Silicon Valley Power Reshuffle: Apple, Google, Anthropic, Meta, OpenAI

00:00:00: I think that the robots is a little bit premature, but the Europe agents where we move beyond a conversational chat model to having an action-based model for the relationship between humans and AI.

00:00:13: I think that this is the year when that starts to happen.

00:00:18: Hello and welcome to a new episode of the Big Bang Tech Report.

00:00:22: Today's guest is on the way to Davos, where... world leaders, entrepreneurs, and technologists will gather to talk about the future.

00:00:32: He's one of the most recognized voices in global AI, Alvin Greiland, a speaker, strategist, and advisor who helps shape the international conversation around artificial intelligence.

00:00:47: Yes, it's been a couple of weeks since we last chatted, so there's a lot of stuff going on.

00:00:51: I look forward to our chat today.

00:00:54: Thank you.

00:00:55: What is your message for this year's World Economic Forum?

00:00:59: Yeah, so I'm actually lucky enough to be on seven different speaking sessions this week, this next week.

00:01:05: So there's a lot of things going on, as you know.

00:01:09: The world is in a very strange place.

00:01:12: And unfortunately, I think a lot of the people who are making decisions are just not informed enough about what's going on with the technology and how it's going to affect it.

00:01:21: So my key message to them is really We really need to reduce the competitive mindset between countries and also between companies because at the end of the day, we don't live in a zero sum world and we're behaving like we do and if we keep doing these things of fighting over a mythical race, we're going to to create a negative outcome for both the countries involved, the companies involved, and humanity as a whole.

00:01:50: So there are just a lot of things that we need to do to prepare for what's coming and to actually work together.

00:01:58: But right now the world is going to the other side.

00:02:01: If you look at Greenland, if you look at Venezuela, Russia, Iran, so right now the world is to make America great again, make, I don't know, great again.

00:02:12: So

00:02:13: this is exactly the problem is that I think the people who are making these decisions are still running on rules of the prior century.

00:02:23: And what we're creating now is technology that can take us into the long-term future.

00:02:28: So unless we can change their minds and these are the right people are going to be there, there's going to be sixty to eighty state leaders there.

00:02:35: Hopefully they will hear the message and start to at least leave a seed to think about doing things differently.

00:02:43: Let's look back, let's look at the start of this year with CS and Las Vegas, where NVIDIA CEO Jens Wang made headlines with one bold prediction.

00:02:54: Twenty-twenty-six will be remembered as a chat GBT year for human art robots.

00:03:01: Do you agree with that?

00:03:03: I would have to say that I think humanoids are going to take a little bit longer, probably a lot longer for it to be... totally relevant, particularly in a non-industrial perspective.

00:03:20: I can understand why he's talking about it, because right now there's more and more competition for the GPU side.

00:03:26: So creating a new realm where he can compete in terms of creating physical AI, I think it's a great narrative for NVIDIA.

00:03:37: But with anything physical, it takes time to develop, it takes time to hone, and it takes time to make safe, and it takes time to get the regulatory approvals for it to get out to market.

00:03:50: And there just is not as much training data as there is for digital AI.

00:03:58: which means that it's going to take longer for it to mature.

00:04:01: So all of these things are workable and it's going to be very important, but it's going to definitely not happen in twenty twenty six.

00:04:10: A friend of mine, he was at the CES from Germany and he was not really happy.

00:04:14: He said, okay.

00:04:16: The robots are at seventy percent, sixty percent or something like that, because once they can do something, but then when they have to change, they have to do another thing, they can't do it, because they can't take cards with hands and all this stuff, you know.

00:04:32: It's there, but it's not complete.

00:04:35: I think we can see the progress on the physical, just the physical forms, and you see how capable they are, how strong they are.

00:04:43: Samulam can climb and jump, and it looks very impressive.

00:04:48: But the underlying logic for them... these vision language action models that are associated with the robots, they're not ready yet.

00:04:58: And the data sets that we have to train them is a fraction of what we have to train the language models.

