- 2 days ago
OpenAI CEO Sam Altman explains how AI is evolving, why regulation and safety matter, and what superintelligence could mean for the future of work and discovery.
Category
đč
FunTranscript
00:00Sam, you are a pioneer of artificial intelligence. You are a genius.
00:05What kind of essence will artificial intelligence never be able to use?
00:30We still have to figure out what to do, what other people want, what other people will find useful.
00:34And I think that will be quite important.
00:37Die kĂŒnstliche Intelligenz entwickelt sich ja rasant.
00:40Wann genau, glaubst du, wird es eine super Intelligenz geben, die in allen Eigenschaften schlauer ist als der Mensch?
00:48I think in many ways GPT-5 is already smarter than me at least.
00:53You know, I think a lot of other people too.
00:56GPT-5 is capable of doing incredible things that many people would struggle with or find very impressive.
01:01But it's also not able to do a lot of things that humans could do easily.
01:05And I think this will be the course of things for a while, where we will see that AI systems can do some things incredibly well,
01:12struggle with some others, and humans use these tools and bring their sort of human insight, creativity, ingenuity to bear in ways that are really important.
01:22I expect, though, the trajectory of the capability progress of AI to remain extremely steep.
01:29So, you know, we've seen just in the two years or three years since ChatGPT has launched, it's almost been three, how much more capable the models have gotten.
01:37And I see no sign of that slowing down.
01:40I think in another couple of years it will become very plausible for AI to make, for example, scientific discoveries that humans cannot make on their own.
01:49And I think, to me, that'll start to feel like something we could properly call superintelligence.
01:54And, and, and hast du, hast du eine, hast du eine genaue Vorhersage, in welchem Jahr du dieses Superintelligence erwartest?
02:01One of the things that I have learned continuously is although we can say the ramp will be very steep, very precise, you know, it'll happen this month of this year are difficult.
02:11But I would certainly say by the end of this decade, so, you know, by 2030, if we don't have models that are extraordinarily capable and do things that we ourselves cannot do, I'd be very surprised.
02:23And also if in 2026 we don't see a similar rate of progress we've seen in 2024 and 2025, I'd also be surprised.
02:30And that means that by the end of 2026, I would expect models that, if we had them today, would be quite surprising.
02:38Viele Experten glauben ja, dass ganze Berufsbilder verschwinden werden, vom Buchhalter bis zum Bankberater.
02:45Wie viele Prozent der Jobs, die es heute gibt, werden aus deiner Sicht in absehbarer Zeit einfach verschwinden?
02:52Well, in 30 Jahren I would expect a lot of change, but in 30 Jahren jobs change all the time.
02:56I mean, if you think about the jobs that we did 30 years ago that may not exist at all today or new jobs that were kind of difficult to imagine 30 years ago that are now commonplace today.
03:05I remember reading a statistic once that about every 75 years, half the jobs in society change over.
03:11That's even without AI.
03:13It may happen.
03:13I expect it will happen faster now.
03:15The thing that I find useful is to think about the percentage of tasks, not the percentage of jobs.
03:23There will be many jobs where a lot of what it means to do that job change.
03:27AI can do things much better.
03:28It can free up people to do more and different things.
03:31There will, of course, be totally new jobs.
03:33And many existing jobs will entirely go away to be replaced by these new jobs.
03:37But I think the more interesting thing is of everyone's job, what percentage of that, what percentage of the tasks you do every day will be done by AI.
03:45And I can easily imagine a world where 30, 40% of the tasks that happen in the economy today get done by AI in the not very distant future.
03:53You've become a father this year.
03:55You've become a father this year.
03:56What would you recommend your son to do that, so that his job in 30 years is not just from KI?
04:02The meta skill of learning how to learn, of learning to adapt, learning to be resilient, to a lot of change.
04:12I mentioned this earlier, but learning how to figure out what people want, how to make useful products and services for them, how to interact in the world.
04:20I'm so confident that people will still be the center of the story for each other.
