Andrej Karpathy Tries to Rebrand AI

0:00Today on the AI Breakdown, NVIDIA is getting prepared to manufacture chips that come in under US governmental restrictions for the Chinese market. But does the Chinese market even want them? Before that on the brief, a new attempt to rebrand AI by OpenAI's Andre Karpathy. 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 newsletter, and our YouTube channel. Welcome back to the AI Breakdown Brief, all the AI headline news you need in around five minutes. We start today with something that may seem like idle chatter on Twitter slash X, and really it probably is, but I do think it's reflective of a broader struggle to define the parameters of what this artificial intelligence industry really means and what its purpose is in the world.

0:50Now, the genesis for all of this was a post from Simon Willison called It's OK to Call It Artificial Intelligence. Simon writes, We need to be having high-quality conversations about AI, what it can and can't do, its many risks and pitfalls, and how to integrate it into society in the most beneficial way possible. Anytime I write anything that mentions AI, it's inevitable that someone will object to the very usage of the term. Strawman. Don't call it AI, it's not actually intelligent, it's just spicy autocomplete. That strawman is right, it's not intelligent in the same way that humans are. And spicy autocomplete is actually a pretty good analogy for how a lot of these things work.

1:26but I still don't think the argument is a helpful contribution to the discussion. Now basically what Simon goes on to say is that when it comes to functionally describing and getting the reader into this broad area and field that we're talking about when we talk about AI, that popular designator matters more than precision of scientific definition. And that's certainly something that I agree with as well. Now one other piece of this however is that he suggests a distinction between what we talk about as artificial intelligence right now and AGI. Simon writes, If we're going to use artificial intelligence to describe the entire field of machine learning, generative models, deep learning, computer vision, and so on, what should we do about the science fiction definition of AI that's already lodged in people's heads?

2:06Our goal here is clear. We want people to understand that the LLM-powered tools they are interacting with today aren't actually anything like the omniscient AIs they've seen in science fiction for the past 150 years. Thankfully, there's a term that's a good fit for this already, AGI for Artificial General Intelligence. This is generally understood to mean AI that matches or exceeds human intelligence. AGI itself is vague and infuriatingly hard to define, but in this case I think that's a feature. ChatGPT isn't AGI is an easy statement to make, and I don't think its accuracy is even up for debate.

2:35The term is right here for the taking. You're thinking about science fiction there. ChatGPT isn't AGI like in the movies. It's just an AI language model that can predict next tokens to generate text. Now, this is where Andre Karpathy from OpenAI comes into the conversation. He quote tweeted Simon's X post about this and said, I'm playing around with calling our tech as it is today, IA, intelligence amplification, instead of AI. IA have the vibe of tools for thought, needing human interaction, and resemble a lot more of what we actually have today. AI feels more like independent, long-running agents.

3:08Now, Andre went on to expand that, writing E slash IA, intelligence amplification. Does not seek to build super intelligent god entity that replaces humans, builds bicycle for the mind tools that empower and extend the information processing capabilities of all humans not a top percentile and faithful to computer pioneers. Now, why would this matter? We are right now in the midst, really just the beginning, but still in the midst of a very important societal conversation around what we actually want when it comes to AI. From the technologist side, there's sort of an a priori belief that we always want more and better technology, and that more and better and more advanced technology is always a net benefit to humanity.

3:49This is, very roughly speaking, the accelerationist side of the conversation. On the other hand, when it comes to AI, there is serious pushback. There are more questions now than we've had about just about any of the recent technology movements around how much it's actually serving the interest of people, and whether there are reasons to want a meaningfully slowed or even neutered version of this technology. When it comes to that type of debate, language actually matters. Being able to easily say we want this type of a thing but not that type of a thing and having names for each of those things that make them easily communicable can be the difference in getting broad-based public support.

4:22It may be that Andre Karpathy trying to define a term that makes it clear that it is quote not seeking to build super intelligent god entity that replaces humans is actually much more significant in the future than it seems right now. Anyways, I think it's a fascinating little undercurrent. It's something that I'm going to be watching. I don't anticipate this IA shift to make a big splash or anything. It's more the underlying narrative debate that I see it representing that has me interested. And now to the rest of our stories for today's brief. Staying in the realm of OpenAI for just a moment, Microsoft has picked their executive Dee Templeton to join OpenAI's board as an observer.

