AI Power Moves: Genesis Mission, Google & Deepseek, and How you can prepare.

00:00:00: who's going to win this AI race, right?

00:00:01: Every day, there's a new model coming out.

00:00:04: Every day, there's a new benchmark being broken.

00:00:07: In fact, just a couple of days ago, Sam Altman said to his team, we're at Code Red.

00:00:14: We're not doing anything else.

00:00:15: We're just going to make our models bigger and better.

00:00:17: And we're going to stop advertising and doing some of these other things for a while.

00:00:21: And I think that that really shows that there is a sense of urgency.

00:00:29: Welcome to Big Bang Future Lab, the show that brings clarity to a world accelerating at light speed.

00:00:37: We cut through the noise and explain the technologies reshaping our lives.

00:00:42: All in under twenty minutes.

00:00:44: I'm Jens de Boer, a journalist and author covering technology, business and politics across Europe for more than three decades.

00:00:54: Hi, and I'm Alvin Graylin.

00:00:56: I have been for thirty five years working in the technology space and AI and XR and semiconductors across China and US and Together we're going to make the unclear much more clear.

00:01:09: So let's dive in

00:01:11: The unclear more clear.

00:01:12: I need it as a journalist.

00:01:13: I love this sentence, you know, because that's my job to make the unclear more clear and to have guys like you who explain it to me.

00:01:22: So first, first of all, if you go out tonight to have a beer and a pop, what do you guys, your tech guys, what are you talking about?

00:01:30: What is on the agenda?

00:01:32: The most important thing?

00:01:34: I mean, right now, it seems like everywhere in the tech circles, people are saying, you know, who's going to win?

00:01:40: say iris right every day there's a new model coming out.

00:01:43: every every day there's a new benchmark being broken.

00:01:47: in fact just a couple days ago Sam Altman said to his team, we're at cold red.

00:01:53: We're not doing anything else.

00:01:54: We're just going to make our models bigger and better.

00:01:57: And we're going to stop advertising and doing some of these other things for a while.

00:02:01: And I think that that really shows that there is a sense of urgency.

00:02:06: Whereas a few months ago, really, I think everybody thought Open AI was ahead.

00:02:12: They're going to win it all.

00:02:14: And I think right now what's happening is that it's showing that the winner is up for grabs.

00:02:20: And it looks like actually maybe some people believe that Gemini is going to be, Gemini is three from Google is probably going to be the winner.

00:02:31: But we still have another month left in this year.

00:02:34: things are going to change again.

00:02:36: In fact, here's something that's very interesting.

00:02:39: We heard about the deep seek moment earlier this year.

00:02:42: They've been very quiet for about five, six, seven, eight months.

00:02:45: And just a couple of days ago, they just launched their deep seek, three point two.

00:02:50: And it's a model that is open source and is as good as the leading frontier models and is available to everyone.

00:02:58: And now it has agent capabilities built into it.

00:03:01: So, and it's much, much smaller than all of the other leading edge models, which means you can run it at home.

00:03:09: I feel like these are the kind of things that are happening every day, and it's quite an exciting time in this industry right now.

00:03:18: But here in Germany, we do not need it, or we do not use it very often because we say, oh, it's a Chinese instrument.

00:03:24: And so it could be dangerous and so on.

00:03:27: And you in the US, you use it?

00:03:30: You have deployed it?

00:03:31: Well, yeah.

00:03:32: I mean, I use it.

00:03:33: In fact, I was just playing yesterday with several data models.

00:03:36: I haven't gotten access to the latest from DeepSeq, but I was using some of the latest Quen models.

00:03:43: It's the Alibaba model.

00:03:45: And it works really well.

00:03:46: And actually, I I have a quantized model that can run on my MacBook, and I don't even need to go outside.

00:03:52: And I was using it, and I was very impressed with the quality of what you can get with just a thirty or forty B model compared to a six trillion or seven trillion parameter model that is coming out from Google or from OpenAI or from GROC.

00:04:06: These are giant models, and essentially a hundred times smaller models are performing similarly quality.

00:04:17: pretty amazing and that's democratizing.

00:04:18: So you're not afraid of it because it's coming from China and that they will take all your data, bring it to China.

00:04:26: That's some of the rumors.

00:04:28: Well,

00:04:29: the thing to remember is that when these models are available, they're open source, so they're available for download.

00:04:35: You download it, you put it onto your own on-premise cloud, or you put it onto your own home computer, there is no Message going back to anywhere else in the world, right?

