The AI Daily Brief: Artificial Intelligence News and Analysis - Apple's AI Strategy Revealed?

Episode Date: September 13, 2023

At the iPhone/Apple Watch event yesterday we learned about Apple's new, more powerful chip and neural engine. NLW argues that the chip has lots to tell us about Apple's AI strategy. Before that on the Brief, the latest AI announcements from Salesforce, plus AI marketing from Coca-Cola and Disney. TAKE OUR SURVEY ON EDUCATIONAL AND LEARNING RESOURCE CONTENT: https://bit.ly/aibreakdownsurvey 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 talking about how AI sneakily showed up at the iPhone event yesterday. Before that on the brief, Salesforce's Big AI announcement, new AI-related marketing for Coca-Cola, and a preview of the upcoming closed-door summit between big tech CEOs, senators, and civil rights leaders. 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 newsletter, our YouTube, and our Discord. Welcome back to the AI breakdown brief. All the AI headline news you need in around five minutes. Today we kick off with a little bit of a review of Salesforce's Dreamforce conference,

where, as you might imagine, AI was the big theme. Now, there were a number of announcements which we're going to talk through quickly, but I think that even more interesting than the announcements are what they tell us about the state of enterprise AI. Now, Salesforce first introduced an artificial intelligence layer that they called Einstein, all the way back in 2016. In March, just a few months after ChatGPT came out, they introduced an updated version called Einstein GPD

that, as you might imagine, basically allowed users to ask questions about Salesforce and natural language throughout the platform. And if Einstein GPT was Salesforce's first try at bringing generative AI across their entire suite of tools, co-pilot is their next generation, and it goes a lot farther. Clara Shee, who's the CEO of Salesforce AI, said,

we are launching Einstein co-pilot, which is a conversational AI assistant for companies, employees, as well as their customers to securely and safely be able to access generative AI to do their jobs better, faster, more easily, and to augment and amplify their own abilities, their skills, their work, and be more efficient and be more productive. So basically, this is just a more advanced version of a chatbot that lives inside the Salesforce software software. The examples that TechCrunch includes are a Salesforce researching new accounts, a newer customer service rep asking how to deal with a return over 30 days, and a product manager asking how to create a customized storefront for a new product launch. Now, one of the big

advantages for Salesforce is that the Einstein co-pilot can be connected to and powered by data that's actually coming from Salesforce's own data cloud, which they introduced at their Dreamforce event last year called Jeannie. They also introduced something called Einstein Co-Pilot Studio, and effectively this includes three elements, a prompt builder, a skills builder, and a model builder. She further explained. The first piece is the prompt builder, and this is for customers who want to customize the prompt templates that have been included in Einstein GPT. The Skills Builder means that Einstein co-pilot is now no longer just accessing your data and answering questions on the data. Companies also have the ability to control and designate which workflows they want

copilot to have access to and run. And finally, the model builder allows enterprises to bring their own model or use one of the supported third-party offerings from companies including Anthropic, Cohere, Databricks, OpenAI, and Google Cloud's Vertex AI. So it sounds like the model builder part of Copilot Studio is exploring similar space as something like Amazon's bedrock. So taken altogether, what does this tell us about the state of enterprise AI software? I think a couple things. First of all, it's more clear evidence that AI and generative AI specifically are going to be a part of a reimagined enterprise suite basically across every type of enterprise software that exists. Second, I think it shows that the farther into this we get, the more that

these software companies are going to want to offer their customers' customization opportunities. That can be very heavy customization, such as the model builder and co-pilot studio, or it can be the much more lightweight customization such as the prompt builder. The point, though, is that different enterprises are going to have different needs, and so a lot of the value proposition is in and around the tooling and customizability of these offerings. A third trend that I think this Salesforce announcement reinforces is just how well-positioned, frankly, existing B2B software providers are. Basically, because AI requires so much data, if an enterprise understands,

that the more data, including proprietary and private data they give an AI model access to, the better results they're going to get, then when it comes to working with third parties, they're likely to prioritize companies that they already trust with their data, as opposed to, for example, brand new startups. It makes sense then that even these bigger companies are moving so, so fast to meet that demand, because they just don't want to allow an inch for those enterprise customers to have to go look elsewhere for these new solutions. This is one of the first really big technological changes that could be distributed largely through existing businesses, simply because they're working so hard to adapt as fast as they are. Now, one more note from Salesforce, in an interview on