00:05:05: So again, it's important, but it's going to take a lot of time.

00:05:10: And it's going to take a lot of time to build the data, because the data that we have today is not the right type of data for what it needs to do.

00:05:20: So let's go through a few major stories from the AI world in just the last few weeks.

00:05:27: And I won't jump in if I forget something.

00:05:30: So I made two strong moves.

00:05:33: First, they acquired Manos AI, a startup focused on autonomous agents, and they are from China.

00:05:40: What is that?

00:05:42: So Manus actually made a big news middle of last year when they started to create a very impressive agent framework and it's actually interesting.

00:05:51: this is the first time that a major US company has actually acquired or is trying to acquire a Chinese firm.

00:05:59: Although the Chinese firm, they moved to Singapore to avoid some of the political issues.

00:06:05: But right now, it looks like China might actually be blocking the acquisition, just like what America has done for multiple Chinese related acquisition type.

00:06:20: in the past.

00:06:20: So it could actually become a more political issue.

00:06:24: But it makes sense what they're doing because agents are going to be increasingly important and Manus is one of the major players.

00:06:31: And this is also an area where meta is actually significantly behind.

00:06:35: And then they bought the limitless pendant, a very good memory device for AI supported productivity.

00:06:42: So what's going on with Meta?

00:06:44: Well, I mean, just like Meta a few years ago, they missed out on the phone.

00:06:51: And so they became captive to Google and to Apple in terms of the interface to the consumer.

00:06:57: They wanted to now find another way, another device.

00:07:01: They try to get the VR devices and they're doing glasses.

00:07:05: I think the pendant is a... a potential competitive area where you have a direct interface to the consumer.

00:07:13: OpenAI is trying to do the same thing right now.

00:07:15: They've also announced multiple devices that are coming out of their cooperation with Johnny Ive because of the same reason.

00:07:23: They don't want to be captive to the major hardware players.

00:07:27: So I think strategically it makes sense.

00:07:31: I've actually been using the Limonless Pendant for almost a year now.

00:07:35: So it's actually disappointing for me that they got acquired because now I have to decide whether or not I keep using because I really don't want to have all of my conversational data.

00:07:46: What is it?

00:07:47: What does it do?

00:07:48: Because I'm not used to it.

00:07:50: Actually, what it does is it transcribes all your conversations and then it helps give you notes.

00:07:54: You know, essentially gives you perfect memory to recall whatever you talked about.

00:07:58: And also, you know, ask it, you know, how could I have been a better speaker yesterday when I talked to Jens?

00:08:05: Or, you know, how could I have been a better father to my child in my conversation with her?

00:08:10: So it's actually a very useful tool.

00:08:13: I did find it very useful.

00:08:14: But again, I'm now reconsidering how much I'd use it.

00:08:18: How do you wear it?

00:08:20: It's an appendix.

00:08:21: It's essentially, yeah, it's, you know, I wear underneath my shirt and it just, it's just always on and it holds my conversation.

00:08:30: So I can go back and look at things I talked about six months ago and I'll remember what I said.

00:08:35: Wow, not bad at all.

00:08:36: Okay, and you can buy it everywhere in the US?

00:08:40: Yeah, it's been available for a while, although now that they're acquired, they stop selling it.

00:08:48: I think they said they'll give you one more year of support on it, but it's probably going to go away.

00:08:53: Okay, so another huge deal.

00:08:54: NVIDIA is acquiring GROC, a company known for building ultra-fast chips for AI.

00:09:00: How big is this for the future of AI infrastructure?

00:09:03: No, that's actually a very big deal, because NVIDIA has really made its moat in terms of the training side of the business.

00:09:12: This is where all the data center chips are focused on, because that's where it can do better than anybody else.

00:09:19: There has been a lot of talk saying, hey, if the world future moves to inference-based compute, training becomes less important.

00:09:26: And this is where GROC is actually specialized, is that they make special chips that are focused on making inference better, faster, and more power efficient.