00:04:46: And and they are allowing these models to be You know fine-tuned and adjusted however you want.

00:04:52: in fact last week Singapore just announced that they have chosen Ali Baba's model for their sovereign national model Right for a country to then choose a open source model as their national model and and you know essentially a friend of America's to choose a Chinese model.

00:05:11: that that's saying a lot.

00:05:13: Last week, we have had the news, the breaking news or something like that.

00:05:18: It's called the Genesis mission.

00:05:21: What is a Genesis mission?

00:05:22: Can you explain it a little bit and then people compare it with the Manhattan project?

00:05:27: And why can you compare it?

00:05:29: Because that's the atom bomb.

00:05:31: Yeah, well, I mean, so.

00:05:35: I like the idea behind the Genesis mission.

00:05:38: It's essentially, it's America's push to say, hey, we need to become a science leader again, and using AI to integrate with all of the national labs of America.

00:05:51: There's, you know, like, sixteen or seventeen of them.

00:05:53: And it's being run by the Department of Energy, the same group which ran the Manhattan Project.

00:06:00: So in fact, in the paper, that talks about this announcement, it specifically says this is our manhand project for AI.

00:06:09: I like the concept of open science and trying to create new breakthroughs.

00:06:14: But calling it a manhand project, I think it is a very bad choice.

00:06:18: In the sense of manhand project was to create a bomb.

00:06:21: And there's only one use for a bomb.

00:06:24: kill people, right?

00:06:25: We really want to try to say how.

00:06:27: maybe we can call it the Apollo project because that's really more about discovery and more something that is that is a common good, right?

00:06:35: And that's what I think this should be positioned as.

00:06:38: This should be saying, hey, we're going to make all the data and all the research of the labs in America, national labs available to all these AI labs.

00:06:48: But then let's do it where we're trying to cure cancer, or we're trying to find new materials, or we're trying to find new ways of communicating or something.

00:06:57: Where that data is used for good.

00:06:58: Right now, when it's positioned as a Manhattan Project, what is essentially creating is an adversarial relationship with anybody that is part of the allies of America, or you're against America.

00:07:10: And essentially, whatever comes out becomes a bomb that can be used against somebody else.

00:07:16: And I feel like that's That is the wrong framing.

00:07:19: And it is an unfortunate position that America is positioning itself in today of creating something that should be a public good into something that becomes a weapon.

00:07:31: But with the Manhattan Project, we have two things.

00:07:34: We have the bomb, and then we have energy too.

00:07:37: And so we have two things, and we call I think dual use more or less.

00:07:43: You can have it.

00:07:44: Yeah,

00:07:45: I think most technology is dual use.

00:07:48: But remember, within a month or two of us developing the bomb, it was dropped and killed hundreds of thousands of people.

00:07:56: I don't know if that's the outcome.

00:07:57: And in fact, pretty much all of the major researchers that were part of it regretted what happened from their research afterwards.

00:08:06: were part of it in terms of trying to win the war.

00:08:10: But we don't need to get into the politics of it on this show.

00:08:13: But I think we need to really reframe what AI is for.

00:08:19: And how do we take the benefits of AI and really only use it for the good side?

00:08:24: Dual use means you have a choice.

00:08:27: Let's choose the positive instead of choose the framing of creating an AI atomic bomb.

00:08:34: You mentioned before some old plans.

00:08:36: Sam Wildman is the CEO of OpenAI and he has a code red.

00:08:40: What is it?

00:08:40: A code red?

00:08:43: Why did he launch it?

00:08:45: I mean, I think, you know, for the last year or two years, essentially, anytime anybody else come up with a model or release, you know, open eye would have something stored in its pocket to say, oh, you know, I'm going to make them look bad.

00:08:58: I've got this thing that I'm months ahead or years ahead of everybody.

00:09:01: And then they would just dish this out to disturb other people's announcements.

00:09:06: And, you know, this is the first time in two, three years where, essentially, there was no answer, right?

00:09:14: what Google created.

00:09:15: And that scared him.

00:09:18: I think it showed that they don't have the lead anymore.

00:09:22: And if they don't have the lead, then does their valuation make sense?

00:09:26: They also don't have a revenue stream like Google has.

00:09:29: Google has a hundred billion dollars a year revenue stream of profits coming in that allows them to fund all of these things, whereas Open AI is losing two or three billion dollars every quarter, maybe more actually now.

