Anthropic’s New Model, the Robot Economy and Apple Health

0:00The most effective people at work aren't working harder than everyone else. They're working smarter, inside better systems. Superhuman Go, from the makers of Grammarly, is the AI assistant that works inside every tool you already use, always ready, already aware of what you're working on. It's a teammate whose only job is to help you be better at yours. With Go working with you, you can show off what you do best. See what Superhuman Go can do at superhuman.com. Some people treat ChatGPT like some kind of smart search engine, and some use it to get work done. ChatGPT Work is a new way of working in ChatGPT that can take action across your apps and files, stay with a project for hours if needed, and turn a goal into finished work.

0:46It's designed to help you move from a chaotic starting point to a reviewable first version. So all the source materials, briefs, and scattered information that you have to grind through to turn into something useful can just become something useful. Put ChatGPT to work on your most ambitious ideas and projects. Get started at ChatGPT.com by selecting Work Mode, available on Plus and Pro plans. This is Robert Smith from Business History. If you're listening to this, there's a good chance you're a small business owner. And like every small business owner, you started with a dream to do what you love and watch it grow.

1:21What you probably didn't dream about? Keeping up with cyber threats. That's where MasterCard can help, with access to tools that help identify cyber threats to better protect your business. Building your dream business? Priceless. For cybersecurity in a changing world, there's MasterCard. Learn more at MasterCard.com slash small business. Bloomberg Audio Studios. Podcasts. Radio. News.

1:54Bloomberg Tech is live from the heart of Silicon Valley with Ed Ludlow in San Francisco. This is Bloomberg Tech coming up and Thropic out with a faster, cheaper AI model as the startup tries to stay ahead of its rivals before its IPO. Plus, we speak with Adam Jonas of Morgan Stanley about SpaceX, Tesla and the robotics boom and what it means for U.S. jobs and manufacturing. And Bloomberg reports Apple aims to take aim at Whoop with its own screenless health and fitness tracker as the company adds new wearables technologies. I want to get right to our top story. Anthropic out with a faster, cheaper AI model as competition heats up ahead of its potential IPO.

2:37This is Claude Opus 5.5, which promises performance similar to its most powerful Fable model, but at a significantly lower cost. Bloomberg's Rachel Metz has the details. Interesting timing on the release of this. What do we need to know here? Yeah, so Anthropic, what they're doing here, and we've been seeing this increasingly lately, they're releasing a model that they're saying is very capable, but it's also more cost efficient. We've been seeing as more and more companies are using AI from companies such as Anthropic, OpenAI, Google, they are realizing, wait, this can cost a lot of money. And some companies have even imposed limits on how much they're allowing people to do work.

3:18So this is an attempt, as you noted before, the IPO to say, hey, we can be a little more cost conscious with our technology. Compared to the prior generation Opus 5, they're saying 40 % cheaper. Pricing is$4 per million input tokens,$20 per million output tokens. Explain why the economics of that matters when a generation of models release like this. Sure. So when you're talking about tokens in and out, you're talking about things that you're putting into a model like a prompt versus what the model is generating. And these costs, they may be lower, they may be different. I think it really also is important to keep in mind that it depends on what kind of tasks you're doing.

4:02Some tasks are going to be more computationally heavy than others. So results may vary depending on who the user is and what they're doing with it. it's not as simple as just looking at a chart to figure out, oh, this will cost me this much less or this much more. Rachel, I want to go back to the timing of this. How normal is it for Anthropic to release different generations of Claude with some cadence? The backdrop is obviously, in the last two weeks, the essay from Anthropic CEO Dario Amodei about pacing the frontier, but they've just released a new model, which there's an emphasis of performance on.

4:38Sure. So, I mean, as you noted, there's a few things going on there. They've got the IPO looming, like in the very near future, as we and others have reported. Dario Amadei also said, hey, we should pace the release of the most cutting edge AI. What they're doing here is releasing a model that is not comparable to their most capable publicly available models right now. So it seems to fit in with those kinds of things. Both the IPO wanting to show investors, hey, we're still doing really important work here and we're thinking about our customers, but also with pacing how quickly they're releasing most cutting edge things.