4:55You'll remember that as part of the agreement that brought Sam Altman back to the firm, Altman and his ally Greg Brockman agreed not to join the board at least initially, and Microsoft agreed not to have a formal board member, but they would get board observer rights. In other words, the ability to sit in in a non-voting capacity, but at least have visibility into everything that's happening on the board. Dee Templeton has been at Microsoft for more than 25 years and is currently the company's VP for Technology and Research Partnerships and Operations. Right now, the voting board members on OpenAI's board are Brett Taylor, who was formerly the co-CEO of Salesforce and before that at Meta, Larry Summers, who's the former U.S.

5:29Treasury Secretary and President of Harvard, famously the President of Harvard when Mark Zuckerberg was going through that school, and Adam D 'Angelo, who's the CEO of Quora and who was the only holdover from the previous board. It remains an open question how large this new board will be, and some see it as an interesting proxy fight for Sam's authority within the company in this new version. Now, from a product perspective, OpenAI continues to scream ahead. News reported in TechCrunch today, ChatTPT is finding its way into more vehicles, this time in Volkswagen cars and SUVs. Announced as part of CES, which goes off this week in Las Vegas, starting in the second quarter of this year, ChatGPT will be integrated into Volkswagen's onboard computer for many European models, although they won't be initially in the US.

6:11Now, VW is not the first auto manufacturer to put ChatGPT in cars. That distinction goes to Mercedes-Benz, who started experimenting with it last June. Now, one company that is also rolling out a chatbot but not ChatGPT for its employees is Big Four consulting firm Deloitte. Around 75 ,000 of the firm's staff members are being given access to their Pear D tool. Interestingly, this chatbot was not built in collaboration with some external startup or lab, but was developed entirely internally to Deloitte. The goal of the tool is to help increase staff productivity in areas like answering emails, drafting written content, automating code writing tasks, creating presentations, creating meeting agendas, carrying out research, and more.

6:51Now, I tend to see usage of AI inside big consulting firms as a leading indicator of where the enterprise is going in general, because to the extent that those firms actually develop and enshrine new workflows based on the use of AI, the more likely they are to bring those innovations to their clients going forward. You've heard me say it lots of times, but I'm quite sure that 2024 is going to be a year of deep integration when it comes to the enterprise and AI. Last up today, Magic the Gathering creator Wizards of the Coast, which is a subsidiary of Hasbro, is in hot water after using generative AI in a set of marketing materials after explicitly making its policy that its artists could not use AI in Magic the Gathering related work.

7:33Basically, there was a set of images released via Facebook ads that showed cards from the upcoming set inside of backgrounds that were pretty clearly created by AI. When people asked them about this, they said no, it wasn't created by AI. But then they actually found out that it was. On Sunday, they went full mea culpa writing, Well, we made a mistake earlier when we said that a marketing image we posted was not created using AI. As you, our diligent community, pointed out, it looks like some AI components that are now popping up in industry-standard tools like Photoshop crept into our marketing creative even if a human did the work to create the overall image.

8:05While the art came from a vendor, it's on us to make sure that we are living up to our promise to support the amazing human ingenuity that makes magic great. We already made clear that we require artists, writers, and creatives contributing to the magic TCG to refrain from using AI-generative tools to create final magic products. Now we're evaluating how we work with vendors on creative beyond our products like these marketing images to make sure we are living up to those values. So, a couple things here. One, part of the issue is that Magic had taken this stance against AI in its art. It'd be one thing if the company was throwing open its floodgates and saying, sure, we'll use AI everywhere, but they haven't.

8:37They've said that they didn't want AI inside the Magic experience. So, of course, this came off as very hypocritical. Second, they're certainly trying to make it seem like it's a process error and has to do with issues with a third-party vendor, rather than a mistake that they made themselves. They also tried to point to tools like Photoshop as part of the culprit. it. Now, on the one hand, this is a story about community uproar and the negative PR costs associated with using AI for a company that indexes high on creative and artistic material like Wizards of the Coast does. And of course, to the extent that the Magic the Gathering community turns this into a more general uproar, there could be meaningful negative costs associated with using AI that offset any productivity gains that they might have otherwise realized.

