Anthropic Raising at a $20B-$30B Valuation

0:01Today on the AI Breakdown, we're discussing reports that Anthropics is raising new money at a 20 to 30 billion dollar valuation. Before that, on the brief, Morgan Stanley thinks that more than 40 % of workers could be impacted by AI in the next three years. 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 Discord, our YouTube channel, and our newsletter. The world's biggest YouTuber has taken to his socials to warn people of a deepfake scam that uses his likeness, a video of him in fact, to try to get people to send money in exchange for an iPhone.

0:38Basically, the video claims that Mr. Beast is giving away 10 ,000 iPhones for just$2 each, and that if you're seeing this ad, you're one of the lucky winners. If you're watching this video, you're one of the 10 ,000 lucky people who'll get an iPhone 15 Pro for just$2. I'm Mr. Beast, and I'm doing the world's largest iPhone 15 giveaway. Click the link below to claim yours now. Now, of course, this ad was not created by Mr. Beast, but by artificial intelligence. Beast tweeted, Lots of people are getting this deepfake scam out of me. Are social media platforms ready to handle the rise of AI deepfakes?

1:12This is a serious problem. Matthew Kobach tweeted, First AI scam I've been served as an ad. I wonder if social media platforms are prepared for this. This fake Mr. Beast could just as easily be a fake Trump or Biden. Now, some argued that this was going to get worse with the rise of open source AI, which is part of the subject of our main episode. A Reverend Labs co-founder, Rahul Sood, said, It is a problem, but this will only get worse with open-source AI, as not everyone runs with high integrity. This is another reason why AI companies should be a protected class in the US, and regulations should be designed to protect and foster innovation here.

1:44Now, of course, this isn't the only news we've had recently about a celebrity warning of a deepfake ad using their likeness as part of a scam. Over the weekend, Tom Hanks posted to his 9.3 million Instagram followers that quote, there's a video out there promoting some dental plan with an AI version of me, I have nothing to do with it. CBS anchor Gayle King also posted a video with the big text on top of it fake video and said, people keep sending me this video and asking about this product and I have nothing to do with this company. I posted this video promoting my radio show on August 31st and they've manipulated my voice and video to make it seem like I'm promoting it.

2:17I've never heard of this product or used it, please don't be fooled by these AI videos. I think it's safe to say that the game of deepfake whack-a-mole has officially begun. Now, moving over into the world of labor, everyone has a sense that artificial intelligence is going to have a big impact on the economy and the jobs market. The question is just how much? Well, a new report from Morgan Stanley analyst says that 44 % of the labor force is likely to be affected by artificial intelligence in the next three years. They estimate a$4.1 trillion economic effect on the labor force. The note read, We see generative AI expanding the scope of business processes that can be automated.

2:51At the same time, the input costs supporting Gen.AI functionality are rapidly falling, enabling a strongly expansionary impact to software production. As a result, generative AI is set to impact the labor markets, expand the enterprise software TAM, and drive incremental spend for public cloud services. More subjectively, Janet Yellen, the U.S. Treasury Secretary, also added her thoughts on AI, saying that its quote, unbelievably rapid development could have a big impact in boosting the economy. Now, she noted that American productivity has gone down. Earlier this This year, Fortune published a piece, American worker productivity is declining at the fastest rate in 75 years.

3:27Yellen said that the explosion of AI could supercharge productivity. She said the progress in this area is unbelievably rapid and it could make a significant difference. Now, overall, she positioned AI as part of a larger shift through which she said we're restoring U.S. leadership in science and tech. Of course, in order to have all of those productivity gains, we have to help people actually learn about AI and adapt to the new AI economy. And on that front, it seems that LinkedIn is doubling down. TechCrunch writes, LinkedIn goes big on new AI tools for learning, recruitment, marketing, and sales powered by OpenAI.

3:59Now, LinkedIn has had a steady drip of new AI products. Back in March, for example, it announced AI-powered writing suggestions that were for both posts as well as for direct messages. And recruiters have also reported seeing a series of tests around job description support using AI. And that appears to be getting a little bit formalized with this latest set of announcements. The company announced something called Recruiter 2024, which is a new AI-assisted recruiting experience. Basically, what it allows for is recruiters to use more conversational, natural language to try to hone in on the type of employees that they're looking to find.