5:20Boombo's Rachel Metz on Anthropics' latest generation of core model. Thank you very much. Sam Altman is taking his AI safety push to the United Nations just a day before he's set to attend President Trump's state dinner for Chinese President Xi Jinping. The OpenAI CEO is set to call for global AI standards today as the US and China kind of grapple with how much to manage the risks of increasingly powerful AI. By the way, Anthropics' Dario Amodei will also join that meeting via video feed. Bloomberg Shringafari is here with the details. We have a sense of Sam Altman, what he will say, what he will tell the UN Security Council.

5:59Try and summarize it for us. That's right. It's in line with what Sam Altman and OpenAI have been saying publicly over the past few weeks, if not months, which is basically this idea of having some kind of global framework to set standards on AI safety would help AI companies make sure that their products are indeed not going to cause great harm to the world. And so, of course, the UN being an international body is an apt forum. The company thinks to present these ideas. That being said, we understand he is not planning to go so far as to name a single body that should be doing this specifically.

6:37but there are several ideas floating around along these lines. And if you remember, DeepMind co-founder Demis Hassabis also has floated some similar ideas about having some INRA-like agency, but for AI that sort of sets some standards. That's right. 24 hours ago, one of the headlines on the Bloomberg terminal was OpenAI to let outside groups evaluate AI models at earlier phase. That's the idea of kind of independent evaluation, which Anthropic and Daria Amodei have also talked about. Let's zero in on Sam Altman. Do we have a sense today of where OpenAI is netted out in its vision of what pacing the frontier means?

7:21I think it's very much still up for debate what exactly pacing the frontier means or will look like tangibly. I think that's something that you have different interests. There are companies that are open source, companies that are closed source like Anthropic and OpenAI. There are smaller players, bigger players. There are different governments. So if you think about getting everyone on the same page, how does industry with governments, blessing or not, go about actually potentially slowing down development? I think that's going to be no easy feat. And no one has a very clear answer right now.

7:54I do think there's a lot of discussion and clearly a lot of interest, anxiety, both within these companies, within their staff and from outside, including the UN, political leaders, etc., about if this goes wrong. So, ladies today, Sam Altman and Dario Amodei, although via video feed, will address the UN Security Council. Bloomberg Shereen Ghafari is the author of our Q &AI newsletter, which I recommend you subscribe to, and we'll be tracking that. Coming up in the program, AI is moving beyond the screen and into the physical world. We're going to speak with Adam Jonas of Morgan Stanley about this coming robotics boom.

8:29what that means for U.S. jobs and manufacturing, why the race with China matters in the context of physical AI. A news story this morning, take a look at shares of Disney now flat to lower. Bloomberg reported the company raising prices on several of its streaming subscriptions, all according to sources. The price of the flagship Disney Plus streaming service without ads is going to rise 13 % to$21.49 per month, an increase that kind of brings the premium Disney Plus product more in line with the similar plan that's sold by Netflix. The stock went to a session high after that Bloomberg report, kind of came back down now, just creeping into negative territory.

9:13This is Bloomberg Tech.

9:20AI has largely lived in the digital world. Chatbots, coding agents, general agents. Morgan Stanley's Adam Jonas says the next chapter is physical AI, robots, autonomous vehicles, machines operating in the real world. But the U.S. has a problem. China dominates much of the supply chain needed to build that. Adam Jonas is Morgan Stanley's global embodied AI and robotic strategist and is back with us on Bloomberg Tech after quite a long time. It's been so interesting to read your research in the month of September. I think that the question that I kept coming back to is how you see this as being so economically consequential for the United States, right, this move to physical AI.

10:01Yeah, thanks, Ed. We think historians of the future are going to write about this time now. Where were you when the robots came? Where were you when we really started to industrialize and make use of infrastructure in space? So the robotics, when you combine AI and robotics, we think that that's economically multiplicative, like an 8x or 10x global GDP before you and I retire. It's militarily deterministic and potentially socially destabilizing. So we have to get this right. In the process, we think the story of physical AI and robotics is the story of on-shoring and rebuilding American manufacturing resiliency for the first time in many, many decades.