9:18At the same time, it certainly feels a bit like a fighting against the tide. And even the artists who are making a stand and trying to say that they won't work with Magic the Gathering anymore have a little bit of an air of having already lost a battle that they're still fighting. Overall, I think it is an interesting case study and something we're going to see a lot more as 2024 develops. For now, however, that is going to do it for today's AI Breakdown Brief. Up next, the main AI breakdown. Quickly, a brief word from today's sponsor. As a listener of this show, I suspect you like to stay up to date on all things AI and tech, Which is why you have to check out the chart-topping podcast, Web3 with A16Z Crypto.

9:55Produced by venture firm Andreessen Horowitz, Web3 with A16Z is the perfect companion podcast to the AI breakdown. Web3 with A16Z Crypto is your definitive resource for the future of the internet. Whether you're interested in the convergence of AI and crypto or simply curious about what's next. If you need a place to start, they recently released an excellent episode with Stanford cryptography professor Dan Bonet and former Google X engineer Ali Yaya in conversation with host Sonal Choksi about the intersection of AI and crypto. From fighting deepfakes and proving humanity to large language models like ChatGPT, they cover it all.

10:28I highly recommend checking it out, especially if you'd like to learn more about how AI and crypto will impact our everyday lives. Beyond crypto and AI, this show is for creators seeking more ways to truly own their work, for business leaders trying to prepare for the future today, and for innovators exploring trending tech topics. Don't miss out. Follow Web3 with A16Z Crypto on Apple Podcasts, Spotify, or your favorite listening app. Welcome back to the AI Breakdown. Once again, NVIDIA is working around or trying to work around US AI chip export restrictions, newly updated as of the end of last year, in order to try to keep China as a going concern and a market for their chips.

11:10The company is planning to begin mass production on their new H20 chip, the most powerful of three China-focused chips that NVIDIA has developed that meets the new restrictions that were announced in October. Now, originally NVIDIA planned to launch these chips last November, but those plans had delays, and there are questions around how big this production run will be at the beginning. Writes Reuters, one of the people said initial production volume will be limited, with NVIDIA set to primarily fulfill orders for major customers. Now, you'll remember that this is not the first time that NVIDIA has gone through this.

11:42The first time that export restrictions were announced was back at the end of 2022, and within about a month of those restrictions, NVIDIA introduced the A800 and the H800 chips, which were at the time alternatives that still met the updated rules. With even further restrictions coming last year, these new chips have been neutered even further. Now, this is part of something that The Economist recently called the era of AI nationalism. They wrote, sovereigns the world over are racing to control their technological destinies. Here's the way that piece kicked off. The hottest technology of 2023 had a busy last few weeks of the year.

12:17On November 28th, Abu Dhabi launched a new state-backed artificial intelligence company, A171, that will commercialize its leading LLM Falcon. On December 11th, Mistral, a seven-month-old French AI startup, announced a blockbuster$400 million funding round, which insiders say will value the firm at over$2 billion. Four days later, Kutrium, A new Indian startup unveiled India's first multilingual LLM, barely a week after Sarvam, a five-month-old one, raised$41 million to build similar Indian language models. Now, basically the gist of this story is that AI is becoming a major vector of geopolitical competition.

12:51And if you've been watching the AI breakdown for some time, this will not surprise you at all. In some ways, this is an extension of what we've seen with other technology movements. Certainly there has been competition around things like communications infrastructure, and a lot of the fault lines in the US-China relationship do run along things like broadband and 5G. But it's certainly getting to a different level when it comes to artificial intelligence. There are a couple reasons for that. One of them is that I think that people have a sense that there are higher stakes than ever when it comes to this technology because of the way in which it potentially erases subcultures and perpetuates a dominant monoculture.

13:30Think about it just from a linguistic perspective. If LLMs start to power business functions in every type of company around the world, and those LLMs are trained predominantly on English language resources that come from American sensibilities, it creates a force that can increase the perpetuation of the cultures that shape the inputs of those LLMs at the expense of those that don't. And thus, wanting AI models trained on the particular artifacts and languages and documents and histories of particular cultures and subcultures becomes even more important. The other piece of this, however, is just the simple fact that people understand that it's likely to be even more significant than previous technology movements in shaping the future of the world.

14:10And yet, sitting on top of all of this is the fact that it has already become another huge wedge issue between the United States and China in what is undeniably the most important superpower competition of the 21st century. Recently, Exhibit A in that superpower competition, and specifically the AI dimension of it, has been Emirati firm G42. Back at the end of November, the New York Times wrote a long, long investigative piece exploring how much U.S. intelligence agencies were concerned with the relationship between G42 and large Chinese companies. Now, one of the big loopholes that was left open in the 2022 version of the White House's AI export restrictions were the international subsidiaries of Chinese companies that weren't targeted by those rules.