00:09:45: to try to do this research.

00:09:49: So I think that's why he's very afraid now and wants to refocus.

00:09:56: And in the beginning years, a month ago, everybody thought that this technology is a threat for Google.

00:10:03: is a threat if we have AI there that you do not need anymore a search machine because the AI will do.

00:10:11: and right now it's the opposite.

00:10:13: It's the stock exchange, the stocks are very high and everybody is talking about the winner, the winner is Google.

00:10:20: Yeah, I mean I think that in fact this was probably the reason why Google didn't even release the large language models when they were the ones that invented the transformer.

00:10:31: They had it for several years and they decided not to do anything with it because they were afraid of cannibalizing their search business.

00:10:38: What they're finding is that now by integrating a good search, a good AI model into their search, into their video systems, into all of their other pieces, they're actually now increasing their revenue share.

00:10:52: And they also have now a lot more data than anybody else because they have their own users with all their emails and all the Google drives and YouTube.

00:11:02: Whereas other companies are relying on partners or the general public internet to be able to get their data.

00:11:08: So Google's in a very very good position.

00:11:12: And in fact, probably something that I didn't talk about was the TPUs.

00:11:16: We talked about in the last show.

00:11:18: And Gemini III was trained completely on TPUs.

00:11:23: And TPU is the GPU, essentially ASIC, that was made by Google specifically for AI processing.

00:11:31: And they owned the entire stack on it.

00:11:33: So they designed it, they go build it.

00:11:35: I mean, they're still using TSMC to build it.

00:11:39: Or it was a Broadcom, I think, Broadcom to build it.

00:11:43: But the fact that they have the entire stack, they own the instruction sets, the tools, and they're able to do it at three or four times more efficiency in terms of cost and in terms of energy compared to Nvidia chips.

00:11:59: I think when you put that together, they'll now look like the the eight hundred pound gorilla in the AI space.

00:12:06: That's a good picture, eight hundred pound gorilla.

00:12:10: That's a real big one.

00:12:13: this step, Google is becoming a serious competitor to Nvidia?

00:12:17: Yeah, I mean, I think no, there's no, no doubt about it, because you can see what happened with Nvidia's stock price over the last few weeks.

00:12:25: And you look at Google stock price and a lot of it was based on the fact that, you know, there's now a competitor for NVIDIA, right?

00:12:34: Because before, really the only choice was going to NVIDIA.

00:12:39: Now, in fact, Meta now has made a deal with... Google to use their TPUs for future training.

00:12:46: And the fact that that happened was essentially a sign that, hey, there is choice.

00:12:51: And when there's choice, that means there's a reduction in margin.

00:12:55: If you look at the twenty six, so in one month and a new year is starting, do you see something real with transformative potential for the economy or society?

00:13:07: Is there something we say next year there will be a milestone or there will be a really.

00:13:13: big, huge thing?

00:13:15: I mean, there's so many breakthroughs every day.

00:13:18: There's so many things I can name.

00:13:21: But I think the people talk about this year being the year of agents.

00:13:25: I actually don't think that was the case.

00:13:26: This year is still really the year.

00:13:30: of these systems figuring out how to make the existing models work, how to make standalone AI work.

00:13:39: The next year, I think, agent technology will really start to blossom.

00:13:46: And another thing that just recently happened was Ilya Suskova, who's one of the co-founders of OpenAI, he did an interview.

00:13:55: And he said something, I think, very interesting, which I agree with.

00:13:58: He said, The last few years have been the years of scaling and all people talked about was making models bigger and making data centers bigger.

00:14:08: But now it is going back to the age of research.

00:14:12: And what he means is it's now up to the humans who make new algorithms and those new algorithms is where the breakthroughs will come from.

00:14:21: And I think the combination of these two where We're going to get new algorithms coming out and we're doing this every day.

00:14:27: In fact, the reason that DeepSeek is able to do the things it is, is because they've restructured how they look at the... the MOE model, how they look at the sparse attention models, so where they can use a lot less resources to do a lot more.

00:14:47: We're going to see more and more of these type of breakthroughs come in.

00:14:51: When these happen, and then you mix in the capability for agents to happen, now you have very high intelligence on very accessible hardware, now having the ability not just to think and tell you something, but to actually start doing things.

00:15:06: And when you start to combine these two, I think we're going to see a much larger impact on our economy, on the workforce, on the productivity of individuals.