10:48And it's going to bring a lot, I think millions, of human jobs along with it. So to make this artificial life, Ed, we need a lot of vocational and even mid and high skilled labor in the U.S. and with our economic partners. We're talking about this from a thematic research point of view. And when we have your industry colleagues on, they say two things. China is ahead in deployment, real world production, be that humanoid robots or autonomous vehicles. But the supply chain element is a big point of focus. Where is your research netted out on that part of China's story? It's a fact. credit to China for having a leadership in the manufacturing ecosystem, the supply chain, all the way up to critical minerals and materials, rare earths.

11:38I know your program's talked about those things. You can't have a robot, you can't have a data center or any sophisticated electronic product without China super, super involved in the ecosystem. And that's not going to change overnight, right? So again, part of the paradox is in order to make robot life, we need human jobs. And in order to diversify away from China, to create a little more balance, we're going to need to continue to work with China. In our simulations and scenario analysis, there's no real credible way you could build a wall and just wall off China from the US ecosystem. I mean, if you think inflation is rough now, just imagine what it would be if we attempted to do that.

12:28So I actually think the story of physical AI and onshoring in a way can lead to a, yes, re-architected and still sensitive, but we think a very prosperous entanglement between U.S. and China relations. Okay, so what role does regulation play in this United States physical AI effort? Obviously, this week you have the UN General Assembly. Since the Pacing the Frontier essay from Anthropic, we've obviously looked at regulation of AI from the model side and the software side. Have you modeled for, pardon the expression, but modeled for a regulation outcome for physical AI from the United States? It's deterministic.

13:13In our base case, there's no, again, plausible scenario where we don't have physical AI very regulated. It's not binary. We can debate whether the regulation, how the regulation phases in and is sequenced in, whether it's informed by industry experts, elected officials, probably a combination of both. probably in response to things that are progressing in unpredictable ways as we're seeing over the last couple of weeks. Those same concerns do pass on into physical AI because of any robot with an AI brain attached to it as an independent agent, it does have a dual purpose. Like any piece of super advanced technology, there is a potential military element You're seeing it in the campaigns in Ukraine and Russia.

14:10You're seeing it in the Straits of Hormuz as well. And the topic, again, addressed by your show of how our defense ecosystem is also in need and currently in process of being re-architected in a physical AI world. That kind of stalking horse of China on the competitive side and potential deficiencies on the military side is creating a sense of urgency that, yes, we will have regulation. But in our opinion, the regulation will change the form and the cadence, but not the destination. We are going to have tens of billions of robots in our daily lives within the next one to two decades. the company case studies of embodied AI for you Adam are SpaceX and to a degree Tesla and you wrote on Sunday that Tesla and SpaceX have separate but synergistically linked physical AI capabilities and you go on to write the companies are working to gradually close the gap between them inevitably the question that our audience puts to you is your scenarios and your base cases for the merger of SpaceX and Tesla under that assumption?

15:23Sure. It's a fair question. Elon gets that question a lot as well. I won't be able to comment on a specific probabilities of a transaction between two publicly traded companies, but the relationship seems deterministic at this point. Tesla, as a manufacturing and data collection layer, They make the robots and can scale that intelligence into the physical world. SpaceX is the connectivity in the AI layer, also having very impressive manufacturing chops themselves. And Elon and others, and some of our work as well, working with our colleagues across the auto team. Andrew Bercoco, who covers Tesla, and our internet team, led by Brian Nowak and the internet team, collectively working on what's that next iteration, regardless of strategic transaction.