00:09:35: So the fact that they're acquiring this company gives them essentially another weapon for them to compete against the other chip vendors that are more focused on inference.

00:09:49: So I think strategically it makes a lot of sense.

00:09:53: They're not even doing a real acquisition, they're doing kind of a partial licensing slash acquisition to avoid some of the antitrust issues.

00:10:02: But it makes sense.

00:10:04: We'll see how the integration works out.

00:10:06: And also, if there's any kind of antitrust issues that the government puts up.

00:10:12: So for you it's a smart move.

00:10:14: and what's the next move?

00:10:15: A smart move too.

00:10:17: What about Apple?

00:10:18: Apple has said that it's choosing Google's Gemini AI and that it's not working with OpenEye together.

00:10:27: So what about the new iPhone?

00:10:29: What about the future of Apple?

00:10:30: They are struggling a lot with the AI, with Siri and all stuff.

00:10:36: They are not a frontier model, right?

00:10:38: Yeah.

00:10:38: I mean, I think that's been kind of the big criticism for them for the last couple of years, because a year and a half ago, they were super big noise about Apple intelligence and they never delivered.

00:10:49: In fact, they're kind of invisible today on the large language model.

00:10:53: But in some ways, I feel like they may not have missed a boat and what they're doing right now with Google makes a lot of sense and both economically because they didn't waste a hundred billion dollars like some of these other companies you know building giant data centers or or training models that become commoditized.

00:11:11: they're essentially going to kind of a best-of-breed solution and then having a a custom support from Google.

00:11:19: This also gives them time to maybe then build their own solution.

00:11:23: But I think that the key difference between what they're doing versus what OpenAI is doing is that they have control right now of the interface to the consumer, which means that they can decide how much data to make visible to the AI, which is something that I think OpenAI will find increasingly difficult.

00:11:44: And this is probably also the reason why they decided not to work with OpenAir, because OpenAir is actually now creating competitive products to Apple or trying to.

00:11:52: And so they're having a much more competitive relationship, whereas Google and Apple has been working together for a long time on the search side.

00:12:03: So they already have a good kind of B to B working relationship.

00:12:07: So, I think, again, it's a smart move.

00:12:10: It will make Apple relevant again, and it actually will allow Apple to deliver services that other vendors like OpenAir or Grog or Anthropic cannot do on the consumer side.

00:12:24: From the US, let's go to China.

00:12:27: We are hearing rumors about DeepSeek V-Four.

00:12:31: So some say it could be a major leap forward.

00:12:35: Avan, what can you tell us about it?

00:12:37: Yeah, so they're expected to launch in February, so right before Chinese New Year, which is about a month from now.

00:12:45: And apparently the rumors are there is testing very well and maybe outperforms Claude, which is the leading coding model.

00:12:55: They've actually specialized this version very much on coding and their coding and coding agent supposedly is very high-performance.

00:13:02: And the interesting thing is it also is completely trained using... Huawei chips, which means that this now is another chip that now displaces the need for NVIDIA.

00:13:18: We talked last time about how the Gemini chip, the Gemini model from Google was trained on TPUs.

00:13:25: Now you have Deepsea being trained on Huawei.

00:13:29: In fact, I think the GLM model from from Z.ai just recently released, and they also said that they trained on Huawei chips.

00:13:39: So this really shows a bifurcation in terms of the entire AI stack of the need for an NVIDIA stack or a non-NVIDIA stack.

00:13:52: And it's an interesting trend that we're going to have to watch this year.

00:13:56: Okay, we will focus on it and see what's going on.

00:14:01: So let's get into a hot topic, AI agents.

00:14:05: We hear a lot about digital assistants, but agents are supposed to go further, they act, they plan and execute for us.

00:14:14: When will we see agents that really book our trip, manage projects, or negotiate for us with little or no human help?

00:14:22: Yeah, I think this is coming very soon.

00:14:26: In fact, this week, they just announced the cloud co-work, which is an agent framework for the anthropic cloud system.

