The AI Power Game

00:00:00: the positive sentiment on AI is only twenty percent, which means eighty percent of people see AI as a negative to their life.

00:00:09: Once again how much?

00:00:10: Eighty percent are negatives?

00:00:13: Yes, eighty percent are negative and twenty percent are positive.

00:00:16: maybe there's some neutral in the middle but there's a lot more negative right now than then.

00:00:21: there Both the wrong narrative that's being pushed by AI leadership, as well as news coverage of all the layoffs coming on.

00:00:42: The energy price increases are coming because of data centers and inflation is happening And everybody blaming AI for a lot these things.

00:00:55: Hello, hello.

00:00:56: Let's talk about big bank tech report.

00:00:59: this is our new episode.

00:01:01: and Yeah we will talk about the AI power game About Trump about see you about Google and about Europe's choice.

00:01:10: And I'm very happy that Evan Grayland is here in my show an hour Show.

00:01:16: Happy to see your album.

00:01:18: Yeah, good to see you again Jens.

00:01:19: We've been off for about three weeks because we were both traveling but I'm...I think there's lots to talk about today.

00:01:24: You are travelling a lot and you have been in Europe!

00:01:28: What did you do in Greece?

00:01:32: So last week i was at the Delphi Economic Forum And you know, gathering of a lot both state leaders as well as thinkers in Europe and actually around the world.

00:01:43: So it was interesting to hear from perspective what people are seeing what the topics were at Davos just a few months ago.

00:01:53: You know, I think the Iranian war has really changed people's perspective of both reliance on the US as well as relationship with other countries around the world and also technology sovereignty energy sovereignty even food security where all as well defense we're all major topics.

00:02:14: that was talked about this week.

00:02:15: What is your takeaway from Greece?

00:02:19: My takeaway is that Europe is a little bit confused.

00:02:26: It's still not sure what it needs to do, there's a lot of debate internally.

00:02:30: some are still hoping when Trump leaves things will go back normal and I think also action says we cannot rely on the US anymore.

00:02:40: We have to be independent and on all those fronts that we just talked about.

00:02:46: And I think the second group probably makes more sense, but this region takes time to make decisions because when you have a twenty-seven country trying

00:03:01: Yeah, but we are asking ourselves will Europe?

00:03:05: Will it shape the AI future?

00:03:07: or would read the press release after America and China has made a deal.

00:03:14: How do you see that?

00:03:16: I think That was one of the big topics is how can Europe play a role?

00:03:23: I think the ones who are saying, hey we can be a leader in AI and Europe.

00:03:29: In terms of innovation an invention?

00:03:32: They probably heading down their own road.

00:03:34: they just isn't resources both in terms of chips or energy essentially human talent to do all that.

00:03:44: but what does make sense is working with open source community because thats three million person deep community around the world that is building AI and leveraging what's available.

00:03:58: And there are some very, very capable models that are coming out of both US China and Europe in this space.

00:04:05: so rather than trying to reinventing create something amazing why not work with that community?

00:04:13: When you hear keywords tech restrictions chips tariffs trade What do you think about it?

00:04:24: Yeah, so you know in a couple weeks there's going to be a meeting between Xi Jinping and Donald Trump in Beijing.

00:04:31: You believe that it will happen?

00:04:33: It'll happens.

00:04:34: this meeting or

00:04:36: I think it will...it was postponed last time because it happened kind of right at the beginning.

00:04:46: Trump wants to have some good news we'll talk about.

00:04:48: He wants a win, he wants a deal.

00:04:51: so you know there's definitely a lot of things at stake right now both the terrorist export control issues recently yesterday I think or maybe today before Manus just had their deal revoked by the Chinese government saying that they need to, acquisition by Metta is now blocked.

00:05:14: So they have to unwind their whole deal and I think that's really more of a game in a pond between

00:05:20: U.S.,

00:05:20: China say here.

00:05:21: so another leverage that I haven't.

00:05:24: it says signal two Chinese companies.

00:05:27: be careful about skirting the rules to go out of the country.