16:20We think the two companies are investors should expect continued cooperation as the shared mission is essentially converting energy to intelligence at scale to achieve the most efficient intelligence per watt per dollar per second. That race of intelligence and density of energy and then getting that time to power quickly. We think both companies bring something important to the table. I think like a lot of people are asking themselves what the end result be, the net result, right? So if you take Optimus, a humanoid robot, I guess the question for you, Adam, is like, does Optimus work? Does it exist in the real world in industrial settings or in the at-home setting that they've also mentioned, without Starlink connectivity, without inference at the edge powered by Orbital Data Center, how much is one a precursor to the other?

17:18Well, we think any machine that can be automated will be. Any machine that can have an AI edge inference computer in it will. Any machine that does not may have a much more rapid obsolescence curve, if not really have any useful life at all. The humanoid form factor that you bring up gets a lot of attention. We think it's more of a recruiting tool and a capital raising tool. In our modeling of our global robot model out to 2050, it is an important form factor. We do get into the billions of, in many of our scenarios, of humanoids. But we think it's just one of maybe thousands of different stratum of low-altitude robots, terrestrial robots, autonomous vehicles, mobile robots, industrial, and everything in between.

18:05And yes, the connectivity connecting those robots together in a swarming intelligence that could do potentially the vast majority of the world's inference work. Elon refers to this as the distributed inference cloud connected to the distributed orbital cloud of the space-based AI infrastructure. We're still going to have data centers for the foreseeable future, but we like to say at Morgan Stanley, using a data center or a GPU for an inference task is like using a Ferrari to pick up the milk. We need to start to activate more inference at the edge, kind of like neurons in your brain, so we can do the compute vastly more efficiently than me speaking into a phone and having a rather simple query generating tokens somewhere in Memphis.

18:56Your job is to be the global embodied AI and robotic strategist at Morgan Stanley. Your history was more an origin, it's like more focused on the automotive part of Tesla. I just would love to understand where Tesla as an embodied AI company, separate from whatever happens with SpaceX, ranks in your worldview of that industry against all of its peers in China, other players here in the United States? We think Tesla, not singularly, but Tesla and Tesla and SpaceX represent one of the best chances to help revive and give Western industry a chance to keep up, if not to surpass our geopolitical rivals over time in robotics and physical AI.

19:46Tesla specifically right now, I'll tell you today, I was driven from Westchester to midtown Manhattan in my Tesla, and I did not have to touch the wheel. Interestingly, I was delivering a robot dog made by a Chinese company. I brought it back to my office to return it to the company we borrowed it from. We think autonomous cars are solved. And this was something that even Apple and a lot of the physical AI propagators have said was the world's hardest problem. In our opinion, when the historians of the future write about when did we solve autonomous cars, it was 2023 when Waymo pulled the driver from Phoenix.

20:29So is it to the level of safety? Do we have enough nines to make it as where we want it to be? know that's going to continue. But we think that on that modality, if we solve that and we have insurance companies start to give you a discount for letting the car drive, that will open up the floodgates for other form factors to follow. I'll just say very quickly, I'm a daily FSD user. I do about 60 miles a day. I log regularly. I log regularly when weird things happen and they're on X. We're out of time, but Adam Jonas and Morgan Stanley back on the network, back on the show, really grateful for your time.

21:01Thank you very much. Coming up, Meta finds its muse. The new AI assistant is climbing the charts, but can it win users' trust? This is Bloomberg Tech.

21:15The people who seem to get more done than everyone else, they're not working longer hours or running on more caffeine. They've just stopped wasting time on the stuff that doesn't move work forward. Switching apps, re-explaining context, hunting for files. Those aren't small inefficiencies. They're hours wasted every week. Superhuman Go gives you those hours back. From the makers of Grammarly, Go is an AI assistant that sits inside every tab and tool you already use. Always available and already aware of what you're working on. Ask it to draft something, summarize a long thread, pull up a file, or prep you for a meeting.

21:56Go handles it without you ever leaving the page you're on. This is what it looks like when AI actually fits into your work, instead of adding to it. It's like having a teammate whose only job is to help you be better at yours. Go keeps up so you can move forward. With Go working with you, you can show off what you do best. Superhuman Go. Find out more at superhuman.com. That's superhuman.com. Some people treat ChatGPT like some kind of smart search engine, and some use it to get work done. ChatGPT Work is a new way of working in ChatGPT that can take action across your apps and files, stay with a project for hours if needed, and turn a goal into finished work.