00:14:40: So the last month or so, code, which is the agent framework for coding, has been super popular among the developer community.

00:14:50: And people are starting to use that agent outside of work, outside of development.

00:14:56: And the interesting thing is, one of the developers at Cloud, at Anthropics said, hey, you know, why don't I develop something that can help this idea be used outside of developers.

00:15:07: And he asked Cloud code to build this framework for him.

00:15:11: And it took a week and a half.

00:15:13: The agent actually built the entire solution, and then they released it as a product called co-work.

00:15:20: So now it's only available right now to premium users, but this is actually very interesting because this will now essentially give very high-performance, long-term tasks make it available to the consumer class.

00:15:37: And maybe also for business users, for non-coding users.

00:15:40: And then just yesterday, Google just announced what they call personal AI.

00:15:47: And that also is available right now just to the premium users.

00:15:51: But that now allows you to have an agent that goes into your photo library, your email systems, your YouTube histories, everything on your essentially Google ecosystem to better understand you.

00:16:09: and then to be able to give you very highly personalized service.

00:16:13: I think this is essentially changing a person's relationship with AI now on a personal level.

00:16:21: So essentially solving the kind of things that you're talking about.

00:16:24: So if you want to book a trip, you could do that.

00:16:26: If you want to analyze your communications to you to outside world, you can do that.

00:16:31: If you want to do a psychological analysis on you based on your data, you could probably do that.

00:16:36: So it's getting very And just.

00:16:39: also last week was released the Universal Commerce Protocol, UCP, from Google, which now allows you to have direct agent to commerce transactions through agents.

00:16:54: and without human intervention.

00:16:56: So now you can actually do the type of purchases that you're looking for.

00:16:59: And they're making this a standard, which will go outside of the Google ecosystem.

00:17:05: So now together with the Anthropic MCP system from last year, now with this UCP from From Google, you put them together and it allows you essentially to connect to any app on the planet and to connect to almost any store on the planet and to be able to do real work and services to consumers and businesses.

00:17:27: This is why last year when I said this is the year of agents, I'm really seeing that coming true and it's only January right now.

00:17:36: Okay.

00:17:37: This is for you too.

00:17:38: A year of agents or what is this year for you?

00:17:41: Because, well, as we have talked, it's not for humanoid robots.

00:17:45: So what do you think?

00:17:46: Yeah.

00:17:47: So I think that the robots is a little bit premature.

00:17:50: But the year of agents where we move beyond a conversational chat model to having an action-based model for the relationship between humans and AI, I think this is the year when that starts to happen.

00:18:04: In fact, I was looking at all my schedule of the various hundred-plus meetings I'm having at Andavo's, and I was like, how am I going to schedule all of this?

00:18:14: And I just asked the AI agent to look through all these different events, find the ones that are most relevant to me, go sign up for me, and then put into my calendar.

00:18:23: And it generally worked.

00:18:24: So being arrested, what happens to the rest?

00:18:26: You cancel it.

00:18:28: The system is canceling it.

00:18:29: No, no systems are casting, but it's essentially putting together all the things that would have taken the hours to scroll through all of these events.

00:18:36: It finds the most relevant ones for me and then it highlights it for me and said this one's relevant because of this reason.

00:18:42: All you need to do is click submit to submit your application to join.

00:18:46: And so it makes my life a lot easier and then it adds it to my calendar so that I don't miss it.

00:18:51: So

00:18:52: this is happening today.

00:18:54: And in another few months, I think it will go much more beyond just being able to help somebody schedule your time.

00:19:02: I predict that you will be on every stage, or not as a visitor, you will be on the stage on every panel.

00:19:10: That's my prediction.

00:19:11: No,

00:19:12: because of the time, quick fire round, two short questions.

00:19:16: First, who is leading the eye race right now in the US?

00:19:19: Google, or may I meet someone else?

00:19:22: What do you think?

00:19:23: I mean, right now, I think on the coding side, Anthropic is leading.