00:05:32: And, you know... To be fair I think that US have actually done similar things in terms of blocking Chinese from buying American companies, there's been sixty to eighty block deals over the last ten years.

00:05:44: This is a first time Chinese has actually blocked one of the US acquisitions by an American company right?

00:05:51: So I think it will be interesting...

00:05:55: You can understand and you see okay that pretty normal this would happen!

00:06:00: It's normal in fact surprising.

00:06:02: they waited long enough you know, essentially that game piece in this overall negotiation.

00:06:10: So same as the Rare Earth thing they had that rare earth piece to play or card for a long time and then played until just earlier last year.

00:06:25: I think there will be some positive news that comes out of it, because both sides want to show something good.

00:06:31: And there's going to be multiple meetings later on this year between these two leaders and they are going to setting the agenda for the rest of world at least in a near term.

00:06:42: What do you expect?

00:06:44: Is there anything big or small?

00:06:51: I mean, they're going to bring a number of US corporations to China.

00:06:55: They are gonna get some trade deals and buy more soybeans in planes or chips and other things.

00:07:01: It'll look like big numbers from the U.S perspective And Chinese will probably get concessions on trade restrictions.

00:07:13: So both sides have been talking for awhile.

00:07:16: The US have actually definitely been a little bit less prepared.

00:07:19: I've talked to people on both sides, and it feels like... A lot of things are very last-minute from the U.S..

00:07:28: Do you think that NVIDIA and Jens Noang will throw a big party after Mr Trump was there?

00:07:35: Because, well they are suffering.

00:07:38: They can't sell all their chips to China.

00:07:41: in this summit I

00:07:48: think there probably will be some movement.

00:07:49: In fact, if you look right now There was supposed to even at the beginning this year... ...there is a lifting of restriction on age two hundreds but up till now.. ..there has not been single chip shipped to China illegally yet.

00:08:02: So the fact that illegally or side grey market and i'm sure many chips have been sold Or at least shipped in smuggled then.

00:08:14: But when that restriction was lifted and no real sales has happened, I'm sure Jensen is not particularly happy about it.

00:08:24: He's been very vocal the last couple of months trying to say why China... Why the US needs to give China at least one generation back chip?

00:08:34: because allows technology stack to continue be used.

00:08:39: But if he starts talking, he praises all the time president of United States and then is talking for us.

00:08:47: it's really incredible what you guys are doing first with the politics to have more.

00:08:57: Yeah, I think this is unfortunate but it feels like what's required these days that leaders' egos are sometimes important than reality of business.

00:09:12: You have read some headlines, riots or something like that.

00:09:16: Molotov cocktails in San Francisco to the house of Sam Oldman and especially the generation said they are not convinced that AI is a part their world but happy with what's going on there?

00:09:35: Yeah, I mean i think this is actually a really big problem and probably A bigger problem in the us than almost anywhere else In the world right now.

00:09:42: The positive sentiment on ai Is only twenty percent which means eighty percent of people see AI as a negative to their life.

00:09:51: Once again how much eighty percent?

00:09:53: eighty percent are negative.

00:09:55: Wow.

00:09:55: Yes, eighty percent a negative and twenty percent are positive.

00:09:58: maybe there's some neutral in the middle but But There is A lot more negative right now than there is.

00:10:03: then there Is positive And I think that That really a bad sign.

00:10:09: It' s both The wrong narrative thats being pushed by the AI leadership as well As the news has coverage of all the layoffs that Are coming on you know, the energy price increases that are coming because of data centers.

00:10:29: The inflation is happening and everybody blaming AI for a lot of these things.

00:10:34: so it's It's A problem that needs to be addressed by the government right now And also buy some of these private companies.

00:10:45: in fact I think tomorrow there's going to be a big session where Bernie Sanders, one of the Democratic senators is going to hold a town hall meeting essentially at the Capitol building about this topic.

00:11:03: So it's something that in the news almost every day now and needs to be addressed.

00:11:09: Yeah but we have some experts.

00:11:12: they asked for CERN for AI.

00:11:15: What is that?

00:11:17: Can you explain it for us or our audience.