CNBC, CEO Mark Beniof talked a little bit about their plans for Slack and AI. Beniof said, quote, the big news is Slack is really starting to wake up with its own AI. It holds so much data for our customers. I think Slack is going to be the promise of AI for a lot of our most important customers. Now, obviously, any conversational medium like Slack is just an absolute trove of data, and so it's probably worth keeping a close eye on what Salesforce decides to ultimately do with all of that information living in Slack. Now, one of the things that has been notable so far with AI is that it hasn't really made it into the marketing hype cycle, as, for example, NFTs and Web3 did a couple years ago. That apparently is starting to change, and Exhibit A is Coke's new Y-3000

flavor, a Coca-Cola creation that is designed by artificial intelligence. Basically, Coca-Cola asked a number of humans to describe what they thought the future tastes like, and then turned it over to AI to determine what flavor pairings and profiles would work to capture that essence. Now, obviously, this is a gimmick. Obviously, this is just for fun. And of course, it's a good experiment for Coca-Cola to understand what new types of possibilities and differentiated thinking turning something like a recipe over to AI might produce. But it is notable as one of the first big examples of artificial intelligence in this post-chat chip-a-tie-t world really being reduced. to a marketing gimmick.

Speaking of AI and marketing gimmicks, during NFL's week one game between the Chargers and the Miami Dolphins, a number of patrons were surprised to see that they were surrounded not by other people, but by AI-powered robots. This was, once again, a marketing stunt designed to promote a new science fiction movie coming up from Disney called The Creator. The plot of The Creator is focused on a future war between humans and artificial intelligence, and given that the Chargers are a Los Angeles team, playing in Los Angeles's SoFi Stadium, this does kind of seem like a natural place for a Hollywood promotion.

Anyways, I think that to the extent that AI continues to be in the public eye, we can expect ever more AI-related marketing. Lastly, today, the biggest thing happening in artificial intelligence in many ways today is Senator Chuck Schumer's behind closed-door summit that has secured the participation of some of the biggest names in AI and some of the most important big tech leaders to discuss alongside labor leaders and civil rights leaders, the challenges and opportunities at the technology,

and what a good path forward might be when it comes to U.S. policy. But of course, that hasn't stopped some folks from coming in and saying that the whole thing is a bit of a sham. Elizabeth Warren, perhaps the politician least able to leave her priors at the door and engage with new situations from First Principles said, these tech billionaires want to lobby Congress behind closed doors with no questions asked. That's just plain wrong. This, of course, gave mainstream media a new angle to pursue as relates to the summit and run with it they did. Now, we are likely going to cover a lot more of what comes out of that event as we learn more, but just to get a sense of who is participating. We've got Sam Altman from OpenAI, Jack Clark from Anthropic, Clement the founder of Hugging Face,

Bill Gates, the former CEO of Microsoft, Jensen Huang, the CEO of Invidia, Alex Karp, the CEO of Palantir, Elon Musk, the CEO of being Elon Musk, Satya Nadella, the CEO of Microsoft, Sundarpe Chai, the CEO of Google, and perhaps if it was just those CEOs, even if the optimistic or non-cinical take would be that this is a signal of the importance of this issue that so many of these people would change their schedules around to be in the same place at the same time, even if it was just those CEOs, it would be one thing. But it's not. Other participants include Tristan Harris, who's the co-founder of the Center for Humane Technology, who has been beating the drum around AI safety, Elizabeth Schuller, the president of the AFL-CIO, Meredith Stein, the president of the Writers Guild, Randy Weingarten,

the president of the American Federation of Teachers, and Maya Wiley, the president and CEO of the leadership conference on civil and human rights. My two cents and my reminder to Senator Warren and anyone like her that's tempted to take an a priori cynical approach to anything relating to any company touching money or big tech is that this is too important an issue, with too much significance for how society and the economy are going to develop, to just make it a place to wage old wars and capture cynical soundbites. I am going to be watching closely for any any hints of what comes out of this conversation, but I can't imagine a world that isn't better off for this conversation having taken place. That's going to do it for today's AI breakdown brief.