4:29In addition to helping recruiters refine their searches, Recruiter 2024 will also have more suggestions outside of what recruiters might think that they are conventionally looking for. Over on the learning side, there is a new AI chatbot that they're calling a learning coach that will give advice around soft skills to job seekers, job switchers, and current professionals. Many are speculating that if this learning coach has positive results, it might be extended to recommending content from LinkedIn's extremely large catalog of learning videos and other educational resources. Finally, both the marketing and sales aspects of the platform will also get an AI upgrade.

5:02Like so many of the other big tech social platforms, The marketing tools are having AI integrated specifically with a new product called Accelerate. And on the sales side, the new AI features promise to help salespeople find the right contacts more efficiently. While the cynical take on all this might be that this is just throwing the AI kitchen sink at everything LinkedIn does, I sort of think that that's just the phase that we're in right now, where basically every company is going to be experimenting with artificial intelligence across the suite of what it does in an effort to see where it actually makes a difference and where it doesn't.

5:31I'm not sure which of these are going to stick, but I think that the trend line of trying all these AI features is pretty clear. Now, not all the news in AI land is particularly great. One of the bits of news that contributed over the summer to a sense that AI hype might be dying down were job cuts at once buzzy startup Jasper AI. And now on top of that, the company has cut the internal value of its common shares by 20%. The company had previously raised a venture round at a$1.5 billion valuation back in June. Now, in many ways, while Jasper started as exemplary of the Silicon Valley excitement in AI startups, It has instead come to represent something very different, which is a warning that just slapping a user interface wrapper around someone else's technology might not be the way to long-term success in this rapidly evolving space.

6:15Now, while this valuation cut has not been confirmed, it is clear that there is some turbulence inside the company. Last Thursday, for example, they announced that CEO and co-finder Dave Roganmoser was stepping down, and former Dropbox president Timothy Young would be replacing him. Bringing it back all the way around to where we started with deepfakes, one of the potential answers that has been posited to the inevitable rise of deepfakes has been AI watermarks. In other words, technological solutions to identify content as having been created by artificial intelligence. Now, there are a lot of different approaches to watermarking, some that are visible to the human eye, some that are invisible.

6:49But new research from the University of Maryland suggests that whatever solutions we have right now, they are not sufficient. As Wired summed it up, researchers tested AI watermarks and broke all of them. Computer science professor Sho Helfezi said, We don't have any reliable watermarking at this point. We broke all of them. Now, one take on this, of course, is that watermarking is just useless, but another take is just to understand that there are limitations to how effective they are. Jonas Gepping points out that watermarks were never likely to solve the issue of AI being created adversarially that was purposefully meant to deceive.

7:21As he writes, It is unreasonable to expect watermarking to prevent instances of targeted misuse like fake news campaigns. This is an important message to policymakers who sometimes seem like they're looking only for easy answers. If nothing else, between this and the Mr. Beast news, it's very clear just what a challenging cat-and-mouse game this is likely to be. But for now, that's going to do it for today's AI Breakdown Brief. Thanks as always for listening or watching, and I'll be back soon with the main AI breakdown. Welcome back to the AI Breakdown. Well, Anthropic has dominated the news for the last week, and interestingly, they are back in the headlines again.

7:55Today, we're going to talk through the new report, as well as explore what it says about the state of the competition for foundation models. Now, just to bring you up to speed with recent developments, on September 25th, Amazon and Anthropic announced a major collaboration that included an initial$1.25 billion investment into Anthropic that could be expanded up to$4 billion, as well as a collaboration with Amazon Web Services. Andy Jassy tweeted,

8:29Many very reasonable people then started to outline the new parameters of the AI battle as Microsoft plus OpenAI, Google plus DeepMind, Tesla plus XAI, Meta plus its internal Meta AI, and now Amazon and Anthropic. Now, one thing that Anthropic was clear on was that, as they put it, as part of the investment, Amazon will take a minority stake in Anthropic. Our corporate governance remains unchanged and will continue to be seen by the long-term benefit trust in accordance with our responsible scaling policy. In some ways, I read that as them specifically saying this isn't quite as close as, for example, Microsoft and OpenAI might be.

