$550B In Net Worth Met In D.C. To Talk AI: Did Anything Get Done?

0:00Today on the AI Breakdown, we're exploring what happened at that closed-door session between senators and big tech leaders yesterday in Washington, D.C. Before that on the brief, Stability releases a new audio and music generation model, Goldman Sachs says there's no AI bubble, and much, much more. 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.

0:30Welcome back to the AI Breakdown Brief, all the AI headline news you need in around five. minutes. We begin today with one of my favorite categories in the entire artificial intelligence space, which is the new text-to-audio, or really more specifically, text-to-music space. Now, this is an area that has lagged perhaps a little bit behind text-to-image and text-to-video, but is really starting to come on strong. Over the last few months, we've gotten Google's Music LM, which remains in a research and testing version. Same with Meta's Audiocraft, which also is mostly available for testing and playing with right now and doesn't really have an application to go along with it.

1:06At the same time, you're seeing more startups like Cassette AI who are trying to bring this type of experience to the general public. And of course, as you know, although Stability AI is constantly releasing new models, they're almost always really strong. Now, Stable Audio is being released in a slightly different way in a slightly more advanced state than Google LM and Meta's Audiocraft. The announcement post writes, Stable Audio is a first-of-its-kind product that uses the latest generative AI techniques to deliver faster, higher quality music and sound effects via an easy-to-use web interface.

1:36And by the way, that web interface is of course what makes it different. Stability AI offers a basic free version of Stable Audio, which can be used to generate and download tracks of up to 45 seconds, and a Pro subscription which delivers 90-second tracks that are downloadable for commercial purposes. Now in terms of some of the details, Stability writes that this model was trained using music and metadata from AudioSparks, which is a music library that includes over 800 ,000 sounds. The blog continues, Stable Audio is the first music generation product enabling the creation of high-quality 44.1 kHz music for commercial use via latent diffusion.

2:09The latent diffusion architecture uses audio conditioned on text metadata as well as audio file duration and start time, allowing for control over the content and length of the generated audio. The Turing Post also points out that Stability gets much faster inference time compared to raw audio because they heavily downsample the latent representation of the audio. What that means is that stable audio can render 95 seconds of stereo audio in less than one second on an NVIDIA A100. But of course, this is really all about what it sounds like, so let's look at the samples they gave. The first is Trance, Ibiza, Beach, Sun, 4AM, Progressive, Synthesizer, 909, Dramatic Chords, Choir, Euphoric, Nostalgic, Dynamic, Flowing.

2:56second is warm soft hug comfort low sense twinkle wine and leaves ambiant peace relaxed water

3:17The third is drum solo.

3:39Sometimes the simplest prompts are the best ones. Now, I think that there are going to be a ton of uses for this. My guess, in fact, is that the first iterations of it aren't actually used as much for people producing end-to-end musical tracks, as much as they are used to produce new generative audio as the soundtrack for other multimedia creations. In other words, I think you're going to see a lot of stable audio soundtracks for Runway and Pika Labs-generated videos. But who knows? Certainly the speculation is cooking. For example, Brian Romley writes, Stability AI just announced Stable Audio, a new AI product that can generate high-quality music and sound effects from simple text prompts.

4:17As I said, by 2039, most of the music you listen to will be tailored to AI music designed to, or by, you. Next up, we move over to another set of generative AI tools that have now become more broadly available. Earlier this year, people got really excited about some of Adobe's new tools that are part of their larger Firefly generative AI suite. The most notable one, of course, was generative fill. Generative fill is, of course, a tool by which a person can take an image and make a specific change using natural language to just a part of an image. So, for example, you can put a dog in a patch of grass.

4:53You can also expand a photo, and this is something that people were doing quite a bit when it first was released. But when it was first released, it was only available in beta. That meant you had to download, for example, the test version of Photoshop to use it, and that has now changed. Adobe's Firefly AI tools are now generally available to pretty much all of their users with the exception of places like China where there are laws that prohibit those models from general release. On top of that, Adobe is also launching a standalone Firefly web app that will give people the ability to use some of these generative AI capabilities without having to subscribe to the specific Adobe Creative Suite applications.

5:27And one that you have to think many businesses will be excited about is that Adobe Firefly for Enterprise is also now widely available. Now, one of the things that businesses were excited about with this release is the fact that because Adobe's generative AI image model is trained on Adobe stock and public domain content, theoretically, it's much safer from copyright claims, which is something that enterprises are worried about given how many lawsuits are currently flying around based on how different AI companies have trained their models. What's more, Adobe has also said that they will foot the bill if a company is hit with a copyright complaint.

