The Danger of an AI Counterreaction

Today on the AI Daily Brief. Could AI lead to an increase in state power? The AI Daily Brief is a daily podcast and video about the most important news and discussions in AI. To join the conversation, follow the Discord link in the show Notes. Hello friends. It is time for a long read. Today I am excited to share the first ever crossover long reads. In other words, the first person to have their work read on both the breakdown and now the AI Daily Brief. And this piece comes from early Core Weave investor Nick Carter. Piece was published on Twitter earlier in the week and gets at something that I think feels like a very natural implication that's still kind of under discussed when it comes to state power and artificial intelligence. So let's read the piece and yes, you will be able to tell from my covidty voice that it is actually me reading at this time, not 11 labs me. And then we'll discuss it after that. Nick writes, this Nvidia rally has gone from incredibly impressive to actually scaring me a bit. Not for AI safety reasons, I'll explain. I'm lucky enough to be an early investor in Core Weave, one of the most incredible startup stories I've ever seen. One of the most interesting things about proximity to Core Weave is simply having a pulse on which AI use cases are taking off at any given time. Back in 2019 the team told me this thing called AI Dungeon is hammering our servers. It was a text based fantasy adventure game built on GPT2. Of course I quickly jailbroke the game and was able to get it to return arbitrary queries rather than simply following the game's intended design. Even on a fairly primitive LLM, the experience of an interactive and sophisticated text model absolutely blew my socks off. At that moment I felt something had changed forever. Many people later had the same aha moment. When ChatGPT came out in 2020, I read Gweran's scaling hypothesis, one of the most prescient and important blog posts of all time, in which he pointed out that simply throwing more data and compute at these models can plausibly create AGI or something close to it. In 2022, stable diffusion came out and that blew my socks off again. I spent countless hours learning, prompting. I realized that AI was truly multimodal. The early image models weren't impressive by today's standards, but the direction of travel was obvious. Imagegen would be perfected and then text to video, which we are on the cusp of now. At that point I felt that ImageGen was too important to ignore and seriously explored the idea of incubating an imagegen startup at this point I had developed the conviction that I was dreadfully underexposed to AI. Even despite my core weave exposure. I was determined to Change this. In 2023 I was lucky enough to meet Vishal Mani and I wrote my largest ever LP check into Mythos Ventures, a VC firm focused on new applications unlocked by AI. I also dramatically shifted my angel activity towards AI and wrote my biggest ever angel check into Avi Schiffman's Friend.com, an AI wearable startup. The reason I leaned so heavily into AI was because of a few beliefs I had developed. First, AI would dramatically empower capital relative to labor. AI means that companies simply need fewer employees while maintaining the same level of productivity. I noticed this already in my own practice. I can now do programming or data science tasks myself. Whereas I might have previously needed a software engineer or a data scientist, I noticed that with certain data analysis tasks I was 100 to 1000x more efficient using AI tools. I noticed the same efficiency and cost savings gained with image generation. This is true in a variety of modalities. This is profoundly disruptive for society and massively accelerates an ongoing trend of automation and devaluing of human capital, particularly in professional services. The point is, I felt that the balance of power was shifting away from people selling their labor to companies and in favor of shareholders and firms. My action items? None. Because as a VC I am already on the capital side. To put it crassly. Next, investors overlooking AI would miss out on the biggest theme of the decade. The foundational models are not, in my opinion, the way to play this though. If you're an early stage investor you benefit significantly because AI drives down the number of staff required to run a startup. Solopreneurs are now a thing. A relatively smart individual with no programming experience can now build things on their own. My action items lean heavily into LPing into AI VC funds, doing AI angel deals out of my pa and looking at hybrid AI crypto deals at Castle Island Ventures. Next, AI will permanently put an end to the post truth era this is the subject for another post, but clearly our prior epistemic standards no longer apply. The cost of creating arbitrary image or video content is effectively zero, so unsigned content will no longer be considered reliable once people have learned to latently mistrust online content. To be considered reliable in the future, content will have to be signed, attested to and timestamped, likely on a blockchain. So our post truth era will end. Not because content is now no longer reliable by default, but because all content will be assumed fake unless attested to. My action items invest in startups like Tab Friend AI wearables that can create an attested digital alibi. Startups like Witness an on chain attestation tool Next the AI boom would rescue the US from its demographic malaise and its significant debt overhang. After World War II, the US was in a similar situation with regards to indebtedness, but we found our way out through a combination of high and variable inflation, a baby boom, and a productivity boom. I currently believe that in the US AI will add 2 to 4 points of GDP growth for a decade and help us grow out of our debt crisis we are facing. Even absent the favorable demographics. I believe the stronger growth in the US relative to the rest of the developed world is at least partially a function of the AI boom we're seeing. Luckily for the US and unluckily for the rest of the world, the epicenter of AI development is here. And that causes me to have a new level of optimism about US fiscal prospects that I simply didn't have before. I think that