00:15:17: When do you think, when we will see companies that are largely operated by AI agents with only one single CEO and I don't know, one hundred and two hundred agents and they are working for it and maybe listed on NASDAQ?

00:15:33: Is it a dream?

00:15:34: Is it a vision?

00:15:35: What is it?

00:15:36: I mean, I know people have been talking about this for a while.

00:15:40: I'm actually less, I guess I'm less focused on having a single owner who does everything because at the end of the day, whether you have one or two or three people as the founding team, it doesn't affect you economically, but it does affect you in terms of how capable you are as a team.

00:15:59: I think there's very few individuals that are both technically savvy, business savvy, and marketing savvy.

00:16:08: I think there may be a few of those people, but they're rare and they usually actually need decades of experience to get there, right?

00:16:18: So I actually think that what's more likely is we're going to get two or three people type teams that will be able to leverage technology, maybe one business founder and one technical founder.

00:16:28: And then together, they will then leverage these agents and outsource services, both human and digital services to create businesses.

00:16:38: I mean, you already have that today in terms of let's say a lot of the algorithm trading shops.

00:16:42: They're not very big organizations.

00:16:44: You could do that today with one or two, no, probably two or three.

00:16:47: people and a lot of computers.

00:16:51: And you can see that we're already doing that soon with programming because you can essentially have one or two people helping automatic coding agents.

00:17:01: You're going to have that with anything that's really done in front of a screen, primarily in front of a screen, graphics design, marketing agencies, those kind of things will happen first.

00:17:11: I think for it to get to the mass market, it's going to take some time.

00:17:19: So I don't think twenty twenty six is the year where you you see that a listing of a company like this, but maybe maybe twenty twenty eight.

00:17:27: I think that's that's probably possible.

00:17:30: Okay.

00:17:30: And But there are fantastic times for entrepreneurs.

00:17:34: You do not need a lot of money to start your business because, well, if you have an agent, they do with your business plan.

00:17:41: They can do your logo and all this stuff.

00:17:44: And you can start immediately.

00:17:47: And what do you think about this future?

00:17:50: Do we see much more entrepreneurs because it's easier to go to market?

00:17:55: Yeah, I think that the costs and the power structure of entrepreneurship has completely changed over the last few years.

00:18:02: It used to be, it was all about the VC.

00:18:06: They were the pickers.

00:18:07: They get to choose who's the winner and they get to choose who they give the money to.

00:18:11: What we're finding now is that it doesn't take much money to start a business now.

00:18:15: It doesn't take much money to get things prototyped, to start deploying because of the capabilities and the efficiencies.

00:18:24: that you get from AI systems, from outsourced cloud, from online giant databases and sources of customers.

00:18:34: And so when that happens, it's really the entrepreneur who has a real idea, who understands the market, who understands the problem, and has the drive and tenacity to stick with it.

00:18:47: And they now have the power because there's a lot of money now chasing deals.

00:18:52: So if you are able to come up with an idea that really makes sense and then execute on it, it really empowers people.

00:19:02: And like I said earlier, I don't think it's just a function of being young anymore because a lot of people say, oh, it's the young people that have the advantage.

00:19:11: I actually think that if you've got five, ten, fifteen years of experience in industry, you're probably more in the position.

00:19:20: to leverage these technologies with maybe a technical founder, then a fresh out of the school person because they may understand the technology and the tools, but they don't understand the market.

00:19:31: They don't understand the customer pain and how to sell and how to deploy.

00:19:37: And these are things that can only come with actual real world experience.

00:19:40: The problem is for a lot of people.

00:19:43: People ask me, Jens, what can we do to use agents?

00:19:48: What is the first step?

00:19:50: What do we have to read to look?

00:19:52: How can we manage it that agents will support us?

00:19:56: What do you recommend these people?

00:19:59: Like I said, I think it's very difficult for one person to be an expert in everything.

00:20:03: So rather than saying, hey, you're going to take a weekend course and become an expert in deploying agent technology and build your own scaffolding and become an expert MCP.

00:20:14: Don't do that.

00:20:15: Just go find a technical person you trust that is capable.

00:20:20: You guys work together.

00:20:21: Let them do what they're good at.

00:20:23: You do what you're good at.

00:20:25: And I think that is actually a better approach to starting a business today than saying, I want to do everything myself.

00:20:34: And if you have a good business, whether you have fifty percent of the business or a hundred percent of a business, a hundred percent of a failed business is not worth anything.

00:20:42: So we can send to the people your email address and you will fix it for them, all right?