00:11:20: Yeah, so actually the Sun for AI concept has been around for a couple years.

00:11:23: It was also.

00:11:24: I discussed this in my beyond rivalry paper at the end of last year and my Stanford paper.

00:11:29: But the idea's that rather than having each country Make their own AI and try to win at everything.

00:11:35: We should try to consolidate resources Just like CERN did with particle physics, you know we don't have a particle collider in every country You know.

00:11:43: so where?

00:11:44: Well, you have essentially A neutral party Like Switzerland.

00:11:48: were you can everybody invest hundreds of millions or billions of dollars And when the results come out from this it's shared with the world.

00:11:58: I actually think that really the right process or methodology to build AI, because when you do that then turn AI into a global public good.

00:12:10: Right now it is being built as either an advantage for a certain company to win the market or A weapon for domination by country.

00:12:21: Yeah, too To win in geopolitics.

00:12:25: And when you do that?

00:12:26: You're essentially duplicating efforts across all of these labs your duplicating energy build and data center builds and you're splitting the Human resource.

00:12:38: so if we can actually if Europe actually can take a bigger role I think this could be a place where Europe doesn't have to, beyond the sideline.

00:12:47: Doesn't have wait they can actually help to unite that resources of world around common cause

00:12:54: But we are not used to play a bigger role.

00:12:57: if you watch Iran right now.

00:13:00: If you watched other things, AI... It's almost that we're reacting and do not move forward at the moment.

00:13:09: And one question again is what is special with CERN?

00:13:14: What is very interesting for you in Switzerland?

00:13:22: Well, CERN something has been around for many decades but it is the largest particle collider and one of things outside of the particle physics side.

00:13:30: One thing that came out as CERND was actually the worldwide web right?

00:13:33: So we wouldn't have internet in a world wide web without what were some other things that come from CERM?

00:13:39: so I think this even though you know, even though Europe has played a kind of background role the fact that it is seen as neutral party between US and China allows to be connector bridge.

00:14:00: It has so much talent in, you know that's in Europe.

00:14:04: That is separated between all these countries and now You can actually put them together And maybe bring some of the international talent as well?

00:14:12: This also may be allows International The countries to then invest into Europe To build some infrastructure.

00:14:19: that Is much needed right now because Europe it this way behind In terms of compute infrastructure versus versus the US.

00:14:28: So there's a lot of benefits to doing this, and not to mention reducing global conflict which I think is a benefit not only to Europe but really to the world.

00:14:40: If we have look at AI engines... There is nothing that Europe is contributing.

00:14:46: We see right now that China with DeepSeq what going on deep-seq?

00:14:51: And where do you see the race?

00:14:54: Yeah, so I mean you know this is the The interesting and also frustrating thing about AI as right now every few days There's a brand new release of something amazing.

00:15:04: Right?

00:15:04: And yeah had over the last Last two weeks we've had that deep seek for four before which people been waiting for very long time.

00:15:11: We've had the GPT-Five point five, which right now is essentially the top performing model in world.

00:15:17: We've had Minimax two point minimax to point six Kimmy two point six.

00:15:24: we've had The new GLM models and you know Essentially every.

00:15:29: every company actually and show me yesterday just came out with a new model that Is essentially?

00:15:34: Top open source model now.

00:15:36: so every week every few days, there's a switching of the top.

00:15:40: And it's a little exhausting actually for this industry.

00:15:43: but also what is shows us that deep-seek as company no longer really core story for AI in China.

00:15:51: It now just part.

00:15:53: half dozen companies are essentially about strong and all making independent innovations.

00:16:04: There's a race, there are different models.

00:16:07: But do you see breakthrough or something?

00:16:09: Okay every time when we're talking say oh the races going on but for me in my life nothing has changed over the last month.

00:16:19: Yeah so I think if your not an industry You can't tell all of things that is happening.

00:16:27: Really i feel like theres been few major breakthroughs this year.

00:16:30: Which one One is the agentic capabilities of the harnesses that are around agents and how you're able to now do jobs.

00:16:40: It used be at end-of-last year or middle last year with somewhere like an hour independent work for these things.

