Anthropic Launches iOS App and Teams as Altman Says He'd Burn $50B a Year to Get to AGI

Today on the AI Breakdown, Anthropic has released a new iPhone app as well as a new teams plan. Before that, on the Brief, Sam Altman says he'd burn $50 billion a year if it got us to AGI. The AI breakdown is a daily podcast and video about the most important news and discussions in AI. Go to breakdown.network for more information about our YouTube, our Discord and our newsletter. Welcome back to the AI Breakdown Brief. All the AI headline news you need in around five minutes. Well, Sam Altman is on a bit of a media tour right now. He's done a number of interviews on college campuses lately, and while there's nothing hugely new, it feels to me like there is a different tone, a sort of declarative, energetic, profound tone that's coming through in all of these interviews. A couple of the highlight quotes that have been pulled out by various media outlets reinforce Altman's commitment to getting to AGI, his continued emphasis that GPT4 ain't it, and a focus on AI agents as the future. Let's talk about this AGI commitment first. In one recent interview, Altman said, Whether we burn 500 million, 5 billion or 50 billion a year, I don't care. I genuinely don't. As long as we can stay on a trajectory where eventually we create way more value for society than that, and as long as we can figure out a way to pay the bills we're making, AGI, it is going to be expensive and totally worth it. Now this of course is coming from a guy who also was reported to be out raising $7 trillion for a network of AI chip fabrication plants. So ambition is not a particular issue here. Now, what about ChatGPT? Almond is clearly almost embarrassed about how underperformant it is versus where we'll end up. He said, Once again, GPT4 is the dumbest model any of you will ever have to use. Now, one really important thing though, from a technical perspective is that Altman also said that GPT5 is going to be a lot smarter than GPT4 and GPT6 is going to be a lot smarter Than GPT5, Altman said. We are not near the top of this curve and it is always going to get better. There have been a lot of questions given how long GPT4 itself or GPT4 class models have been state of the art. If we're coming up against some scaling limits, some prominent researchers have suggested that we are, but Altman is clearly planting his flag in a different camp. Additionally, Altman is clearly pointing to AI agents as the next big thing, the MIT Technology Review recently wrote. A number of moments from my brief sit down with Sam Altman brought the OpenAI CEO's worldview into clearer focus. The first was when he pointed at my iPhone SE, the one with the home button that's mostly hated, and said, that's the best iPhone. More revealing, though, was the vision he sketched for how AI tools will become even more enmeshed in our daily lives than the smartphone. What you really want, he told MIT Technology Review, is just this thing that is helping you. He described the killer app for AI as a, quote, super competent colleague that knows absolutely everything about my whole life, every email, every conversation I've ever had, but doesn't feel like an extension. It could tackle some tasks instantly, he said, and for more complex ones, could go off and make an attempt, but come back with questions for you if it needs to. Now, of course, a lot of AI hardware companies are experimenting with exactly this sort of thing, recording ambiently, for example, everything that you're interacting with or promising to be able to go off and actually do things for you. However, Altman isn't so sure. He said that while he thinks AI hardware will be cool, that for this AI agent future vision to come to fruition, quote unquote, I don't think it will require a new piece of hardware. As for juicy details, when one reporter asked if he knew when the next version of GPT was slated to be released, Altman simply said yes. Beyond these interviews, OpenAI just made a major update to their website as well. When you go to OpenAI now, the first top panel is a ChatGPT field. When you fill in a question and press go, it brings you to the normal ChatGPT app. But there are also three other top panels, which many feel is giving an indication of how OpenAI sees its priorities. Next to the ChatGPT box is a research link building safe AGI that benefits all of humanity. Next to that is OpenAI for business. Add the power of AI to your products, operations and workforce. Fourth, interestingly, is Sora. Given that this has just been launched, we don't know how much these panels will actually rotate and whether they really do indicate a hierarchy of priorities or just if they're there right now. For example, will Sora be rotated out for whatever next cool thing they announce a little later on? If you scroll down the page, there are sections for once again, ChatGPT, Research, Business and Developers, as well as a new story section that seems to be about use cases that are showing off how OpenAI is helping the world in a variety of ways. Finally, there is a news section. AI creator and educator Greg Camerot writes new OpenAI homepage. What's your read on what they signal with the new changes? Seems like a signal for a much larger consumer play, possibly getting ready to capitalize on the GPT5 priority. The AI for Business account responded, A lot of good thoughts here. Just want to call out the change in art style from bold colored shapes to soft abstract watercolors. Feels like they're moving away from just being a programmer's playground to being more accessible and humanistic. Good play for business acceptance. Now. Interestingly, we will talk about that in the context of Anthropic and their recent announcement in the main part of today's episode. Ultimately we don't know, but it does all feel like OpenAI is getting itself ready to make some moves at some point in the not too distant future. Speaking of use, we continue to see AI rolling out in a variety of different contexts. The latest is Sam's Club. The way that Sam's Club works if you haven't been there, is that you have to have a membership to shop there. When you leave the store, a person checks your physical receipt to make sure everything is kosher. Well, now they are rolling out an AI version of that check, having