The AI Daily Brief: Artificial Intelligence News and Analysis - Apple Reveals It Spent Billions on AI R&D

Episode Date: August 7, 2023

During last week's earnings call, Apple CEO Tim Cook said that Apple anticipated putting generative AI into every product, and intimated that they had spent billions on artificial intelligence research and development. Before that on The Brief: after the community questioned some art from an upcoming book, Wizards of the Coast and D&D Beyond are clarifying their AI art policies; Morgan Stanley says AI bubble may be peaking, University College London study says 27% can't recognize deepfakes, and more.  Today's Sponsor: Supermanage - AI for 1-on-1's - https://supermanage.ai/breakdown ABOUT THE AI BREAKDOWN The AI Breakdown helps you understand the most important news and discussions in AI.  Subscribe to The AI Breakdown newsletter: https://theaibreakdown.beehiiv.com/subscribe Subscribe to The AI Breakdown on YouTube: https://www.youtube.com/@TheAIBreakdown Join the community: bit.ly/aibreakdown Learn more: http://breakdown.network/

Today on the AI breakdown, we're looking at new information about Apple's AI strategy. Before that on the brief, Morgan Stanley says Bubble and Dungeons and Dragons is changing its AI artwork policy. 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. Hey, hello friends, welcome back to the AI breakdown. One additional note before we dive in. Sponsorship is officially open on the AI breakdown. I am filling out slots for the fall right now,

and so if you have an AI project or know of an AI project that is looking to get in front of what is easily the smartest AI listener base in the world, either DM me or send a note to sponsors at Breakdown.network. Looking forward to working with some great companies, and now let's get to the show. Welcome back to the AI Breakdown Brief. All the AI headline news you need in around five minutes.

Quick apology for those of you watching on YouTube, Descript decided to delete these files, so I am doing a redone version that is audio only. We'll be back to our normal visual tomorrow. Today we are talking deep fakes, Congress learning about AI, Morgan Stanley warning of a bubble, and a big Dungeons and Dragons dust up. We kick off with a new study from University College London

that argues basically that deepfakes are getting too good for people to realize. The TLDR is that using a text-to-speech program, the researchers from UCL created 50 deepfake speech samples in both English and Mandarin. Those samples were then played for over 500 study participants, and in around 27% of the cases, the listeners thought that the speech was not from AI, but was in fact real.

Techniques the researchers used to potentially increase the recognition of those samples didn't really make a big difference, nor did listening to the clips several times or listening to shorter clips. Now, on the one hand, you might be thinking to yourself, huh, if people couldn't recognize deep fake speech 27% of the time, that means they could 73% of the time. That's still firmly in the majority, right?

Well, researchers pointed out two things. First, this is a category of technology that is getting better constantly, so that 27% fake-out rate is pretty much destined to do nothing but go up. Second, they point out, it means that if you're just thinking in terms of something like phone scams, one in four could be successful. Putting a pretty fine point on their conclusions, the study is titled, Warning, humans cannot reliably detect speech deepfakes. Basically, the conclusion that researchers come to is that humans cannot be relied upon to detect AI, so, so we need automatic detectors instead. Unfortunately, the crowd's overall performance in these studies was on par with the best automated detectors we currently have and was in fact less likely to fail as circumstances changed.

So right now, it's not clear how viable the automated detection solution really is. Next up, as regular listeners of this show know, artificial intelligence has jumped onto the U.S. legislative agenda in a major way. To help with those efforts, Stanford University is hosting a boot camp this week. The boot camp is designed for congressional staffers to get up to speed across a variety of AI-related issues. In fact, it's literally called Congressional Boot Camp on AI and is hosted by the Stanford University Institute for Human-centered Artificial Intelligence. The experience promises a multi-day bipartisan boot camp consisting of many sessions unpacking what AI means for international security, future of work, health care, and it also includes field trips to Stanford labs and interactive

experiences. Stanford says that the program is for mid-to-senior-level congressional staff, with a, quote, keen interest in AI and emerging technology policy. Now, I think this is a hugely important type of effort. Staffers are, of course, the front lines for any elected official. They help them figure out what to think about new issues and serve effectively as an extension of them as they have to deal with such a diverse array of issues in any sort of elected role. Giving those staffers resources, then,