5:58Now, no doubt there is lots and lots of excitement around new AI tools, and that excitement, as we well know, has extended to the stock market, where AI stocks have led a rally this year in spite of so many negative macro factors, from bank crises to increased interest rates to geopolitical instability. And this, of course, has had many questioning whether AI stocks were already in a bubble. However, at least for now, Goldman Sachs' answer is that no, they are not. On September 11th, they published a report why AI stocks aren't in a bubble. Chief global equity strategist at Goldman Sachs Research, Peter Oppenheimer, writes, writes that the valuations of stocks leading the market are, quote, not as stretched as in previous periods.

6:38He says, quote, we believe we are still in the relatively early stages of a new technology cycle that is likely to lead to further outperformance. One of the things that the report points out is that tech stock valuations have gone up this year despite rising rates, which is the opposite of what happened last year as we saw a significant sensitivity to rising rates in the prices of technology stocks that had gone up quite a bit in the previous couple of years. To put some numbers around it, the blog post summing up the report writes, technology sector valuations are certainly high by historic standards.

7:06Compared with the 10-year median and range, the current price-to-earnings ratio for the US tech sector is right at the top. But that's not the whole story. The seven biggest US companies seen as leaders in the race to commercialize generative AI technology have an average PE of 25. That compares with a PE of 52 for the biggest companies at the peak of the internet bubble. What's more, they write, the current crop of technology leaders are already very profitable and generate cash, meaning they are investing at a high rate even in an environment of elevated interest rates and borrowing costs. Their cash as a percentage of market capitalization is double what companies had during the internet bubble.

7:40Their return on equity and average margins are also nearly double what was seen during the 1990s run-up. Says Oppenheimer, this has made these companies relatively defensive in terms of their revenues and earnings. TLDR, according to Goldman Sachs, we are still at the beginning and not yet in overheated territory. And yet, valuations are certainly climbing. Bloomberg reports on a new round for drone startup Shield AI, through which they will raise$150 million at a$2.5 billion valuation. Now, of course, one of the big themes that we have talked about on this show is how, as politicians hem and haw about how AI should develop or not, the US military and other militaries around the world are absolutely racing forward with autonomous technology and new AI-powered weapons and defense systems.

8:22Now, it's worth noting that Shield AI is not some Johnny-come-lately trend-jumping company. It was started all the way back in 2015, and earlier this year announced a partnership with Boeing. Another example of AI finding its way into big business comes from EY. The services and consulting company has spent$1.4 billion developing an AI platform that includes developing its own LLM, EY.AI EYQ. Now, this follows after competitors and peers KPMG said in July that they plan to spend$2 billion on AI and cloud services over the next five years. And in the same month, Accenture announced a$3 billion investment plan to expand its investment in AI as well.

8:58Now, in my estimation, far from being just a jump on the trend kind of move, these major consulting firms actually have a really sensible use case for custom or at least customized LLMs that take advantage of their incredible trove of data. If you think about how the general consulting contract works, all of the data and information and learnings are siloed off from the teams that didn't work on it. This creates an opportunity where you could theoretically use an LLM and other AI technology to give people across the company access to better information and learning from the collective experience of the company rather than just the projects that they previously worked on themselves.

9:33Now, how exactly that will impact their work, Whether it will result in better performance or lower costs, or simply advanced capabilities, remains to be seen. But on a base level, I think it makes a lot of sense for these big consulting firms to be investing in AI the way that they are. And so, friends, the AI march continues on. There is going to keep being lots to talk about. But for now, that is going to do it for the AI Breakdown Brief. I'll be back soon with the main AI breakdown. Welcome back to the AI Breakdown. The biggest event in AI policy this week was undoubtedly Senate Majority Leader Chuck Schumer's AI Summit.

10:08Today, we're going to talk about everything that happened, or at least what we could ascertain given that it was behind closed doors. But let's start off just with a recap of who actually participated. As Bloomberg's headline blared, this summit had about a half trillion dollars worth of wealth,$550 billion in net worth in a single room. That included Elon Musk, Bill Gates, Mark Zuckerberg, Jensen Huang, Eric Schmidt, and so many more. Now, as we noted yesterday, the meeting also included people from the world of civil rights, organized labor, and a number of other sectors, but it's hard to deny that the featured headliners of the discussion were the big tech billionaires themselves.

10:44It sounds like from the reporting that there was quite a bit of tension in the room. Part of that was about disagreements around policy, and part of that was personal. For example, according to people who were in the room and then reported it to Bloomberg, there were fairly significant disagreements between Zuckerberg, OpenAI CEO Sam Altman, and Bill Gates around the risks of releasing models as open source. And there was also apparently a tense moment between Berkeley researcher Deb Raji and Tesla CEO Elon Musk, where Raji accused Musk of downplaying the concerns that many have around AI-powered self-driving cars.