00:19:28: On the consumer side, I think Google is winning.

00:19:33: you know, open as the market share, but their market share is actually coming down significantly.

00:19:39: And I think very soon people will start to be taking them as a competitor and start to make their life more difficult, you know, because when I just talked to you about the the request to do the review of the schedules and putting my calendar.

00:19:56: I did it in OpenAI.

00:19:57: It didn't work.

00:19:58: It found two events and then I said I can't open this file because it's a Google file.

00:20:02: I did it on Google and it actually worked.

00:20:04: So Google is actually blocking OpenAI from using their systems, which will make life very difficult for them to offer that next level of agent-based interactions.

00:20:15: Okay.

00:20:16: Second, do we have a personal dark horse?

00:20:19: A company people should watch in twenty-six.

00:20:22: Yeah, this is actually a very good question.

00:20:24: And I'll give you three dark horses.

00:20:26: Okay.

00:20:27: One is on the model side.

00:20:31: People should start keeping an eye on liquid AI.

00:20:34: It comes from my alma mater from MIT.

00:20:36: It's a small team, but they've recently raised a significant round.

00:20:40: And they are creating models that are one-one-hundredth the size of the large-language models, but they are very suitable for real-world operations in robots and control systems, and starting to also be able to run on phones so that you don't need giant data centers.

00:20:57: I think that's something that will be quite interesting.

00:21:00: On the China side, actually, the models there, three models that recently either went public or is going public or just did almost a billion dollar rounds are Minimax, Moonshot, and Z.ai.

00:21:17: All three of them are now very well funded.

00:21:20: They are all within the top two or three or four of the open source.

00:21:26: stack or a leaderboards and they I think will be a very important force in twenty twenty six in terms of making things happen.

00:21:35: and one thing that you know I didn't mention as much as is on the hardware side.

00:21:40: I actually think that Huawei is going to be a major player both from both from the chip offering.

00:21:48: They're already offering chips that are training these Chinese models, but soon they're going to be making chips that will export out of the country.

00:21:55: There's most people that supply a one point six million GPUs this year, which is sizable volumes compared to the last year.

00:22:06: And they have probably the best networking in the world between chips.

00:22:10: So they can do both the training side and the inference side.

00:22:14: So I would keep an eye on how they're doing.

00:22:17: Also recently, there was an announcement that China is now doing EUE machines and Huawei's got a big plan there.

00:22:23: So I think that the combination of their hardware and their ability to link with the Chinese infrastructure will make them a much more significant player in the twenty twenty six.

00:22:35: Evan, you're a Wikipedia of AI.

00:22:37: It's incredible what you know and what you can describe.

00:22:40: Thank you very much.

00:22:42: Thank you for your time.

00:22:44: You are right now in Seattle.

00:22:46: I am in Seattle, but I will be heading out in a day or two.

00:22:51: It's always inspiring to talk to you, to all our listeners.

00:22:56: Thank you for tuning in.

00:22:58: If you enjoyed the episode, send us a message.

00:23:01: If you didn't, send this feedback to Evan.

00:23:06: And well, I'm just kidding.

00:23:08: We both love hearing your thoughts.

00:23:10: And don't forget, join us this September.

00:23:14: It will be soon in a few months in Berlin at the Big Bang AI Festival.

00:23:19: And over eleven thousand guests will be there, more than four hundred speakers.

00:23:23: And we will talk about AI, AI, AI, and maybe about

00:23:27: robots.

00:23:30: Thank you very much and see you soon.

00:23:32: Yes, and thanks for these amazing questions.

00:23:34: You've done a really good job wrapping and I always enjoy our talks.

00:23:39: For our audience, please let us know what else you want us to talk about.

00:23:43: See you in a few weeks.

00:23:44: Yeah, please send me at www.dbuhr.de and so I will answer you immediately and I will send everything to Alvin.

00:23:59: Okay, thank you very much guys,

00:24:01: see

00:24:01: you soon.