I'll be back soon with the main AI breakdown. Welcome back to the AI breakdown. Yesterday was the big Apple iPhone 15 unveiling, and we also got a new Apple Watch, the Apple Watch Series 9. And of course, if I'm talking about it on this show, you know that I am exploring the artificial intelligence dimension of it. Well, to get a sense for how to think about if and where AI made it into the presentation, I think the title of this Reuters piece about the event does a pretty good job of summing it up. The piece is titled AI quietly reshapes Apple iPhones and watches, and so that's what we're going to be exploring today. Now, to kick things off, let's talk about how our understanding of Apple's approach to

AI has changed over the last few months. How and when exactly Apple was going to jump into the AI race has been one of the big questions in the big tech sphere throughout 2023. In July, we got a little bit more of a peak of what was going on behind the scenes, with Bloomberg reporting that Apple was testing something they were calling Apple GPT, as well as working on their own generative AI tools and LLMs. The piece reads, Apple Inc. is quietly working on artificial intelligence tools that could challenge those of OpenAI, Google, and others. But the company has yet to devise a clear strategy for releasing the technology to consumers. So the framework that Apple was working on, was called Ajax, and the chatbot that they were using internally, they were calling Apple GPT.

According to the Bloomberg article, the push inside Apple for AI had grown in importance in recent months, and what's more, that executives were getting more worried about figuring out Apple's slant on the whole shift, again from the Bloomberg piece. 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 technology. 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 they continue Apple began laying the foundation for AI services with the Ajax framework as well as a chat GPT-like tool for use internally. However, also according to the piece,

Apple GPT doesn't really have any features that any of the other publicly available tools like it don't have, nor any really novel technology. Indeed, they write, Apple is still trying to determine the consumer angle for generative AI. While the company doesn't yet have a concrete plan, people familiar with the work believe Apple is aiming to make a significant AI-related announcement next year. All in all, the piece basically presented a picture of a company that knows that it needs to do something, but that isn't exactly sure what it should be doing. Now, this was kind of reinforced by another report from the information a couple weeks ago. That piece was called Apple boost spending to develop conversational AI, and the big new piece

of news was that Apple was now spending several million dollars per day training their internal models. And on the one hand, even though they were spending more money, the information piece still made it seem like there were big questions internally at Apple about exactly what the Apple's slant on AI would be. From the information, questions linger over how Apple can incorporate LLMs in its products. The company's leaders prefer running software on devices, which improves privacy and performance, as opposed to on cloud servers, but that could be difficult to achieve. Ajax GPT, for example, has been trained on more than 200 billion parameters, and an LLM with more than 200 billion parameters couldn't reasonably fit on an iPhone. So basically, what you have here is a

demonstration of how Apple's core operating philosophies, which are privacy-centric, which focus on on-device versus cloud models, are running up against the particulars of the generative AI space. At the same time, the piece did give some hints about how Apple might start to bring AI into its product line. From the very beginning of the article, one of Apple's goals is to develop features such as one that allows iPhone customers to use simple voice commands to automate tasks involving multiple steps. The technology, for instance, could allow someone to tell the Siri voice assistant on their phone to create a GIF using the last five photos they've taken and text it to a friend. Today, an iPhone user has to manually program the individual actions. So the update,

effectively from that July Bloomberg report was one, a seeming increase in resource expenditure, two, a continuation of the big questions about how to Appleify AI, and three, a sense that at least the initial steps might be focused on improving core experiences on key Apple products like the iPhone. There were some who thought that the announcements from the worldwide developer conference in June also pointed in that direction. A piece in the Atlantic called Apple as an AI company now argued that, quote, lots of tiny AI tweaks are quietly taking over the iPhone. And basically this pointed out that a number of these small feature updates that have been announced, such as a new generation of AutoCorrect, the Photos app being able to recognize

