems, cybersecurity, biosecurity.
12:49I think we need to not treat this as a race and treat this as a, like, a good future
12:56of the world isn't everyone's interest.
12:57That's not where we are right now, though.
12:59Well, I think we're still in the economic.
13:00I don't think we're, I think we're transitioning to a world where we have to think about that.
13:04I think we're still in a, this is mostly an economic story.
13:06But as the risks, potential risks have increased in the last few months or the last year, I have been
13:13very heartwarmed by, you know, a new recognition among leaders of companies and governments that, hey, there's a, there's a
13:22category here we have to treat differently.
13:24You know, I read the news about Trump's visit to China recently, and I know they talked about this.
13:28So I think people are taking this seriously.
13:31Yeah.
13:32I mean, when we talk about a race, Sam, you're also in a race with your competitors as well.
13:37One of them's Anthropik.
13:38I'm sure you heard today they filed.
13:40It used to be confidential.
13:41Apparently, it's not anymore.
13:42I just heard.
13:43For their, you know, I'm curious as to what goes through your head.
13:45Is there a race to be the first to come public?
13:49I don't think, no, not for that.
13:52I think there is a race to deliver the best technology and build the best business.
13:57But, you know, going public is a financing event, and I don't think that's one that we're focused on the
14:02timing of.
14:03We'll do it when we think it makes sense.
14:04But you will do it as well, I assume.
14:05We'll do it someday, yeah.
14:07At some point.
14:08And when it comes to sort of that competition, I mean, again, back to sort of where we started with
14:12how much money is being spent.
14:14Are you confident that all, you know, that you, that it's not a winner take all?
14:20I'm confident it will not be, yeah.
14:21I think the world will demand, this is going to be such critical infrastructure for so many things, that the
14:27world will rightly demand robustness in the system with multiple providers.
14:31I think that's a very good thing.
14:33And you think that compute, you know, the last week I was hearing about compute, for example, companies starting to
14:38wonder, well, what are we spending it on?
14:39Our bills are going through the roof, and it's not clear to us exactly what predict, you know, in other
14:44words, I know a lot of my spend is going well, but I don't know which part of it.
14:48So I think this is the most fair criticism right now of AI, which is you hear companies saying, I
14:54am spending a ton of money on AI.
14:58And I know some great stuff is happening, but I know there's a ton of waste.
15:02And, you know, how long do I have to wait for it to really show up in revenue, and how
15:06long do I have to wait to really get the costs under control?
15:08And I assume that the industry will figure that out pretty quickly, but I think that is a fair issue.
15:13You do?
15:13Yeah.
15:13Quickly being?
15:15I would bet that by another year or two from now, there is a much better rationalization of companies' spend
15:23relative to outcomes.
15:24And finally, Sam, are we ever going to see things like this up in space?
15:30Ever?
15:31I hope so.
15:32In the short term, I think it's probably better to do it.
15:35You know, there's like, as you can see, this is a huge project.
15:41It'd be a lot to put up there.
15:42Huge project.
15:42Yeah.
15:42Putting this in orbit at current launch costs, or even at lower launch costs, feels difficult.
15:47Also, you know, we know how to cool it here.
15:49We know how to get people to service it here.
15:51We have an atmosphere protecting us from some radiation.
15:59I hope that humanity expands to the stars someday and data centers along with it, but we're going to focus
16:03on building on Earth for now.
16:04It doesn't sound like you think it's happening anytime soon.
16:06It's not a short-term priority for us.
16:09Sam, thank you for your time.
16:10Thank you.
16:11I look forward to further conversations, I hope.
16:13Enjoy.
16:13We'll be right back.
16:13I hope that we will help you to get into this.
16:13We'll be right back.
16:13You've got it.
Commenti