11:15Bloomberg also writes, apart from policy disputes, there was also personal friction among some of the attendees. Musk, the world's richest man, previously proposed a cage fight with Zuckerberg, the world's 10th richest man, and took to his own social media platform to criticize Gates, the world's fourth richest man, for allegedly shorting Tesla stock. Said Humane Intelligence CEO Rahman Chowdhury, that's probably the worst wedding to try to do seating for. Chowdhury reported that there was zero interaction between Musk and Zuckerberg, who were probably quite wisely seated at the opposite ends of a huge table that ran the length of the room.

11:46Schumer began the meeting by saying, Today we begin an enormous and complex and vital undertaking, building a foundation for bipartisan AI policy that Congress can pass. This is going to be one of the hardest tasks we undertake because AI is so complex, will impact nearly every area of life, and is evolving all the time. Now, apparently in the meeting, Elon Musk told senators that were there that AI is a double-edged sword, and he spoke a bit with reporters after the summit as well. He said, I think this meeting could go down in history as important to the future of civilization. In those comments, Elon also reiterated his call for a federal department focused on AI exclusively, and said that although there might have been tensions in the room, overall he thought there was a, quote, strong consensus around the need for regulation.

12:27Fortune reported something very similar from meeting organizer Chuck Schumer. The recap piece they ran reads, Chuck Schumer says he asked Musk, Gates, and others about whether to regulate AI, quote, every single person raised their hands. As Fortune puts it, Schumer's basic point is that Congress should do what it can to maximize AI's benefit and minimize the negatives, whether those be, quote, enshrining bias or the loss of jobs or even the kind of doomsday scenarios that were mentioned in the room. and only government can be there to put in guardrails. Musk agreed, saying the key point was really that it's important for us to have a referee.

12:57It was a very civilized discussion, actually, among some of the smartest people in the world. End quote. Now, in terms of the format, each participant at the meeting had three minutes to speak on a topic of their choosing, and then a group discussion was led by Schumer and Republican Senator Mike Rounds. One notable contrary opinion around the idea of having an AI department was IBM CEO Arvind Krishna, who, according to Fortune, quote, expressed opposition to proposals favored by other companies that would require licenses. Now, for all of the potential positivity, there is very clearly not consensus in Washington, not only on how to legislate, but about how to even go about the process of figuring out how to legislate.

13:35Politico ran a long piece in the wake of the meeting called Senate Starts to Fracture Over How to Govern AI. Notably absent from the meeting was Republican Senator Josh Hawley, who alongside Democratic Senator Richard Blumenthal has just introduced their own legislation that would, among other things, require big tech companies to seek licenses for AI systems that are more powerful than what we have today. Hawley said that he would not attend what he said was a, quote, giant cocktail party for big tech, and continued, I don't know why we would invite all the biggest monopolists in the world to come and give Congress tips on how to help them make more money and then close it to the public.

14:06Going back to that political piece, it begins, cracks are starting to emerge in the Senate's approach to regulating artificial intelligence, its top tech policy focus of the fall. One concern brought up around process was the speed of action. Richard Blumenthal, one of the sponsors of that legislation we just mentioned, told Politico, Congress operates at the speed of molasses in sub-freezing weather. We need to operate at light speed on this issue. Blumenthal had kinder words to say about the summit than his legislation co-sponsor Holly, but did warn that an educational push shouldn't be used as a reason to delay the actual introduction of legislation.

14:37He said, you can't know everything before you do something. You need to do the legislation and learn at the same time. Schumer, on the other hand, said we don't want to just put together legislation. If you go too fast, you could ruin things. Now, you'll note here that some of the standard party lines aren't exactly being held up. Schumer and Blumenthal both come from the Democratic Party and are clearly taking very different approaches to this issue, while Senator Josh Hawley is taking a much more antagonistic tone towards big tech than many of his GOP peers. Candidly, I think it's probably a good thing that thoughts around AI aren't falling crisply along party lines.

15:08It suggests that people are actually, you know, thinking about this issue and making decisions for themselves rather than just taking whatever talking points party leaders are handing them. Reinforcing this, Politico writes, The emerging Senate split over AI legislation doesn't break down along partisan lines. In both parties, there are lawmakers urging fast action. And in both parties, there are lawmakers who believe Congress needs more information before it can move. There's also tension with people who have previously been involved in AI regulatory efforts. For example, Washington Democratic Senator Maria Cantwell said, I set up the National AI Advisory Committee years ago for this very thing.