AI is at least as significant economically as the invention of nuclear power or the Internet, and probably more. However, it will have a profoundly disparate impact and the benefits will accrue to far fewer, which is part of my concern. My action items reduce my internal probability that the US faces a significant debt crisis, at least relative to the rest of the developed world. Retain the US as the nexus of my professional activities Today's episode is brought to you by Fractional Fractional AI is my Go to AI Dev shop. When we wanted to build an AI product feature for our company Superintelligent, we hired Fractional because they're some of the best and fastest AI engineers on the planet. The feature they built turned out great. It's already been released. I'm about to hire them for another project, so I highly recommend them to anyone looking to build AI product features and workflow automations. The Fractional team is a group of senior engineers in San Francisco working on some of the most exciting projects in applied AI. They work with everyone from startups all the way through the Fortune 500. To request a free consultation. Head to Fractional AI if you want help identifying and building AI projects for your business, then I highly recommend that you go check them out. Hit pause on the show, open a web browser, go to Fractional AI and get your free consultation. Today's episode is brought to you by Super Intelligent. Regular listeners know that super is our platform for helping people learn how to actually use AI tools. These are not long, laborious courses. These are fun, fast tutorials that get you actually using the world's most interesting and useful AI tools within minutes. If you want to build a web application with no code, we've got tutorials for that. If you want your presentations to look better than ever and take you less time than ever, we've got tutorials for that. If you want help brainstorming, writing social media copy, and just generally working smarter, faster and better, we've got tutorials for that. We've worked really hard to make it so that there is no better place on the Internet to learn how to actually put AI to work for you, and I'd love for you to check it out. Go to BeSuper AI and use code podcast for 50% off your first month. Once again, that's BeSuper AI. The scale of the AI boom is so significant today that it is running up against new bottlenecks. In 2021-2023, the constraint for AI was the availability of hardware, specifically a 1/ hundreds and then h 1/ hundreds. Today it's the availability of tier 4 data centers. AI data centers have meaningfully different infrastructure requirements from ordinary ones because they require more sophisticated networking, have higher power density, and need more cooling. These take a long time to build, and that's the bottleneck today. If you listen to Zuckerberg on the Dwarkesh podcast, he repeatedly says that the new constraint on AI compute growth is simply power. The level of investment the hyperscalers are talking about putting into AI compute will, in my opinion at least, rival investments in telecoms. Quick side note, even if the hyperscalers are overinvesting on AI clouds and data centers, this isn't wasteful in the same way that the railroad boom was, since AI clouds can incorporate different models based on whatever ends up being best, so duplication isn't a problem. If they overspend, that simply creates a consumer surplus whereby inference is cheaper than it otherwise would have been. Amazon, Alphabet, Meta and Microsoft announced that they will collectively spend $200 billion on AI infrastructure this year alone. AI growth is so aggressive that we are now running up against the literal availability of gigawatt scale power as the new constraint. Why is the Nvidia rally making me nervous at 2.8 trillion market cap and up 135% year to date. Which editor's note has gone up subsequent to that? Obviously we're over 3 trillion and above 150% year to date. Now Nvidia is posting growth numbers that are almost inconceivable for a company this large. The rally is so significant it appears to be sucking capital out of the rest of The S&P 500 and other big tech names. Partially, the rally is driven by investor desire to chase proven growth in a relatively weak economic environment, powered by the belief that Nvidia chips, software and networking are protected by a fundamental moat, which I generally agree with, and so they are the equivalent of a monopolist in a commodity that everyone needs to buy. But I'm also listening to what the market is telling me, which is that Nvidia is the most important company in the world today. The growth numbers Nvidia is posting, at least partially, seem to be justifying the rally. I think the market has realized that AI will be embedded into every application, AI wearables will be ubiquitous, and eventually we'll stop thinking of AI as a distinct category. The same way we don't think of Internet connected devices anymore because everything is networked. We don't have Internet investors, we just have investors. And every startup relies on the Internet. AI will simply be ubiquitous, and this means that the compute requirements per capita will increase by many orders of magnitude over the coming decade. Virtually everyone will use AI virtually all of the time because it will simply be incorporated into every application. As I said before, I think AI dramatically empowers capital relative to labor. This is why as a capital allocator, I significantly pivoted my focus to firms that would benefit from AI on a first and second order basis. But this has a very uneven impact on society today. In my view, human capital has already been devalued to around zero in fields like translation, transcription and summarization, Full self driving works today potentially obsoleting huge pools of labor like taxi drivers, rideshare and eventually trucking. In other fields like programming, web dev and graphic design, AI tools dramatically enhance human productivity and reduce the need for junior programmers doing relatively mundane tasks. In medicine, AI diagnostics are already superior, especially in imaging, although the highly regulated nature of medicine means these improvements will be resisted for some time. In white collar professions like