done so so far at 120 of Sam's Club's 600 locations. Lastly today, a new survey from Accenture around big New York City Fortune 500 attitudes towards AI and man in a world where AI skepticism is on the rise, this study goes the exact opposite direction. Accenture surveyed 500 C level executives in New York City and of them 97% believed that AI would have net positive impact on society. Those executives also said that they believe that the rise of AI skills will drive a trend towards skills based hiring rather than hiring based on, for example educational pedigree. In surrounding research, Accenture found that 63% of the hours worked in New York City could be either automated or augmented by generative AI. That is obviously an enormous percentage of people's work activity and shows the magnitude of this change. Anyway, really interesting to see such a positive focus study, especially compared to some of the other things that have come out. But that is going to do it for today's AI breakdown brief Next up, the main AI Breakdown Attention AI Breakdown listeners ConsenSys 2024 marks the 10th gathering for all things crypto, blockchain and web3. However importantly, this year's agenda will also dive deep into AI Driven Transformation, and the speaker lineup includes the leading minds and innovators at the forefront of this digital renaissance. Don't miss the Consensus AI Summit to cut through the hype to find where true transformation and opportunity lie. Listeners to this show can get 15% off registration with the code AIBREAKDOWN. Visit consensus2024.coindesk.com to learn more. Some of the folks we'll be at Consensys this year include Guillaume Verdon, AKA Beff Jesos, founder and CEO of Xtropic as well as spiritual leader of the Accelerationist movement Neal Stephenson, co founder of Lamina One, and Brendan Eich, the CEO of Brave Software. Again, go to consensus2024.coindesk.com to learn more and get 15% off registration with the code AIBREAKDOWN. Today's podcast is brought to you by Plum. Are you a builder with a vision for an AI powered app but lack the technical expertise to bring it to life? Plum believes that everyone should have the power to create. That's why they've built a no code AI app development platform that lets you go from idea to reality in record time. Whether you're a bootstrapper, indie hacker or product visionary, Plum gives you the tools to build your dream AI app. Make everyone a builder with Plum. Check out useplum.com that's Plum with a B for early access to the future of AI app development. Welcome back to the AI Breakdown. Yesterday, AI users got a very delightful update when Anthropic announced that they had finally dropped an iOS app for Claude. What we're going to do today is talk about exactly what was announced, because that was just one part of it. A little bit about how Anthropic seems to be positioning itself from a branding perspective, and mostly the state of competition between LLMs in the wake of the latest news. So first of all, what was announced? Well, one part of it, like I said, was Claude finally has a mobile app. On the one hand, the app gives you access to a mobile version of the chatbot, but it also has integrated multimodal capabilities so that you can upload a photo and interact with a chatbot. That way, if you've used something like ChatGPT's mobile app, you know that this actually opens up an entirely different world of use cases that can be really, really valuable. You can, for example, take a photo of some hardware thing you're trying to fix and get help figuring out how to actually do it. You can also use it as sort of a personalized tour guide or history buff pointing it at some exhibit or some piece of art in a museum and getting more information. The point being that when it comes to these leading chatbots, I really think that a mobile experience is not just nice to have from a choice perspective. Do I want to do this on the web or on mobile? It really is a whole category of uses that become unlocked when you have that capacity. The app is free for all Claude users, including those who use the free version, as well as those who are pro users who pay $20 a month. And based on that you can use any of the main Claude 3 models. When it comes to community response, I would say mostly people were just excited to finally have access to Claude via mobile, although I did see a couple people point out that there are some subtle ways in which Anthropic was extending its brand position. As Morgan from X puts it, Anthropic unashamedly showing their highbrow in the Claude app store images. Specifically, they give an example of someone who's uploaded a PDF of painting after Durer, a text file of Cumulus poems and a PDF of classical composition with the question Claude, what are five blog titles about these texts? The first one that Claude suggests is the importance of Claude's in Albrecht Durer's Landscape Etchings, a study of classical composition techniques. So yeah, Anthropic is definitely positioning for a certain type of person. What's more, if you look at the 15 second promo video they released on Twitter as well, it's clear that they are trying not just to appeal to white collar professionals, or at least not some generic boring business user. The main text of the video is Claude is AI for architects, lawyers, economists, crafters, scientists, musicians, learners, connoisseurs, artists, academics, caregivers, engineers and more. It feels in its positioning, in its coloring, very sort of artisanal. And it may be that that was a very explicit choice to differentiate it from the other thing that was announced, which was their new team plan. In fact, when it comes to the hierarchy of the announcements, Team Plan was actually positioned first. Anthropic writes the team plan enables ambitious teams to create a workspace with increased usage for members and tools for managing users and billing. It's the best way for teams across industries to leverage our next generation Claude 3 family model. The team plan costs $30 per user per month. So what does it come with for that 50% premium over the Pro plan? 1 is just increased usage. Anthropic says that with this plan versus the pro plan, every teammate can quote significantly increase the number of chats they have with Claude, it comes with a 200k context