04:26And anything in that world, I think, will be great.
04:29And I'm also so confident that human desire for new stuff, desire to be useful to other people, desire to express our creativity, I think this is all limitless.
04:40In all these previous technological revolutions, people wonder, rightly so, what are we all going to do?
04:45In the industrial age, these machines came along and we watched them do the things that we used to do and said, what would be the role for us?
04:50And each new generation uses their creativity and new ideas and all of the tools the previous generation built for them to astonish us.
04:59And I'm sure my kids will do the same.
05:00Du klingst sehr optimistisch, aber es gibt natĂŒrlich auch KI-Kritiker, die vor allen Dingen die dunkle Seite und die Gefahren sehen.
05:07Zum Beispiel der berĂŒhmte KI-Forscher Elisa Jutkowski.
05:11Ich zitiere ihn mal.
05:12Der sagt nÀmlich zum Beispiel, das VerhÀltnis der Superintelligenz zum Menschen wÀre ungefÀhr so wie das VerhÀltnis des Menschen zu einer Ameise.
05:20Wir denken nicht darĂŒber nach, ob wir einen Ameisenhaufen zerstören.
05:23Wir tun es einfach, sofern er uns im Weg ist.
05:25Also wie groà ist deine persönliche Angst, dass die KI uns irgendwann wie Ameisen betrachtet und uns einfach zerstört?
05:33I've heard many people describe many different ways of what the relationship between an AI and humanity will be.
05:40The one that has always been my favorite is my co-founder Ilyas Jutskowar once said that he hoped that the way that an AGI would treat humanity, or all AGI's would treat humanity, is like a loving parent.
05:58And given the way you asked that question, it came to mind.
06:00I think it's a particularly beautiful framing.
06:02That said, I think we, when we ask that question at all, we are sort of anthropomorphizing AGI.
06:12And what this will, I believe, will be is a tool that is enormously capable.
06:16And even if it has no intentionality, by asking it to do something, there could be side effects, consequences we don't understand.
06:22And so it is very important that we align it to human values.
06:25But we get to align this tool to human values, and I don't think it'll treat humans like ants.
06:32Let's say that.
06:33Kritiker werfen dir ja vor, dass du Open AI von einer gemeinnĂŒtzigen Institution, die die Risiken von KI erforschen sollte, zu einem kommerziellen Unternehmen umgebaut hast und da auch teilweise die Sicherheitsrisiken auĂer Acht gelassen hast.
06:47Trifft dich diese Kritik? Stört dich das?
06:49Oder denkst du, man muss einfach manchmal vorangehen, wenn man Fortschritte erziehen will?
06:53First of all, we still have a non-profit entity, and we always will.
06:57I hope that we will have, I believe we will have, the best resourced and hopefully the most impactful non-profit of all time.
07:03And this is very important to our mission.
07:05Also important to our mission is the governance role of this and ensuring that we stick to our mission and that we prioritize safety and the well-being and the maximum benefit of humanity.
07:15We've obviously made some mistakes as we understand this new technology we'll make more in the future.
07:19But on the whole, I'm extremely proud of our team's track record on figuring out how to make these services safe, broadly beneficial and widely distributed.
07:28One of our core beliefs is that if we can figure out how to build this tool, align it with human values and then put it out into people's hands and have them express all of the things they want to do with this,
07:37that will be great for the world and is deeply in accordance with our mission.
07:41In Europa, da diskutiert man ja gerade sehr lebhaft ĂŒber den AI Act, um kĂŒnstliche Intelligenz zu regulieren.
07:49Wie ist dein Blick da drauf?
07:50Sind die Regeln gut?
07:52Sind es zu viele Regeln oder sollte man die einfach wegwerfen?
07:54Du hast dich gestern ja auch mit dem Bundeskanzler getroffen.
08:22Was hast du mitgenommen aus diesem GesprÀch?
08:25Ich war sehr beeindruckt.