15:40What should the government role be? Cantwell told reporters on Wednesday that Congress had already had three years to discuss what AI legislation should really look like. Now, on top of all this, as more members of Congress get deeper into these issues, naturally more of them are bringing their own ideas to the table. Politico again writes, additional AI frameworks are starting to crop up across Capitol Hill. On Wednesday, Senator Michael Bennett wrote to Schumer with elements of his own AI regulatory framework, which contains third-party audits of AI systems, disclosures on AI-generated content, and opt-in measures on user data collection for AI training.

16:10Other senators are also moving, including Cantwell, who plans to soon introduce bills that would address the threats posed by deepfakes, as well as potential disruptions to jobs and education posed by AI. Said Schumer, there's going to be lots of ideas out there. Holly and Blumenthal have one, and many other people have other ideas. It will be our job to put together as comprehensive a plan as we can that A, incorporates many different ideas, and B, can pass. Perhaps this is why Martin Heinrich, a co-chairman of the Senate's Artificial Intelligence Caucus, said that he didn't believe that we would see AI legislation this year.

16:40He said, I would like to have something that we can pass in Congress, some people have said by the end of the year. I don't see things coming together that quickly, but I do think we could see a package in the following year. However, lurking on top of all of this are questions of the trust that the U.S. population has in Congress itself. Wired yesterday wrote a piece, The U.S. Congress has trust issues. Generative AI is making it worse. Senators are meeting with Silicon Valley's elite to learn how to deal with AI, but can Congress tackle the rapidly emerging tech before working on itself? The piece reads, there's a problem though.

17:10Schumer's facilitating the wrong conversation. As generative AI is poised to flood the internet with more and more convincing disinformation and misinformation, many AI experts say the top goal of the Senate should be restoring faith in, well, the Senate itself. Dan Mintz, chair of the Department of Information Technology at the University of Maryland, said, the government is, in my opinion, based on a belief in process over a result, that if the process is equitable, we'll live with the results whether you agree with it or not. But people now don't believe in the process, and they don't believe in the result.

17:37Now, one point that the Wired piece does make, which I think is notable, is that while people have genuine concerns over closed-door meetings and the American population not being able to see what's going on behind those closed doors, the bipartisan group of senators that participated in three private all-Senate AI briefings this summer would suggest that there is some seriousness here with actually engaging with the issues. Basically, as Wired points out, most hearings aren't really about senators learning about an issue. They're a chance to use one's five minutes of time to drop a nice little bomb that hopefully makes for good content for YouTube and social media.

18:11In that way, the fact that these conversations are happening behind closed doors suggests that they may actually be better optimized for learning. But of course, that still hasn't stopped many people from being worried that this will be a trend in how AI policy is made. The AI Now Institute tweets, There's a real danger in developing AI policy behind closed doors where industry outnumbers civil rights orgs, unions, and critical thinkers. If this convening is an indication of where Congress is heading, we risk entrenching big tech corporation power to the detriment of the public. Elevation Partners co-founder Roger McNamee had a slightly crisper way of putting it.

18:42Yeah, I expect Chuck Schumer to shake these guys down for some really large contributions. You're such a cynic. Come on. I'm sorry. That is how the game is played. But if these guys... Really, it's just a fundraising opportunity for the majority leader? I believe that's what it is. And by the way, not just for the majority leader, I think for all of them. You know, the situation here is if they were serious about regulating tech, the thing would be filled with people from civil society. The way the thing is structured, OpenAI is essentially a sharecropper on the Microsoft Azure platform. And at the end of the day, most of the profits from whatever gets done by open AI are going to be captured by Microsoft.

19:23They really have all of the birthstrings here. In many ways, I think the question comes down to what extent this meeting was representative of a goal to give senators the chance to hear from these tech CEOs in a context in which they might be more real, for lack of a better term. By removing the cameras from the room, by removing the public eye from the room, could members of Congress actually get more raw and unfiltered perspectives that ultimately will be better useful for figuring out how to write the right rules of the road for this new technology category? Now, the cynical view, of course, and the concern expressed by the AI Now Institute is that instead, this is representative of regulatory capture and big tech getting to leverage their power and wealth to dictate how the rest of the world has to run.

20:06I think that on this front, the two answers are, one, we'll kind of have to wait and see around which of these polls it really is closer to. And two, let's not forget that it's election season now. And the American public who has strong feelings about these issues, if they have strong feelings about these issues, gets a chance to make those feelings known. In any case, what is undeniable about yesterday is that it shows just how significant AI is in the US policy agenda. This is not a fringe issue. This is frankly not just an emerging technology issue. It's an issue which senators and members of Congress are rightly seeing as touching basically everything that we do as individuals, as an economy, as a society, as a whole.

20:42Interesting times indeed. Thanks as always for listening or watching. And until next time, peace.

20:52Thank you.