to get up to speed as much as possible and as quickly as possible, especially around something as fast-moving as AI, seems really valuable. In fact, the need for AI education is so paramount right now in Congress that Senate Majority Leader Chuck Schumer is also planning a crash course for senators this fall. The plan includes at least nine forums with experts joining to cover issues including copyright, national security, workforce issues, existential risk, transparency, and of course the implications for elections and democracy. So here's hoping that all these educational resources help our elected officials be in a better position to actually regulate what is a very dynamic and fast-moving space. Now, moving over to the market side of the house, we have talked endlessly about how important artificial intelligence has been as a theme to public markets this year. Never was this more notable during the debt ceiling debates when even the impending default of the U.S. government on its debt, which would have been a first, couldn't break the market's enthusiasm for AI stocks. Now, of course, leading that rush has been in video, which is up more than 200% on the year.

And of course, any time you see a major stock go up that much in such a short period of time, you're going to have people start to ask, if there's a bubble forming. According to a new note from Morgan Stanley, if we are using Nvidia as the proxy, then the answer may be yes. Morgan Stanley analysts point to historic precedent in making their argument. They note that bubbles tend to rally a median 154% in the three years pre-peak, and given that Nvidia is up over 200% this year, that might mean that this stock surge is in its later innings. Now, one counterweight to that is that not all AI stocks are up that much. One broader AI benchmark, the MCSI USA Robotics and AI Select, net USD index is up a kickbut but not yet bubbly 47%.

The only thing that's clear right now is that a lot of people are going to make money and a lot of people are going to lose money betting one way or the other on this very hot space. Lastly, today, some interesting intrigue from the world of Dungeons and Dragons. And actually, this is the second time in the last couple of weeks that AI and D&D have been uttered in the same breath. The first time came in a recent partnership announcement where Hasbro, which is the company that owns Wizards of the Coast, which is the company that owns Dungeons and Dragons, talked about how a new partnership would allow them to put AI-driven game mechanics into their

properties, including D&D. Their SVP of Gaming, Adam Beale, said that AI could help them create dynamic experiences that can instantly react to player decisions. Now, if there was a little bit of grumbling from the D&D community around that one, the last couple days have seen a much bigger uproar. The issue surrounds art for a new D&D book called Bigby Presents Glory of the Giants. Some intrepid internet detizens, noticed that there might be extra fingers on one of the giants in the source book and started to ask whether AI had been used to create the image. Basically, when questioned, the artist, Ilya Schippen, said that he had used AI as part of the creative process, but that it was only to enhance details. It wasn't like he entered a prompted mid-jurney and this is what came out.

Now, it's worth noting that this is an artist who was already maybe out of sync with many others in the D&D community and, frankly, the gaming communities in general, who have been extremely hostile to changes in those spaces, such as, for example, NFTs. Ilya is known for using AI art generators. He has an NFT collection on Super Rare. And in many ways, there are really two different issues going on here. One is around disclosure, a source at Wizards, for example, told Gizmodo that they weren't aware that I was using AI as part of his process.

And further, it certainly wasn't something that they disclosed in the book. And the second is around whether they'll allow AI to be used in the D&D art creation process in general. They ended up issuing a statement. Today, we became aware that an artist used AI to create artwork for the upcoming book, Bigby presents Glory of the Giant. We've worked with his artist since 2014, and he's put years of work into books we all love. We weren't aware of the artist's choice to use AI in the creation process for these commission pieces, and we have discussed it with him. He will not use AI for wizard's work moving forward.