00:16:47: The latest models coming out have a meter.

00:16:51: METR benchmark is now up to about forty hours of independent work that it can do without human involvement, right?

00:16:57: That means essentially you can a person's one week of work by sending in one instruction.

00:17:03: One prompt and they will do weeks for what going on.

00:17:07: And its really just when this technology be integrated into the workplace or workflow smoothly How we change the existing company's workflows and processes.

00:17:22: So when that happens, the type of productivity gains is going to be immense.

00:17:30: The other thing it actually memory both in terms a short-end long term memory.

00:17:34: context windows are getting bigger and bigger which means you can fit a lot more content into the consideration set for models.

00:17:42: yet I think all these markdown files now creating longer memory.

00:17:46: So that was one of the biggest missing things in the prior models.

00:17:50: Now, essentially you can have dozens or hundreds files all being organized in different markdown files so that you don't lose context and memory between large jobs.

00:18:02: So it can go and refactor entire giant code base or do big jobs, a big analysis without getting confused?

00:18:11: And that was something that was not possible just six months ago.

00:18:14: But well there's a lot of change.

00:18:16: but if you go out here on the street in Hamburg and Germany and ask people are afraid your job is going on?

00:18:23: I think nobody until now that oh, I'm really unhappy.

00:18:28: what's going on right.

00:18:29: Now we do not see it because people they are quitting their job looking for another job.

00:18:37: Until now there is no frustration.

00:18:38: until now We don't see in the market What's going?

00:18:41: On Germany like

00:18:45: I think Europe is a little bit protected from some of these forces.

00:18:50: What we're seeing in the US for young people, it's absolutely happening right?

00:18:53: People who are graduating from their best schools now have really tough time finding jobs.

00:18:58: The unemployment rate in China for new graduates is somewhere around twenty percent.

00:19:05: In the US, it's from seven to eight percent.

00:19:08: But then under-employment right now for new graduates in the

00:19:11: U.S.,

00:19:11: is around fifty percent.

00:19:13: Seven percent are unemployed but then forty two percent doing jobs that far below what a graduate from my college would be doing, delivering something on Uber Eats or going and working in the coffee shop.

00:19:27: So people are not necessarily being... They may be employed but they're not gainfully at that level where there will be happy.

00:19:36: Now I think we need to really realize is Job displacement, technology displacement isn't a linear graph in the sense of if it goes up ten percent capability you don't get more job displacement.

00:19:51: What happens is what's called J-curve and essentially starts off very slow.

00:19:55: but once that gets to critical number which probably around seventy to eighty percent able to be done automatically, then the replacement goes up a straight line.

00:20:06: And when we are right now if you look at all of the benchmarks for real world work it's getting to that sixty-seventy eighty percent mark and so I think what will find is that displacement is going coming in next year much more significantly than we have seen.

00:20:23: So where not seeing this storm.

00:20:26: thats really underneath

00:20:28: But the storm was already in your family because I know that, I think you're daughter.

00:20:35: She was working at IT industry and now what she's doing right now?

00:20:39: And is she happy?

00:20:40: Yeah so actually i have two daughters.

00:20:42: one of them was laid off last year due to AI Because they were doing data analysis.

00:20:47: They moved into another company but the one who did well at Microsoft decided to quit.

00:20:55: every week Different friends of hers were getting laid off, and there was so much pressure.

00:20:59: And every week they're trying to do more with

00:21:02: less.".

00:21:03: She said I don't like this pressure or being a cog in the machine.

00:21:12: Yes, they're focused on getting people to get off the screen and be face-to-face with people.

00:21:21: And it's doing very well.

00:21:22: by the way I mean every day there are lines out of door thinking about opening up a second location already.

00:21:27: so i'm very happy for her.

00:21:29: she is tired but she was very happy.

00:21:32: She has used AI as business plan to convince banks all this stuff

00:21:39: For legal reviews or some design work Advertising you know, so she has a small team But they look like they have very big presence.

00:21:49: right and I think this is what's going to be different Is that AI will enable small companies To behave And perform large companies.