9:07Well, putting a fine point on that is this new report from The Information that Amazon is talking about a fresh$2 billion round that would be led by Google and others. Now, it's important to note that Google is already an investor in Anthropic. In 2022, Google bought around 10 % of the company. Now, the previous fundraise for Anthropic from March, which was, of course, before this Amazon investment, had valued Anthropic at around$4 billion. But according to the information sources, the company is now seeking a valuation of between$20 and$30 billion. If they were to get that, it would represent a 200x multiple on the company's current$100 million revenue run rate.

9:43Now just by way of comparison, should they successfully complete this raise at that valuation, they would be at a significantly higher ARR multiple than even OpenAI at this reported$80-90 billion round, given that OpenAI is now reportedly at a billion dollar annual run rate. Now that said, Anthropic's revenue is apparently growing fast. Sources close to the company said that by the end of this year, Anthropic is projecting that it will generate revenue at a$200 million annualized rate, or around$17 million a month, and that by the end of next year, it hopes to be at a$500 million annualized rate, or more than$40 million a month.

10:15Importantly, one of the reasons that they're projecting that amount of growth is this deeper tie-up with Amazon, which will see Amazon's bedrock push Anthropic to cloud customers in a more direct way. Now, in terms of how much Amazon is setting the valuation, apparently their investment was done in the form of a convertible note, which turns into stock at the next fundraising round. The information sources also made it seem like a big deal for Amazon was the ability to get their specialized server chips into a major player like Anthropic as a way to start competing against NVIDIA's GPUs. Kind of stating the obvious, the information writes, If Anthropic raises more money from Google, it would raise questions about how the startup plans to split its allegiance between the two rival cloud providers.

10:53Now, one interesting thing that could explain why these companies are trying to raise so much money so fast comes from Anthropic's pitch deck, which reportedly reads, We believe that companies that train the best 2025-26 models will be too far ahead for anyone to catch up in subsequent cycles. In other words, that there is a very distinct moment in the history of the evolution of the artificial intelligence space that will mint winners and losers, and that effectively right now is the only chance anyone really has to compete. Now of course, the other interesting dimension to this battle is the degree to which all of these big tech companies have different philosophies on how the AI space in general should evolve.

11:29As we reported recently and once again from the information, apparently a lot of the tense conversations happening behind closed doors in Washington, D.C. have to do with what the right approach to open-source AI will be going forward. Here's how the information summed up the friends and foes of open-source AI. In the pro-open-source column, they put Meta and Databricks. They summed up Meta saying, Says licenses are not appropriate for current generation of AI models. Released Llama 2, a cutting-edge open-source LLM over the summer. And indeed, it appears from this reporting and from just basically any other conversation that we've read, that part of what has put such a fine point on this question around open source is how fast it's evolving thanks to initially the leak of Llama 1 and then the release of Llama 2.

12:10Moving back to where a company stands, Databricks, they say, says open source AI should not be unduly restricted by regulation. Now, somewhere in the middle are Microsoft and Google. Microsoft, they write, has supported requiring companies to obtain licenses before releasing models, but released FI 1.5 and Open Source LLM in September, and Google, they say, has not indicated whether it supports requiring government approval for launching Open Source LLMs. Its last Open Source release was in December 2022. And the companies that are against Open Source AI, Anthropic and OpenAI share common cause here.

12:40OpenAI, the information says, wants companies to obtain government approval before releasing highly capable AI models. He points out that they have not open-sourced an LLM since 2019. And in fact, Sam Altman has said explicitly that they were wrong about their approach to open-source AI. Anthropic, the information characterizes as opposing open-source cutting-edge AI. Supports independent evaluation for safety of models prior to release, including by the U.S. government. Has not open-sourced its AI. Now, as the article points out, this is not just a principal question about whether open-source should be allowed or not.

13:09In fact, it's much more of an implementation question of answering things like, If policymakers agree that some models are too powerful but other models aren't, how are those lines drawn? What is the process for determining whether a model is too powerful or not? Now, one of the folks who also thinks that we shouldn't be as open as we are with open-source AI is former Google DeepMind co-founder Mustafa Suleiman, who's now the co-founder and CEO of Inflection. Last month, he put out a book called The Coming Wave, Technology, Power, and the 21st Century's Greatest Dilemma. In August, before his book came out, he stirred up controversy by tweeting, The last century was defined by how quickly we could invent and deploy everything.

