OpenAI Hits $2B Revenue Run Rate

Today on the AI Breakdown, the White House wants to cryptographically verify all official communications. Before that, on the brief OpenAI hits a $2 billion revenue run rate 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. One of the questions after Sam Altman was fired and then rehired was whether it would impact the company's ability to sell into the enterprise. Think about it. If the company was subject to these sort of disruptions, would big enterprise or business buyers think to themselves, maybe I should go with someone else? Well, apparently those concerns were a little bit overblown because in December, according to The Financial Times, OpenAI hit a $2 billion revenue run rate that is up significantly from just a couple months earlier and makes the company one of the fastest growing in history. Now on the one hand there is no shortage of companies who are competing for the big enterprise AI model space, but at the same time, right now it feels to me like we are very much in the phase of the market development where the amount of new capital coming into the space, in other words the amount of new demand is significantly bigger than any sort of competition between the existing labs, I. E. It's a time when all of the very small handful of companies that are offering these sort of tools to the enterprise and is likely to do very, very well. Now of course, lots of people have pointed out just how remarkable this is in the context of Silicon Valley and point out that its revenue is growing much faster than previous companies like Amazon, Google or Meta. Now one of the ways that OpenAI makes money is through its partnership with Microsoft, although frankly it makes a lot more money when enterprises go direct as Microsoft keeps a much bigger cut when enterprise buyers work through Azure. Last week Azure announced a set of new offerings all based on OpenAI technology, including the Assistance API, new text to speech capabilities, updated models for G turbo and GPT 3.5 Turbo, new embeddings models and updates to the fine tuning API. Basically this is bringing Azure offered versions of OpenAI services into parity with OpenAI direct services. The relationship between Microsoft and OpenAI continues to be one of the fascinating updates and developments of this AI space with a big element of both friend and frenemy in how it plays out. One other interesting little Microsoft update Apparently they are testing a new super resolution AI upscaling tool for gamers the feature would increase the video and image quality for supported games and do so theoretically without sacrif pricing. Playing quality upscaling is actually becoming a very important use case for AI very early, as you can tell from Magnific and all of these other AI upscaler tools that are coming out as standalone startups, as well as the way that upscalers are coming to MidJourney, Dall E3 and other tools like that as a core feature. Now one last headline referencing Microsoft Yesterday we talked about the company's ad for Copilot in the super bowl, and at least from a download standpoint, it seems like people were paying attention. On Apple's App Store On Monday, all three of the most downloaded free apps were companies that had advertised heavily during the super bowl, those companies being Paramount plus Copilot and Temu. Now of course, how much that was an initial burst versus how much people will stick around with Copilot will be interesting to see. Moving over to the world of patents, there has been an ongoing discussion from the US Patent and Trademark Office around whether AI created works could be patented or trademarked, and on Monday we got a clarification on that, basically saying that inventions that were created with the aid of AI could indeed be patented, but that the important thing about it was that the patent should focus on human contributions. It couldn't be primarily an AI created work because they said patents function to incentivize and reward human ingenuity. Patent and Trademark Office director Cathy Vidal said in a blog post, the right balance must be struck between awarding patent protection to promote human ingenuity and investment for AI assisted inventions, while not unnecessarily locking up innovation for future developments. Effectively, we are living in the gray area here and we're just going to keep having to figure it out as it comes along. Now, in a slightly related story in the copyright lawsuit swirling right now, defendants Stability, Midjourney and Runway have all hit back with a flurry of filings that are arguing that the case should be thrown out with prejudice, VentureBeat writes. Yesterday, lawyers for the defendant's stability, AI Midjourney, Runway and DeviantArt filed a flurry of new motions, including some to dismiss the case entirely. All the companies sought variously to introduce new evidence to claim that the class action copyright infringement case filed against them last year by a handful of visual artists and photographers should be dropped entirely and dismissed with prejudice. Trying to sum up a very complicated case, VentureBeat writes. In a nutshell, the artists argue in their lawsuits that the AI companies, by scraping the artworks that the artists publicly posted on their websites and other online forums, or obtaining them from research databases and using them to train AI image generation models that can produce new, highly similar works is an infringement of their copyright on said original artworks. Continuing, VB writes, the company's new counterargument largely boils down to the fact that the AI models they make or offer are not themselves copyrights of any artwork, but rather reference the artworks to create an entirely new product image generating code, and furthermore, that the models themselves do not replicate the artist's original work exactly, and not even similarly unless they are explicitly instructed or prompted by users to do so. Furthermore, the companies argue that the artists have not shown any other third parties replicating their work identically using the AI models. Now on whether these are good arguments, the VentureBeat author makes clear to remind us that they are not a lawyer, but they say, I do think the latest filings from the web and AI companies make a strong case. And they and that's broadly the sentiment that I've seen on Twitter X as well. Now again, this is just the next chapter in what is going to be a very ongoing saga, so we will just continue to keep an eye on it. Over from the adoption files A new study from a college job search site suggests that Gen Z are thinking that AI could be a way for them to get ahead in the workplace. Axios writes. AI native Gen zers are comfortable on the