08:26Wir hatten eine groĂe Diskussion ĂŒber die Neu- und Infrastruktur in Deutschland, um AI-Services in Deutschland, by Germany, for Germany zu ermöglichen.
08:34Wir waren sehr begeistert ĂŒber unsere recenten Erkennung, gestern mit SAP und Microsoft, um eine sovereign Cloud zu ermöglichen,
08:41um die deutsche öffentliche Sektor zu nutzen, um Frontiermodelle zu ermöglichen und SovereignitÀt zu ermöglichen.
08:44Wir hatten eine groĂe Diskussion ĂŒber das auch.
08:46Du sagst, ihr wollt Infrastruktur in Deutschland bauen.
08:50Das Problem ist, der Strom ist in Deutschland ungefĂ€hr dreimal so teuer wie in den USA und fĂŒnfmal so teuer wie in China.
08:56Ist Deutschland ĂŒberhaupt ein attraktiver Standort fĂŒr OpenAI?
08:59So, certainly, the energy costs are a challenge for AI.
09:04However, we had also a good discussion about how to address the energy needs of AI.
09:10And more than that, I think that use of AI will be one of the best uses of energy,
09:17whether it's used in Germany or, you know, you run on servers that are somewhere else
09:21because there's, you don't want to build the data centers here, which would be a very valid choice.
09:26The importance of delivering AI in Germany to German businesses and German consumers is very important.
09:32Germany is our biggest market in Europe.
09:34It's our fifth biggest market in the entire world.
09:36Virtually all young German people use ChatGPT.
09:39So, AI is here and people are getting value from it.
09:42And we'll keep doing that and we'll address the energy challenges.
09:45Was wĂ€re denn deine Empfehlung fĂŒr Deutschland, wie man die Energiepreise senken könnte?
09:53Ist nicht zum Beispiel der Atomausstieg aus deiner Sicht vollkommen irre?
09:59Look, again, this is up to the German people and I don't understand all of the sort of trade-offs locally.
10:05I personally believe that nuclear energy, advanced fission, fusion, the whole set of approaches
10:12is one of the most promising approaches to energy and that's something that the world should pursue more.
10:19In dem Rennen um kĂŒnstliche Intelligente dominieren ja ganz klar die USA und China.
10:24Deutschland hinkt ziemlich hinterher.
10:27Was mĂŒsste Deutschland tun, um aufzuholen im KI-Wettrennen?
10:31I was very impressed in my meetings with German CEOs yesterday.
10:34I think the companies here really plan to make a big bet on AI, implement it in big ways.
10:40They believe that it can add huge value to what they're doing.
10:43Heard the same message from the government about its ambitions for AI.
10:46So, again, given what we are seeing for AI adoption in Germany, it's like unbelievably strong.
10:52I think it's up 5x in the last 12 months.
10:54Incredible.
10:55I'm optimistic about this.
10:56Die ganze Welt guckt ja auf eine Innovation aus dem Hause OpenAI.
11:02Es soll demnÀchst auch eine Hardware geben, ein OpenAI-GerÀt.
11:05Du hast dafĂŒr einen Designer von Apple eingestellt.
11:08Das heiĂt, wir gehen davon aus, das GerĂ€t wird auf jeden Fall gut aussehen.
11:13Aber was wird das OpenAI-GerÀt können?
11:17ErzÀhl uns doch ein bisschen was.
11:18It will be looking good.
11:20We, I think we've only had one or two big revolutions in how we use computers in a long time.
11:28We had the mouse and the keyboard and the idea of the monitor displaying this sort of windowed system.
11:32And that was a breakthrough for sure.
11:34And then we had the touch devices, which adapted that, took out the mouse, let you just use your finger and have this very personal device.
11:42And this is huge, but, you know, fundamentally, we have never had something as powerful as AI.
11:49Computers really can understand what we want, can think, that has let us reimagine what it could mean to use a computer.
11:55So we're still exploring.
11:57It'll take us quite some time.
11:58This, nothing, don't expect anything very soon.