14:57Another vector that was left open was, of course, partnerships. And throughout the course of 2023, many U.S. intelligence agencies got more concerned that Chinese companies were continuing to get access to advanced cutting-edge U.S. technology vis-a-vis partnerships with companies, particularly those in the Middle East. In early fall of last year, we started to get reports that the U.S. government, even in advance of new restrictions, was putting pressure on big firms like NVIDIA to curb sales to many of their partners in the Middle East, not because the US government was concerned that the United Arab Emirates, for example, had access to that technology, but because they thought it was getting back to China.

15:32Now, the pressure in the case of G42 got significant enough that the company ultimately had to actually pick a side and started disentangling itself from its Chinese relationships. That happened a few weeks after this article came out and once again dramatized the stakes of this competition. Now at the same time, throughout December, one of the big sources of geopolitical drama around AI was actually centered in Europe. The EU AI Act is the most comprehensive legislation that has been voted upon anywhere in the world, but not everyone in the European Union is very happy with it. French leadership in particular have been very loud in their antagonism towards the act, specifically as it relates to last-minute provisions regarding foundation models.

16:11You've likely heard me talk about this before, so I won't get too deep on it, but one of the things that you have to remember about the EU AI Act is that this has been in the works for many years. Specifically, a lot of the work predates ChatGPT and generative AI emerging on the scene in late 2022. And so one of the questions throughout 23 was how this act would actually handle those updated technology areas. Now, in previous cases, the EU has often punted on certain issues that had come up subsequent to the drafting of legislation, saving it for the future. I'm thinking specifically of the EU's crypto rules, which started to come together in the wake of the ICO boom, and because of that kind of actually contain some old out-of-date language, but which made the explicit decision not to deal with things like DeFi and NFTs in the first round of the MECA regulations because they wanted more time to actually try to go out and figure things out.

17:02That's not exactly been the process when it comes to these foundation models, with many feeling like the technology is going to move fast for them to keep up and that they needed to take this chance to actually move on new rules, but the concern of people like French President Emmanuel Macron is that the EU will become the leader in regulation, but at the cost of not having any of the technology actually developed on its shores. Now, of course, the fact that French company Mistral is such a leader in the open source space is adding extra context to that French fight against the AI Act in its current form.

17:32Even beyond the EU AI Act, it's clear that AI is at the top of the European mind. Reuters recently reported about Italy's upcoming G7 presidency, where AI is set to be a major theme. And then of course going on all around this, beyond any of the diplomatic questions or regulation or policy questions, is the actual AI arms race when it comes to global militaries. It seems like not a week goes by when there isn't some major story about the DOJ or the Pentagon and their particular AI initiatives, leading to, for example, this piece from AP back in November, Pentagon's AI initiatives accelerate hard decisions on lethal autonomous weapons.

18:06It seems to me fairly likely that given that this year is an election year, we may fail to see real progress on comprehensive AI regulation in the United States, but that certainly isn't going to slow down the military at all when it comes to their particular use of the technology. And so ultimately, this brings us back to the story of NVIDIA and these slower AI chips. Business Insider's reporting of this is titled Chinese Companies Aren't Interested in NVIDIA's Slower AI Chips. They write, Alibaba Group, Tencent, Baidu, and TikTok owner ByteDance have all suggested they're planning to order far fewer chips from NVIDIA this year now that its most powerful chips aren't on offer.

18:40By downgrading its chips, there is far less of a performance gap compared with locally made Chinese alternatives, meaning the companies may start to shop at home instead. Chinese firms are also concerned about potential increases to US restrictions. Now, of course, this will read very differently depending on whether you're in the NVIDIA boardroom or whether you're in the White House. The White House's rules are making it clear that they care more about denying China access to these chips than they are about NVIDIA's bottom line, but it's also creating a hell of an incentive for homegrown companies to try to race to catch up to U.S.

19:10capacities. Reuters, for example, had previously reported that Baidu had started to shift their ordering to Huawei technologies instead of NVIDIA. And so, of course, what we're dealing with is a situation where restrictions may work, but for how long and at what cost? It is a really interesting geopolitical moment and one that I'm going to continue to watch closely. For now, though, that is going to do it for today's AI Breakdown. I appreciate you listening or watching as always. Till next time. Peace.