We're revising our process and updating our artist's guidelines to make clear that artists must refrain from using AI art generation as part of their creation process for developing D&D. art. Now, this is obviously a very heated topic that gets not just at Dungeons and Dragons, but is about how AI is impacting art in general. The comments on this one are all over the board. Some are the standard rage at corporations. Some are actually questioning where the lines are. Sergio Alonzo writes, we all use AI when doing art, already smoothing, brush color, correction, even upscaling, downscaling. I'll use some kind of AI technology. What's your definition of AI art? Others, however, think that this is inevitably a battle that AI will not lose. Pixie.Eath writes, you'll change your mind in about one year. Anyways, that's going to do it for today's AI breakdown brief. If you're

interested in talking about Dungeons and Dragons and AI Art, come join us on the AI breakdown. You can find it at bit.ly slash AI breakdown, and I would love to see you there. Thanks as always for listening or watching, and I'll be back soon with the main AI breakdown. Before we get into the main AI breakdown, I want to tell you about today's sponsor, Supermanage. If you work in a professional setting, you probably have some version of a one-on-one meeting, either with the people that work for you or the people that you work with. Unfortunately, all too often, those one-on-one meetings become glorified catch-up calls. Don't you wish you could jump right to the stuff that really matters? That's where Supermanage comes in. Supermanage AI magically

distills your team's public Slack channels into a real-time brief on any employee, any time. Catch up on contributions, work in progress, challenges their facing, sentiment, everything you need to show up ready for a truly meaningful conversation. And it's completely free. Visit supermanage.a.i forward slash breakdown today to start making the most of your one-on-ones. And thanks again to Supermanage for sponsoring the AI breakdown. Welcome back to the AI breakdown. Today we are talking about new revelations about Apple's AI strategy and how it relates to what the other big tech companies are doing. Now, for those paying attention, Apple has been the most quiet of the big tech companies when it comes to their strategy around generative AI. Indeed, at the beginning of June, Apple held its big worldwide developer conference where they unveiled their Vision Pro. Generative AI was conspicuously absent from that event, so much so that Wired's headline about it was Apple Ghosts the Generative AI Revolution.

Wired writes, yesterday Apple announced new features powered by its neural engine hardware, like call screening that transcribes the first few words of a voicemail live, you can decide whether to pick up a call, but there was no mention of generative AI during the two-hour worldwide developer conference keynote address. The only thing that came close was an update to a feature in iOS 17 that suggests the next word you might want to use when typing on an iPhone keyboard. First Mark Capital investor Matt Turk said, the reality is that Apple is a bit behind others like Microsoft and Google in generative AI, so it smartly chose to position itself as running its own race in AI, as opposed to trying to play catch-up with others. Others have basically said, look, Apple is always late to the party. What matters is what they bring when they finally show up.

However, we also learned a couple weeks ago that quietly Apple had been doing more than it had appeared publicly when it came to generative AI tools. Bloomberg reported that Apple had been quietly testing internally something called Apple GPT. It was built on an LLM foundation that Apple called Ajax, and at the time of the report, internally people had been using it for a variety of different purposes. According to the reporting, however, Apple was very, very unclear about exactly what they wanted to do with it, this despite the fact that it had become a major effort with several teams collaborating on the project. Some of the issues included potential privacy concerns, as well as wanting not to just be another also ran. Bloomberg wrote, behind the scenes, Apple has grown concerned

about missing a potentially paramount shift in how devices operate. Generative AI promises to transform how people interact with phones, computers, and other technologies. And Apple's devices, which produced revenue of nearly 320 billion in the last fiscal year, could suffer if the company doesn't keep up with AI advances. That's why Apple began laying the foundation for AI services with the Ajax framework, as well as a chat GPT-like tool for use internally. Now, at the time of this reporting, Apple employees said that the company's tool doesn't really have anything different from a novel feature perspective

as compared to barred chat GPT or Bing. It didn't even really have a UX, being accessible only as a simple web application. Now, to get a sense of how much Apple's effort, in this space were anticipated. On this reporting alone, Apple's shares went up 2.3% to a new record high. What's more, Microsoft, who of course is OpenAI's biggest partner, went down 1% on the news. And as much as Tim Cook has tried to brush off questions around their AI strategy from the market, it seems increasingly like that is no longer viable given where Wall Street has its attention focused. Which brings us back to last week's earning call, where AI got a big center stage.