cutting edge. From the survey, a third of this year's seniors and more than half of tech majors said they plan to use generative AI in their careers. Members of Gen Z are more likely to want to learn AI skills than boomers by 1.6. X or Gen X by 1.1. X, according to LinkedIn research as well, said Valerie Capers Workman, the chief legal officer at Handshake. We're not seeing nervousness among Gen zers. They see it as an opportunity to be on the cutting edge of a transformational technology. They are actually digging in on AI, she said. The most important headline is they are hyper aware that it is mission critical for them to be able to have the best opportunities in the employment space globally. A LinkedIn survey found that 48% of Gen Zers and 52% of millennials, quote, believe that AI will help move their career forward by providing faster access to knowledge and insights which will help them be more confident at work. Indeed, what we've seen so far too, is that one of the places that AI has the biggest impact is with early career professionals who catch up faster to their mid and late career colleagues with the help of generative AI. Now, given how much of AI policy this year is going to be shaped by AI sentiment, I'm always interested to see this type of survey A couple Quick Feature and Product Updates Adobe and TikTok have announced that TikTok's AI powered creative assistant is now available within Adobe Express, writes TechCrunch. With Adobe Express, creators have access to templates, Adobe stock video clips, audio stickers, and an Adobe Express TikTok video creator. With the new Creative Assistant add on, creators will get access to TikTok insights about trending hashtags and AI powered tools all within Adobe Express. If, as we've discussed many times on this show, one of the big themes of 2024 is integration, this is a great example of how companies are putting their AI tool directly into a context that is for a specific use case or purpose. In this case, TikTok video creation another really interesting feature announcement comes from 11 Labs and is actually more than just a feature. 11 Labs is now opening up a marketplace by which voice actors can create a clone of their voice, which can then be used in projects from users around the world and make them money. As it happens, 11 Labs writes, here's how it works. Head to the Voice Lab and upload 30 plus minutes of audio, name and describe your voice, set your price and usage parameters, add your payment details and get paid when it's used. The response to this will almost certainly be a Rorschach test for people's thoughts about AI. On the one hand, people will be rightly concerned that the companies that choose to use this AI generated version of voices are companies that previously might have paid those voice actors for them to create those voices directly. And almost certainly they get paid less for this AI version than they would for performing something themselves. The counterargument is that this might open up entirely new buyers that would never have bought a voice actor before, but who are now financially able to do so. And so thus this opens up an entirely new stream of revenue that doesn't necessarily diminish that direct voice acting either. Ultimately, of course, this is the type of thing that we will just have to see play out in practice. But from 11 labs standpoint, it seems like a pretty cool, interesting and valuable update that could create some real interesting marketplace network effects around the platform. However, that is going to do it for today's AI Breakdown brief. Next up, the main AI breakdown. Welcome back to the AI Breakdown AI deepfakes were on just about every politico's watch list for a big 2024 issue. Given that we have the presidential elections here in the US and other elections around the world, it seemed like this was going to be a thing that we were going to deal with already. It started in some ways. Just before the New Hampshire primary, many of that state's Democrats got a phone call from someone that sounded like President Biden telling them not to go out and vote because their vote didn't matter until November. Whether it was that particular instance or something else, you knew that there was going to be a lot of discussion around how to deal with the issue of deep fakes, and now it appears the White House is having an explicit conversation around that. What makes this interesting is that if you ask the crypto community around where there is or if there is an intersection between the artificial intelligence world and the crypto world, one of the things they will often point to is the idea that you can use blockchains, which are effectively public truth machines, as a way to cryptographically verify what's real and what isn't. To put it simply for those who haven't spent a ton of time in that world, one way to think about, for example, the blockchain that underlies bitcoin is just as a shared public ledger. That is an agreed upon assessment of who owns what within that system. The whole purpose of the bitcoin ledger is for people to be able to agree that I own X bitcoin while you own Y bitcoin, without there having to be an intermediary who says that it's so. The idea then of using some sort of similar approach, although not necessarily exactly the same, to certify when something is real has appealed to a lot of people. Well, now it isn't just crypto people who are talking like that, but in fact the White House itself. Business Insider published a piece at the end of last week titled the White House wants to cryptographically verify videos of Joe Biden so viewers don't mistake them for deepfakes. This came after the FCC declared AI generated robocalls illegal last week. But of course the whole issue with this is that legality doesn't necessarily stop these deepfakes from having a big issue in practice. President Biden's Special advisor for Artificial Intelligence, Ben Buchanan, said that one of the ways that they're trying to address this is to verify all official communications from bi. Buchanan said the aim is to essentially cryptographically verify everything from the White House, whether a statement or a video. Buchanan pointed out that one of the provisions of the executive order last year had created an AI Safety Institute at the Department of Commerce that had at the core of its mission to create standards for watermarking content to show provenance. Now, this White House effort is apparently separate and more specifically focused on official White House communications, buchanan said. This is a case where we recognize the potential for harm. We're trying to get ahead of it. So of course right now this is just an open conversation. But it matters, I think, more than