00:21:58: We talked A little bit about last

00:21:59: time

00:22:00: but i think This is something That Will increasingly Be so Increasingly easy for Small Companies Because The Interface Would Just Be Natural Language.

00:22:11: Yeah, and you are happy with that situation.

00:22:14: or we've the change because now she's not working as in a well in power of Microsoft.

00:22:20: And she is earning a lot of money and Now she's doing it different job Because there has a father in this and then as society It's different to have an academic job and than to work In a coffee shop or as a founder on a coffee.

00:22:37: Yeah, I mean so.

00:22:38: I mean you know.

00:22:39: as I was I had to be honest.

00:22:41: I was a little bit concerned in the beginning because I said look You spend your entire life?

00:22:46: No being trained to do what you're doing and you have job at one of the biggest And most prestigious companies in the world.

00:22:52: yeah But when would I have to say is after seeing how she's How her life is and how or Her I guess her emotional state and her personal satisfaction is i'm actually supportive now Because I think I think she's doing what she enjoys.

00:23:08: She is finding it meaningful and happy waking up every day to go to work, whereas that was not the case.

00:23:16: .I remember when you're first joined Microsoft ,every week i would have a coffee with her and she'd be complaining about all of the issues as she has in why shes frustrated.

00:23:24: And now its about any advice for how do this or that?

00:23:28: How can we grow our

00:23:29: business?".

00:23:29: You know...i'm really happy.

00:23:31: Yeah,

00:23:33: last week there was a new story in the magazine and Germany.

00:23:38: And that was written that Google is doing quite fantastic job.

00:23:45: people thought months ago That this search machine will be destroyed by AI?

00:23:52: Now it's the opposite.

00:23:53: What are your thoughts about Google?

00:23:58: I think right now Google is in a very good position because they really do have the full stack.

00:24:03: They have users, data and access to a lot of computers as well as phones around the world.

00:24:12: And you know there's business that churns out money with search.

00:24:17: Actually i'm surprised that AI has not affected their search businesses much than expected Because for people every friend I know that's in tech, their usage of search has gone down.

00:24:31: you know fifty sixty eighty percent from what they used to be.

00:24:36: But i think somehow they've been able to maintain the revenue there.

00:24:43: You know, I don't know how long that can last.

00:24:49: That's really the thing.

00:24:50: i'm not sure...I think they're going to be a major player because They actually have their own hardware or not?

00:25:01: But they also have drawn down significantly their free cash flow and their cash reserves, right?

00:25:07: Because you're building so much hardware or so much capex around the world which some of other companies haven't done like we talked about Apple before.

00:25:15: They've essentially...Apple is spending somewhere in order five to ten billion dollars a year on capex.

00:25:20: Google and Microsoft are spending hundreds million dollars.

00:25:31: it could be that, you know this is still early in the game and we may see some issues coming out for Google longer term.

00:25:45: And in fact last week GPT-Five point five just came up with their new image model, the Image Two Point O which significantly outperforms the Nano Banana model from Google text to image model and now no longer true, right?

00:26:03: So the fact that OpenAI has surpassed them I think is alarming for Google.

00:26:11: But it's a race.

00:26:12: maybe in two weeks or three weeks Google will surpass again.

00:26:17: We do not know it right now.

00:26:18: because, no.

00:26:19: Not because we see a lot of change and you mentioned Apple?

00:26:35: Yeah, I think Apple's actually in a better shape than people.

00:26:39: Think everybody says oh apple missed the boat on AI?

00:26:42: I Actually think that they probably made the right move because when when AI?

00:26:47: Right now what it looks like is AIs becoming much more commodity.

00:26:50: every Every model within one or two percent of each other and there are no able to go and license did The best models from Everybody?

00:26:57: for you know essentially their deal with Google was only four million dollars per year to get one of the best AI models, and they're getting paid ten billion dollars a year by Google to provide search on Apple.

00:27:12: So there actually making an exchange nine billion dollars just cash flow right?

00:27:17: But you'll find that Anthropic and probably OpenAI in other companies will come to Apple say hey we want to get onto your list too let me pay use some money Right, so this is what they did with Search.