22:39It's designed to help you move from a chaotic starting point to a reviewable first version. So all the source materials, briefs, and scattered information that you have to grind through to turn into something useful can just become something useful. Put ChatGPT to work on your most ambitious ideas and projects. Get started at ChatGPT.com by selecting Work Mode, available on Plus and Pro plans. This is Robert Smith from Business History. If you're listening to this, there's a good chance you're a small business owner. And like every small business owner, you started with a dream to do what you love and watch it grow.

23:14What you probably didn't dream about? Keeping up with cyber threats. That's where MasterCard can help with access to tools that help identify cyber threats to better protect your business. Building your dream business? Priceless. For cybersecurity in a changing world, there's MasterCard. Learn more at MasterCard.com slash small business. Let's talk about health care for a second. It doesn't always work the way people expect it to. If you've ever waited on a prescription refill or had a hard time getting the care you needed, you know the feeling. The system should just work better for everyone.

23:48That's exactly what the people at Optum are trying to do every day. They're a health care company linking patient care and pharmacy services and using data and technology to drive the whole system so care is connected, not complicated for patients and providers. Things like making it easier to get care that looks at the whole person, from primary care doctors to mental health support and even in-home care, and then using technology to make sure they all work together. Technology designed to help doctors spend less time on busy work and more time with their patients. And those prescriptions? Optum is working to bring costs down, save patients money, and make it easier to get refills.

24:26Little by little, Optum is helping make healthcare work as one for everyone. Head to business.optum.com to see how. It's time now for Talking Tech. I'm Yahaira Anand. First up, Alibaba is accelerating its data center expansion in Europe and the Middle East. The company will set up its first cloud regions in Turkey, Finland, and the Netherlands over the next 12 months and expand its data center footprint in several other countries. Plus, Australian AI startup Firmus is in talks with lenders for a roughly$10 billion financing deal to buy NVIDIA chips for its Indonesian data center. The financing would add to Firmus' expanding capital base ahead of a planned IPO in late October.

25:11And sticking with Australia, the U.S. government criticized the country's plan to clamp down on social media algorithms and harmful content online. In a written response to the Australian law, the U.S. embassy in Canabara said it risks imposing, quote, disproportionate compliance burdens on American companies. Ed. Thank you, Yohara. Let's stay on the topic of social media. Meta has finally an AI hit on its hands. Its new Muse personal assistant has shot to the top of the App Store charts and helped send Meta shares higher now at a September high. But getting people to hand over access to their email, calendars, other sensitive information, that could be a much bigger challenge.

25:52Bloomberg Sarah Fryer leads the team covering the company and the tech. And we've been writing about it in the Tech in Depth newsletter on the equities team. Outline that kind of balance. Well, it's really fascinating, first of all, to see Meta come out with a consumer product that people are gravitating towards. It's a huge win for them. I've watched this company try to put out a popular product with sticking ability for a long time. And here we have it. The question then is whether it has the ability to bring a broad audience in. And Meta still, as you note, has serious trust issues that it's digging out from under.

26:34So the product needs to be good enough not just to get people to try it, but to get people to get over whatever they think about Meta and its security with their data after all of these years and decide that the convenience is worth the risk. So this is an AI agent where you can give it access to, say, your email, tell it to get rid of all the spam, but you need to give it the access. And Riley writes about some examples in that real quick. Yeah, she writes about getting it to help her with prescriptions for her doctor. There are others who have used their assistance to book flights for them or handle a customer service issue.

27:18That means you might need to get it to use some of your online shopping accounts, maybe have access to your credit card information. It's a lot of things that need to be taken into account. What's interesting is we're seeing the market react because there is thought that people will take this on, if not from Meta, then maybe from others. And then some of these companies that rely on customers not making changes could get hurt. Bloomberg, Sarah Fryer, thank you very much. Half time in the program. This is Bloomberg.