and differentiate between the owner's dogs and other dogs, and AirPods getting smarter about adjusting background noise, all were ultimately AI-powered features, even though that wasn't a term that was used to describe them. Indeed, at the Worldwide Developer Conference, the term artificial intelligence was used a grand total of zero times. And so that brings us back to yesterday's iPhone and watch announcement, and the Reuters piece called AI quietly reshapes Apple iPhones and watches. The piece reads, without using the words artificial intelligence to describe the emerging technology, Apple showcased a new line of phones and a new watch that included improved semiconductor designs that power the new AI features. The features largely improved basic functions like taking a call

or snapping better images. So the big thing that people were focused in on was this new A17 Pro chip and what Apple calls the neural engine that powers it. Atilla writes, 35 terraflops of ML compute in your pocket. On-device inference is getting interesting. Mac Rumors writes with the new neural engine S-9 series requests are processed on-divor. to make them more secure and quicker. Danny Acosta writes, Apple just doubled the AI capabilities in the new iPhone 15, 2X faster neural engine, performs nearly 35 trillion operations per second with 16 cores. They can train a whole ML model in your iPhone that learns about you without even going to the cloud. Okay, so what we have here is quite clearly a hardware update that makes it appear

as though Apple isn't quite ready to give up on its foundational principles just because AI is caught. What I mean by that is that if the problem with on-device models is that they're too big for those devices as they stand, well, then maybe what Apple needs to do is just power up what those devices can do. Let's read Danny Acosta's tweet again with that in mind. Apple just doubled the AI capabilities in the new iPhone 15. 2X faster neural engine performs nearly 35 trillion operations per second with 16 cores. They can train a whole ML model in your iPhone that learns about you without even going to the cloud. In other words, part of why it seems like Apple may not be going headfirst into AI, at least in terms of the marketing of AI,

is that it's still building out the technological capabilities to do AI the way that it wants to, on-device, private, trained in relation to the specific user. But that doesn't mean that the new chip won't come with new features available right away. As ours technical rights, the neural engine offers big boost to on-device processing for Siri requests, including 25% faster voice dictation. The other thing that the neural engine enables is a new gesture, that once again, as ours technical puts it, Apple claims watch users will be using every day. So what is that gesture? It's something that Apple is calling double tap and involves a user tapping their index finger and their thumb of their watch hand together twice. In other words,

without having to touch the Apple Watch's display. This will allow users to perform basic tasks, such as answering a call. Apple writes, the new double tap gesture is enabled by the faster neural engine in Apple Watch Series 9, which processes data from the accelerometer, gyroscope, and optical heart sensor with the new machine learning algorithm. The algorithm detects the unique signature of tiny wrist movements and changes in blood flow when the index finger and thumb perform a double tap. So basically, the watch understands how your wrist changes when you do that double tap, and that's a feature that's now available because of the faster neural engine. Now, some users were underwhelmed by this. Cnet writes, in my brief time using the series nine, I used double tap

to scroll through widgets, answer a phone call, start a timer, and toggle the flashlight. It worked accurately most of the time, but there were times when I had to perform the gesture more than once to get the watch to respond. Haptic feedback and a tiny symbol at the top of the screen let you know the double tap is working. It may not be a game changer, but double tap could be useful for dismissing alarms and answering calls when my hands are full. Now, this is a completely reasonable review from an observer trying to be objective and thinking like a general consumer would, but for our purposes, it's hard not to see it as the beginning of a break in how we interact with AI-powered devices made possible by more advanced machine learning and more advanced chips for actually doing

the inference on device at a speed which works for real life. As we talked about in the brief today, one of the big themes of AI at the moment is the sort of privatization and personalization that is a necessary part of the development of the field as people understand and value that the most useful AI models are the ones that know the most about them or their businesses, but that that involves a lot of risk in the form of giving access to personal data. While it may not be ready for primetime yet, the fact that Apple is racing to build devices that are actually prepared to handle the workload of truly private, personalized, AI-powered features is something that could be pretty game-changing. For now, the value may be faster Siri and a better ability to hold pumpkin

spice lattes while using your watch to interact. But in the future, the implications are obviously a lot bigger. Anyways, guys, let me know what you think in the comments or come join us on the AI Breakers Discord. The discussion happens at bit.ly slash AI breakdown. Thanks as always for listening or And until next time, peace.