00:27:33: I think that will probably just do the same thing and put a different model behind Siri.

00:27:38: but Siri is the front end And from that perspective it's cash smart for Apple But you don't lose control of your customer.

00:27:52: Evan What are on our minds these weeks we haven't talked about yet?

00:27:56: Is there something to share with us?

00:27:59: Yeah, so I think you know.

00:28:02: one of the topics has been talked about recently is the mythos model and how cyber security's becoming a bigger-and-bigger issue.

00:28:12: And How AI is going to be a tool for national superiority right in fact this was One Of The Things that Was Discussed Quite A bit When i was in Delphi?

00:28:28: That needs to be taken more seriously because right now everybody thinks that this AI race is really about creating a national advantage.

00:28:38: But the problem with that, when you start to do that... You start creating systems that will be very accessible to bad actors, right?

00:28:49: In fact just in the last two weeks you've had Anthropic.

00:28:53: The company who makes Mythos.

00:28:54: they leaked their own source code for a cloud code.

00:28:58: one of the most important parts of your business was worth billions dollars.

00:29:01: so there's very bad internal data security access security.

00:29:06: They also like just a couple days ago released at their mythos model and had a lot unauthorized access by users expect, right?

00:29:16: So what this is?

00:29:16: showing?

00:29:17: that if you try to weaponize this AI and an AI software it's very different than nuclear weapons or hardware missiles.

00:29:27: That can be easily secured and detected.

00:29:31: This thing essentially one USB stick away from being accessible to anybody in the world.

00:29:38: And when thats the case we really need think about how do work as a world against a common threat, which is bad actor misuse versus state-to-state use.

00:29:49: Because right now the focus has been oh we get the best model then we can use it to then spy on our opponents.

00:29:56: and when you do that you create escalate tension.

00:29:58: You stop talking each other?

00:30:00: You start creating information in terms of where the risk are.

00:30:02: what We need To Do Is really to say okay...we do The same things That we do for terrorist activities For all The hackers that are going around the world trying to take down power grids and do ransomware.

00:30:15: We should be doing this, That level of sharing intelligence between countries And then go create share safety standards Hotlines for making sure we aren't mistaking activities For state activity when they're really bad actors.

00:30:31: This is one thing Xi Jinping and Trump will probably start talking about when they meet in a couple weeks, is how to make safety better.

00:30:40: And how does it makes sure that you know be careful of using it on automated weapons?

00:30:45: but also How do we share safety?

00:30:48: so the world Is a safer place overall even though You maintain certain level competition On national security side.

00:30:58: But against common risk We have work together.

00:31:01: Yeah, sometimes we have the impression that these guys they do the right decisions.

00:31:08: That say...that they uh..do it.

00:31:10: but sometimes if you look at Iran and so on The decision are not quite well.

00:31:16: We will have a look What's going on?

00:31:19: It is really big topic And we need to talk about this.

00:31:22: So for me AI isn't waiting China isn' t waiting America isn''t waiting Big techs aren't waiting and Europe cannot wait.

00:31:29: And you guys are listening, watching us... You have to wait for fourteen days!

00:31:36: Then we'll talk about everything in the next episode.

00:31:40: Wow it's going so fast but I'm really happy that Alvin is the best expert of all time on our show.

00:31:47: Thank-you very much.

00:31:50: You have for the first time in months slept more than ten hours and you are living on a plane.

00:31:59: And it's nice to have you here, let us have new episode in two weeks!

00:32:05: Yeah yeah It is always fun to chat with your audience.

00:32:08: Please send me questions.

00:32:10: Let know what you're interested In share this episode with friends.

00:32:14: I think there will be useful information that world needs hear.

00:32:20: And we see each other live in September, and Berlin the Big Bang AI Festival.

00:32:28: We have already five thousand people that has bought a ticket.

00:32:32: last year there were five hundred.

00:32:34: so There is lot of interest on our topics.

00:32:38: Let's talk about it!

00:32:41: Thank you very much Alvin.

00:32:42: thank everybody.