13:44This century will be defined by our ability to collectively say no to certain technologies. This will be the hardest challenge we've ever faced as a species. It's antithetical to everything that has made civilization possible to date. For centuries, science and technology have succeeded because of an ideology of openness. This culture of peer review, critical feedback, and constant improvement has been the engine of progress powering us all forward. But with future generations of AI and synthetic biology, we will have to accept that this process needs an update. Figuring out how to delicately do that without trashing innovation and freedom of thought will be incredibly hard.

14:14But left unchecked, naive open source in 20 years time will almost certainly cause catastrophe. Physicist and AI founder Kevin Fisher writes, the voice of an authoritarian. Why Combinator President Gary Tan said, lost me at banning open source and matrix multiplication. Bad look and feels grifty. Now, John Carmack, the former CTO at Oculus, had a more nuanced review of the book. He tweeted, This is written without any wild-eye fear-mongering, and I like some of the historical perspectives, but he is clearly a statist. He previously worked in government, expresses concern for the government almost as often as concern for people, and wants to see powers expand in scope and across national borders.

14:48There is an undercurrent of, we obviously can't let the peasants have crossbows. I also don't like the bundling of biotech risk and AI risk. It feels like a big tent play to raise more concern and grow more levers of power. At the end, he suggests making high-end open-source AI work illegal, and imposing censorship on the dissemination of prohibited AI research. Cheers for openly saying what you want instead of leaving it all vague, but I am not aligned. Stability AI CEO Ahmad Mostak quote tweeted that tweet and said, I agree largely with John here and would actually go a bit further. Almost every argument I've seen made against open source AI is ignorant of history, logic, and technology.

15:21Those who give up the liberty this technology will enable for supposed safety will get neither. When someone asked him then why he had signed the statement saying mitigating the risk of extinction from AI should be a global priority, Ahmad responded, The danger of AI extinction is a threat that should be considered a global priority just in case. It is not nuclear-level tech overall, and things like dataset transparency and open-source mitigate threats. Closed AI controlled by states with no alternative is the danger. Among people that agreed with the mod included Epic Games founder Tim Sweeney, who said, Agreed!

15:49It's a new generation of unprincipled company leaders lobbying the government to prevent upstarts from competing with them. Thanks for keeping stability open. Now this is obviously a growing, contentious conversation. And when it comes to what normal people think, I thought subroshi.eth had a really honest point. They wrote, Most of us simply do not understand the topic well enough to make any real judgment, which is a bit scary considering how huge the consequences might turn out to be. Can't blame people for being cautious. Now the reason that I think it's worth revisiting this conversation around open source and AI safety in light of a larger anthropic raise is sort of summed up by a point made on September 25th by Rune, who wrote, Anthropics splitting off without a doubt accelerated AGI timelines.

16:27The point that he was making is the irony that Anthropix leaders left OpenAI because they were concerned about their approach to AI safety, but in so doing simply increased competition, which almost inevitably will lead to faster evolutions in the space. I've called this before competitive accelerationism. It's quite clear from all this fundraising activity that the battle is getting more, not less intense, and that means that creating calm middle space for nuanced conversation about these questions is going to get nothing but harder. Hopefully we can create at least a little bit of that here in our small corner of the AI world.

16:57Lastly, one little side note for those of you who follow both crypto and AI, an incredible denouement to the FTX story is that among all of the crazy, random, dragon's treasure hoard of assets, as Michael Lewis put it, that Sam Bangman-Fried had between his personal accounts, Alameda and FTX, was an investment in Anthropic. If this deal goes through, and Anthropic is able to raise at that$20-30 billion valuation, that puts the stake that FTX owns at between$3-4.5 billion. It doesn't completely do it, but it comes fairly close to covering the hole owed to FTX customers. How wild is it to think that among the two big November 2022 events, the collapse of FTX and the launch of ChatGPT, the tailwinds from the latter might be so powerful that they help clean up the mess of the former.

17:41Pretty fascinating stuff, guys. But for now, that is going to do it for today's episode. If you enjoy this, if you like the AI breakdown in general, I would so appreciate it if you would subscribe wherever you listen to podcasts and leave a rating or a review. It makes a really big difference to how new people find the show, and I appreciate each and every one of you who has taken the time to do that. Until next time, guys. Peace.