12:00But over time, I expect we'll make a small family of devices.
12:04They will look good for sure, but that's not the main thing.
12:06I hope, if we work hard, if we do a really great job, they will change what it means to use a computer and how you do your work and how you play and live your life.
12:17But there's a lot of work and a lot to explore between here and there.
12:21Jetzt hast du mich natĂŒrlich neugierig gemacht, da muss ich nachfragen.
12:24Sag mir eine Eigenschaft oder eine FÀhigkeit, die so ein GerÀt haben könnte.
12:27Ein Beispiel.
12:28You should, right now, if you want to do a task on a computer, you have to, like, click around.
12:34You have to go between a bunch of applications.
12:35If it's a difficult task, it might take you a while.
12:38One of the promises of AI is you can say something complex you need to happen over the course of a day or a month or a year even.
12:45And you can just trust that the computer will understand it, do it for you, come back to you when it needs help.
12:50But you can imagine asking a computer a very nuanced, complex but brief question and then just trusting that it's going to do the right thing and come back to you when it needs help.
13:00And that would totally change what it feels like to use the computer instead of launching a bunch of apps and having notifications constantly come up.
13:08So that's one thing.
13:10In der Tech-Industrie in den USA, da herrschte ja lange ein liberal-demokratischer Grundton.
13:16Jetzt fÀllt auf, viele Tech-Gurus, Tech-Bosse zeigen sich auch mit Donald Trump.
13:22Du warst ja auch direkt am Tag, glaube ich, nach seiner AmtseinfĂŒhrung schon im WeiĂen Haus, hast da eine groĂe Initiative verkĂŒndet.
13:28Wie erklÀrst du dir diesen Vibe-Shift, diesen politischen in der Tech-Industrie?
13:32Warum sind jetzt plötzlich alle doch ganz gerne mit Trump zu sehen?
13:37First of all, I think the Tech-Industrie should work with whoever the American President is.
13:42But in this specific case, I think there have been some welcome policy changes.
13:48The ability to build infrastructure in the United States, which have been quite difficult and quite important.
13:54The companies like ours, President Trump, has done an amazing job of supporting.
13:57And a more general pro-business climate and pro-tech climate has been also, I think, a welcome change.
14:04So, you know, that's kind of how I describe it.
14:06Die USA sind ja sehr polarisiert, geradezu gespalten.
14:10Ich habe selber ein paar Jahre dort gelebt und das miterlebt, wie aufgeheizt die Stimmung ist.
14:15Wie findest du die Idee, statt eines US-PrĂ€sidenten kĂŒnftig einfach eine kĂŒnstliche Intelligenz regieren zu lassen?
14:21I don't think people want that anytime soon.
14:25Okay.
14:25What I expect though is that presidents and leaders around the world will use AI more and more to help them with complex decisions.
14:32But I think we all still want a human, you know, signing off on that at some point.
14:37Letzte Frage zum Schluss.
14:39Du hast vorhin erwÀhnt, wie viele Menschen in Deutschland ChatGPT schon nutzen.
14:42Ich habe mir das mal im Detail angeschaut.
14:44Viele Deutsche holen sich sogar Beziehungstipps von ChatGPT.
14:48Hast du deinen eigenen Bot auch schon mal um Hilfe gebeten in Beziehungsfragen?
14:53I don't use it as much for that as other people do.
14:56I've tried it, but no, that's not one of my big personal use cases.
14:59It's clearly something a lot of people use it for.
15:02Okay, Sam Altman, vielen Dank fĂŒr dieses GesprĂ€ch und Ihnen vielen Dank fĂŒr Ihr Interesse.
15:07Wir machen hiermit Nachrichten weiter.
15:09Das war das Welttalk-Spezial mit OpenAI-CEO Sam Altman.
15:18Wir machen hiermit Nachrichten weiter.
Recommended
26:25
|
Up next
2:14
1:02
15:42
8:09
9:00
0:56
Be the first to comment