just whether we'll get clarity around whether something is actually President Joe Biden or not. It also matters because to the extent that the White House figures out an approach for this that works for them in this particular instance, I think that you could see that then be more broadly influential. In other words, others might want to take seriously this idea of cryptographically verifying information or videos or photos or any sort of information that can be faked by AI. And at the same time, it might also be a catalyst for helping consumers understand how to do this sort of verification on their end, how to demand actual proof before they accept something as real. Ultimately, it is going to be a weird melange of consumer expectation shifts, plus new technological approaches, plus, of course, the legal system that nudges us towards a place where we've adapted to the world of deepfakes. But the White House making it an issue certainly increases the speed with which that might happen. Now, the US is of course, not the only country dealing with these sort of issues. One other really interesting case study in how AI can impact politics is happening over in Pakistan right now. I'm John Massa, the CEO of Replit. Tweets cyberpunk moment Pakistan's ousted, slash jailed former PM have been using AI to campaign for his party from behind bars and now uses AI to deliver his victory. The New York Times coverage of this piece was titled Imran Khan's Victory Speech from jail shows AI's peril and promise. So without having to go too deep into the political situation in Pakistan, the former PM Imran Khan has spent basically the entire electoral campaign in jail. Yet in spite of this, he's been using AI to replicate his voice, which was part of a strategy that was deployed by his party to, as the New York Times put it, circumvent a crackdown by the military. The Times goes on. On Saturday, as official counts showed candidates aligned with his party, pti, winning the most seats in a surprise result that threw the country's political system into chaos, it was Mr. Khan's AI voice that declared victory for The Times. This is actually an example of how AI can be used to prevent political crackdowns, they write, as concerns grow about the use of artificial intelligence and its power to mislead, particularly in elections, Mr. Khan's videos offer an example of how AI can work to circumvent suppression. However, as they point out, it still has the net effect of, as Toby Walsh put it, undermining our belief in the things we see and hear. Now, to me, this isn't a particularly compelling line of argument about why a technology is bad. If our entire discourse about any new technology was, well, it can be used for good or bad, so therefore we should be worried about it, we would be worried constantly about every new technology. Technology is of course neutral, and that doesn't mean the good or the bad use cases are equivalent, and that is of course worth discussing. But not engaging with it because sometimes it might be used for bad ends doesn't undermine the good that can be done with it as well. What's more, an angle that I'm surprised I'm not seeing more from the mainstream media coverage of this is the fact that while the concern around deepfakes is that they undermine people's understanding of what's real and what's fake, in this case we have an instance where everyone knows that artificial intelligence was used to make these speeches and these videos. Their purpose was not to deceive people into thinking that somehow Imran Khan had access to broadcasts from behind bars, but instead to galvanize a party in a public with a voice of a leader that otherwise would be silenced. I think one could make the argument that that sort of introduction to artificial intelligence will actually increase people's acumen about the technology. In other words, get a chance to understand its power without being deceived by it, but also getting familiar with what it sounds and feels like when receiving a message from it. Now these issues are very live and ongoing. Another piece from the Washington Post today writes, AI companies agree to limit election deepfakes but fall short of a band. The Post writes, leading AI companies are planning to sign an accord committing to developing tech to identify, label and control AI generated images, videos and audio recordings that aim to deceive voters ahead of crucial elections in multiple countries this year. The agreement developed by Google, Microsoft and Meta, as well as OpenAI, Adobe and TikTok, however, does not ban deceptive political AI content. Basically, it's an agreement, it sounds like, to try to de risk this sort of content without banning it entirely. This is, of course, in the spirit of the voluntary pledge that the White House got many AI companies to sign last July, where they, among other things, committed to try and identify and label fake AI content on their platforms. The question, of course, is whether this will be enough or whether there will be continued pressure to put even more stringent requirements on how AI shows up on the distribution channels. Interestingly, basically every part of the world right now is contributing to the AI conversation in some way. The EU continues to proceed with its AI act, with lawmakers ratifying a recent political deal on the AI act rules. And over in the UAE, OpenAI's Sam Altman has said that that country could be an AI regulatory testing ground. In a virtual appearance at the World Government Summit yesterday, Altman said, it's very hard to get all the regulatory ideas right in a vacuum. If there was a contained way that I could give people the future and let them experiment with it and then see what makes sense, what went wrong, what went right. That seems like an interesting experiment. I think for a bunch of reasons, the UAE would be set up to be a leader in the discussions about that. Now the uae, of course, is in a very interesting political spot relative to the geopolitics of AI. One company based in that area, G42, for example, has been at the epicenter of the battle between Washington and Beijing and just yesterday told Bloomberg that the company is scaling back its presence in China in an effort to appease scrutiny from dc. So whether it intentionally wants to be this sort of sandbox for AI regulation, or whether it just happens to become a geopolitical staging ground, it does seem like the Emirates are going to have an interesting role to play anyways. Lots and lots of interesting things happening. As always when it comes to AI society, elections, politics, it is something I am watching closely and will continue to do so. For now, however, that is going to do it for today's AI breakdown. Until next time, peace.