window. This is aimed, they say, at processing long documents like research papers and legal contracts, discussing complex topics like financial forecasting and product roadmapping, as well as maintaining multi step conversations like customer support inquiries now they also say that more team features are coming. In the coming weeks, they write, we will be releasing additional collaboration features including citations from reliable sources to verify AI generated claims, integrations with data repositories like code bases or CRMs, and iterating with colleagues on AI generated documents or projects. In fact, the video that they released alongside the team announcement isn't about stuff that's available right now, it's about those collaboration features that are coming. And really it looks to move from a chatbot that has some increased usage to a shared workspace that is AI powered. When it came to the community response here, things that were definitely seen as benefits as compared to particularly ChatGPT's team model was the higher usage limits in the 200k context window. People are also excited to see how this future access to code bases and CRMs will work. The biggest complaints was that Anthropic didn't give many details around data privacy, voice input doesn't seem to be enabled and there is a 5 seat minimum as compared to the 2 seat minimum of ChatGPT teams. Still, I think when push comes to shovel, most people are generally enthusiastic that Anthropic is walking down this path and kind of waiting to see the deeper collaboration features before really pinning down how good they think it is compared to the competition. Now let's speak about that competition. The competitive landscape recently has changed a bit specifically because of Meta's Llama 3. When Metalama 3 started performing very close to the state of the art of both the GPT4 Turbo as well as Claude 3 Opus, it definitely changed the way that some people thought about how this industry is going to evolve. Basically the question is if there's an easily accessible open source version that's close enough to state of the art, does that make state of the art models a commodity? Put differently, will it reduce enterprises willingness to pay a premium for the state of the art when they can get access to a close enough version for cheaper? Lumina Wealth CEO Rahm Alawalia writes, the disruptors are getting disrupted. Devs are replacing OpenAI with Metalama with no degradation. Why? Price is cheaper. GPT4 is $10 per million tokens input meta is $0.60. For that amount, who will pay the $20 a month OpenAI subscription Zuck just took a major swing at Microsoft. Bindu Ready writes OpenAI says they are going to steamroll startups and make them obsolete. Last time I checked, llama3 is 10x cheaper and is as performant as GPT4. Meta AI is free and available on Facebook, Instagram and WhatsApp compared to ChatGPT, which cost $20. Grok and Google have integrated Genai results into their core products. Technically speaking, isn't OpenAI already obsolete now? While many pointed out that that's not exactly what Sam Altman was talking about, the context for that comment was a podcast conversation where he basically said that it was probably not the right idea for startups to try to fill gaps in what OpenAI could do, when it was likely that the capacity of GPT5 or GPT6 was just going to fill those gaps anyways. But I think the broader point that she's making around the new competition that OpenAI faces because of 1 the lower cost of these open source alternatives and 2 the availability for these big tech companies that have big distribution platforms like Google or Meta's family of apps to plug their AI in does bring up questions about the long term economic viability. Even if companies like OpenAI or in the context of our conversation today, Anthropic, are operating at the state of the art, no one can presume to know the answer to this question. And anyone who says confidently that they are sure that enterprises will just work with an open source alternative, or that consumers will by and large just use the models that are plugged into the apps they already use, it's just too early to tell. We don't know how big the premium will be for different types of users on any gap between the state of the art and close to state of the art. We don't know how the AI agent age will change this as well. The gap between state of the art when it comes to agent capacity and close to state of the art could be significant enough that it is worth the premium, because the value derived from the premium version is still so much less expensive than whatever process it's replacing. There is also the simplicity and convenience factor, Professor Ethan Mollock tweeted recently, at least in the sample of firms I talked to, seeing a surprising amount of organizations deciding to skip or at least not commit exclusively to customized LLM solutions and instead just get a bunch of people in the company ChatGPT Enterprise and have them experiment and build GPTs. This is not shocking to me at all. Yes, theoretically you would imagine that cost and customization would make some parts of the enterprise want to go with those customized solutions. But the complexity of that versus just letting people sign on to something and start using it right away and accruing value right away is a really hard trade off. Now the question is, is that just for the early exploration phases where companies are trying to figure out exactly what these tools are going to be good for? Could be. It could be just a byproduct of something I've talked about before, which is the bottoms up adoption that's happening in enterprise AI, but I'm not so sure. I also don't think that they're mutually exclusive. I can see certain functions within the enterprise tending to have preference for customized solutions, while the cost of just giving everyone access to these $30 a month subscriptions for either ChatGPT Enterprise or Anthropic's new teams still just remains worth it for huge swaths of the organization. Anyway, it's a really fascinating time. We are in a transition moment where we simply don't know what's going to come out on the other side. And as of right now, Anthropic is fully in that game. That's going to do it for today's AI breakdown. Until next time. Peace, Sam.