00:20:47: Well, I think we're happy to take comments and feedback from any of the viewers, but I am a little bit tied up working with you during the shows.

00:20:56: Yeah, that's true.

00:20:58: So, in four weeks, you know what will happen in four weeks, more or less?

00:21:02: I think we have a holiday coming up.

00:21:04: We have a holiday.

00:21:05: Yeah, we have Christmas and if we look at this Christmas, I think it's almost more or less the same.

00:21:11: But if we look in ten years, how do Christmas will change?

00:21:16: because of these technologies are coming up and we will have in every life?

00:21:22: Do we have a change?

00:21:24: yeah so i mean ten years is a very long time as you know.

00:21:27: with today's pace and you know usually i tell people in the next ten years you're going to find the world would change as much as has changed in the last hundred years.

00:21:35: so if we look back at how how was life back in nineteen twenty five you know you know essentially pre the depression and you know essentially pre tv's preese, rockets, what did that look like?

00:21:49: I think we're going to find that the world can change that much.

00:21:54: I'm actually a little bit torn because I feel like ten years from now it could be super amazing.

00:22:00: We could all be essentially working one day a week or maybe less and have the time to go and pursue music or art or travel or take care of our children more.

00:22:13: Or we could potentially be in a world where we have a few oligarchs that rule the world.

00:22:19: And they were the ones who made the first models that then used to create trillions of dollars a profit, which then they use to then control the government.

00:22:29: And I think that those two are very both possible.

00:22:34: And it really depends on how our governments in the near term react to this.

00:22:41: technology is coming and whether we see it as something that is a public good that we should share with the world or whether we look at it as a tool for personal gain or for national dominance.

00:22:53: If we look at it that way, then we could actually end up in a very sad place in ten years.

00:22:59: This

00:22:59: is a big picture, but for a lot of people, the small picture or being together with the family, do you believe that In ten years I can have, I can drink a glass of red wine, I can eat meat?

00:23:11: or is there somebody who is saying to me, don't do it because you will lose your insurance, your health insurance because don't drink alcohol, don't do silly things like that.

00:23:23: How do you see that?

00:23:25: I mean, I think if things go well, in ten years, we could have a robot in every home helping you make that holiday dinner and then serving you the wine.

00:23:35: Now, whether or not you choose to say, hey, you should have less or more red wine or less or more meat, I think that's always going to be up to the individual.

00:23:45: There's no way whether a country or an AI is going to be able to go into your home and stop you from drinking or having fun.

00:23:55: But I think from a personal perspective, we probably will become healthier if the AI is there to suggest something, to say, hey, it's OK for you today to have an extra glass because it's the holiday.

00:24:04: But maybe tomorrow you should go take a jog.

00:24:06: Maybe you should go swim a few laps.

00:24:09: And I think that's not necessarily a bad thing.

00:24:13: I wear a smart watch to tell me my sleep schedules and how much steps I have because I want to be able to monitor it and stay healthier.

00:24:23: AI becomes smarter and they see their position as trying to be your advisor, then they will give you suggestions.

00:24:30: And we always have the agency to choose whether or not we take that advice.

00:24:35: A few months ago, somebody told me a story that if you will use your car, the car will tell you, first of all, you have to do some push-ups and then you can do use the car because it's part of your insurance.

00:24:49: Well, I mean, they do have cars for people who have been drunk driving.

00:24:53: They'll put little breathalysers where you have to blow into it to make sure you're not drunk before you can drive.

00:24:59: And for certain people, that makes sense.

00:25:03: But I think in ten years, hopefully, we may even not need to drive anymore because we have autonomous driving.

00:25:08: that will help us do a lot of this work.

00:25:10: So there's amazing technologies and amazing future ahead of us.

00:25:14: If we let it, I think that's the key.

00:25:16: If we let it, we have to make sure that the technology, the perspective we have on technology is that it has to be a public good.

00:25:23: It has to be something that lifts everyone instead of helping a few people become ultra-rich.

00:25:28: Thank you very much, Alvin.

00:25:29: Thank you for sharing these insights.

00:25:31: This is Big Bang Future Lab, where we explore the breakthroughs defining our future.

00:25:37: We look forward to seeing you in the next episode.

00:25:41: Yeah, and don't forget to like and subscribe.

00:25:44: And we will read all your comments.

00:25:45: Please tell us what you would like us to chat about.

00:25:47: Thank you.

00:25:48: Bye.