One part of that was a pretty significant hint that Apple has been spending heavily on AI-related research and development. Apple reported that so far, R&D costs have hit $22.61 billion for 2023. That figure is over $3 billion higher than the R&D spend was at this time last year. Cook pointed to generative AI as one of the big reasons. The CEO told Reuters, quote, We've been doing research across a wide range of AI technologies, including generative AI for years. Obviously, we're investing a lot, and it is showing up in the R&D spending that you're looking at. What's more, Cook said during the earnings call that generative AI was, quote, absolutely critical to us,

and that they viewed AI and machine learning as, quote, integral to virtually every product. Now, part of the reason that there may be a bit of a shift in the communication strategy around AI is the need to compensate for otherwise declining revenue in sales. Q3 earnings were down, and in particular, iPhone sales fell short of what analysts had been expecting. outside of the earnings call, the other place that analysts are reading the tea leaves around Apple's AI strategy has to do with their hiring. Over the weekend, the Financial Times reported that Apple is seeking to bolster its expertise in generative AI on mobile devices. FT points out that Apple is hiring for dozens of roles across offices in California, Seattle, Paris, and Beijing to work on LLMs with a specific focus on compressing models so that they can run efficiently on mobile devices. There are a couple big benefits of that, which get to how many think Apple may be trying to distinguish.

distinguish its particular strategy. FT. writes, The benefits of running AI software on phones without the need for an internet connection or to send data to the cloud are that apps can run more quickly and allow a user's data to be processed in a more secure and private way. One job posting is looking for a senior software engineer to, quote, implement features that compress and accelerate LLMs in our on-device inference engine. Another says that they're looking for someone to bring, quote, state-of-the-art foundation models to the phone in your pocket, enabling the next generation of machine learning-based experiences in a privacy-preserving way.

Apple has of course made a huge investment over the last decade in being the big tech giant that cares about your actual privacy. It's a way to hit at competitors like Google and Facebook for their ad-driven models, which require by definition the capture and accessing of lots of information about users' preferences, behavior, and interests. Because Apple is selling devices and services, they don't require the same sort of data from users, and they've leaned into that trying to build products that are inherently more privacy preserving. Now, one of the big ways that people see the AI space evolving is that instead of people just using chat GPT or other third-party LLMs, it feels likely to many that will have some form of personalized model that's trained specifically on our particular data. You're already seeing startups try to build that now, such as Quiverr, which I've talked about here before, and many think that Apple is in an extremely privileged place to potentially make a play in that area. That would certainly give it that differentiation that changes what it's offering as opposed to Google and meta, who are developing their own foundation models.

and Amazon who's making a bet with their bedrock service, that enterprises are going to want to have access to many different models that they can customize for whatever proprietary purposes they want to use them for. And so while everything isn't totally clear yet, you really are starting to see how different big tech companies' AI strategy is coming to the four, and in particular how it's changing or validating how they've operated in general. Microsoft is, of course, trying to integrate AI everywhere faster than anyone else in a way that in some cases takes advantage of the lock-in they already have with professional users,

such as the $30 cost for co-pilot inside the office suite, and in other cases is trying to use it as a way to leapfrog their products that have been behind, such as Bing AI directly in the browser. Google is likewise trying to put AI everywhere, and is in the midst of figuring out how effectively to disrupt itself by changing the search experience that has been core to the Internet for so many years. Metastrategy is one of the most interesting in that it's a big shift to the open source world, where meta is just constantly pushing foundation models and other models like audio craft,

which I talked about last week, out into the public and trying to get people to build on top of them. Now, how much value that ultimately means accrues to meta is, I think, an open question, but it's certainly notable as a strategic approach. Amazon is staying out of the fray of foundation model building, and instead leaning hard into the idea that enterprises are going to live in a multi-model world. Bedrock is their approach to meet businesses where they actually are without forcing them into one domain or another. And then there's Apple. And so far best we can tell, it seems like the play might be to take advantage of all the things that actual hardware devices allow for in terms of local data while still preserving privacy to offer that personalized LLM experience of the future. However, for now, that is all speculation. But if markets keep up pressure on Apple, we might get some more hints a little bit faster than Apple might otherwise be comfortable with. Anyways, guys, that is going to do it for today's AI breakdown. Thanks as always for listening or watching. And until next time, peace.