law and accounting, AI will be able to replace a lot of the grunt work done by junior staff while AI delivered medical or legal advice seems primitive right now. These fields mostly boil down to ingesting patient data and creating recommendations, or querying large datasets of case law and giving advice. There's no reason AI can't reach parity with the state of the art here. Of course, these white collar professions rely on tastemakers at the Very top to interpret the data they are given, and that won't go away. But most of the process to get there can and will be automated. There seems to be no place to hide. Many draw analogies from the Industrial Revolution and point out that it didn't put people out of work, it just created new jobs. As civilization was able to harness energy more effectively, urbanize and specialize. But this isn't quite true. The Industrial Revolution did make huge labor pools irrelevant animals like horses that suddenly had no role in agriculture. The number of agricultural horses in Europe, for example, declined by about 90% in the 100 years following 1850. Today, taxi drivers, translators and so on are the horses. But you can't literally put these people out to pasture like the horses were. The social contract in developed countries stipulates that they'd be taken care of even if their human capital has been devalued. This, combined with a demographic transition and a shrinking prime workforce relative to total population, deeply concerns me. The other disanalogy with the Industrial Revolution is that in that case we harness new sources of energy to make humans more productive. In this case, we have created superhuman level intelligence, currently specialized in a few domains and within a few years general that far surpasses human capability. Of course, entrepreneurs and creatives will be able to harness these tools to make themselves orders of magnitude more productive. For this reason, I am positive on GDP growth in the startup sector as the number of employees needed to build a startup continues to decline. But it's undeniable that it simply makes a lot of human skills irrelevant. It's my current hypothesis that AI will continue to drive a worsening division between capital and labor to the benefit of capital. In the recent inflationary environment, as asset prices came down temporarily and hourly wages were revalued upwards, labor actually did well relative to capital. I think AI will reverse this short term trend as we see productivity grow and as senior programmers, consultants, lawyers, et cetera are able to use AI tools to do the job that would have normally required five to 10 analysts or more junior staff. In the medium term, entire professions which employ millions of people will simply cease to exist. Society can't just tolerate a massive furloughing of a huge percentage of its workforce. So I expect we will see reprisals against capital, which we are already seeing to a degree. These reprisals could take the following highly regulated AI in an effort to slow its disruptive growth. This currently is underway under the guise of AI safety, raising capital gains taxes and eliminating loopholes like qsbs and the carried interest tax loophole, increasing government spending on entitlements and direct transfers to this newly unemployable sector. Think the COVID era transfers made permanent. This is the side effect of increasing inflation which creates nominal equity gains which are then taxed, creating a wealth transfer from investors to the state directly involving the state in AI development via intrusive laws like California's SB 1047 which would effectively ban open source AI. If things play out the way I expect, then we may also see an empowering of socialist political movements in developed nations as new coalitions of AI affected individuals are formed. These may be more powerful than prior socialist movements as we will see white collar salaried workers included in the set of disenfranchised individuals. As an investor, the AI opportunity is obviously colossal and on a par with the invention of the Internet or railroads in terms of disruption and value creation, but I think it's likely to be too successful in terms of disrupting society. I believe that the effect of AI on the workforce will lead to an empowering of socialist anti capital dynamics in the West. So while the move is to allocate aggressively, you have to consider the reprisals to come so really interesting stuff here from Nick. Again, just published as a Twitter essay, something that I've spoken about quite a bit, not necessarily on this show, but on other people's podcasts, is what my base case for optimism is when it comes to AI. The short of it is that I think that the human capacity for consumption is basically unlimited and that in a world where it takes one tenth of the effort that it does now to create the same amount of stuff, be it the same amount of code, the same amount of content, the same amount of entertainment. I do not believe that means we'll have one tenth of the people creating code, or 1/10 of the people creating content. I think that means we'll have 10 times as much code. I think that means we'll have 10 times as much content. I think that we will simply make more more of everything. We will in fact have so much more that it will totally change what coding, what software, what entertainment, what content looks like because of the absolute glut of availability. Now there are going to be some weird dynamics within that, but net net. I do believe that over the long term there's just going to be a phenomenal amount more of economic activity. So that's my long term reason for optimism. However, I think the point that Nick is making that there are implications in the short term is absolutely true And I don't believe that there's anything mutually exclusive about, on the one hand, being net long optimistic and in the short term being really concerned about the types of disruptions that we're about to experience Now. I do think a lot of roles that we might imagine right now will be just totally gone, will evolve in ways that are much more nuanced than we're understanding. I think different collections of tasks will be obviated first. But I think that in the search to not have to unemploy people, there will be a race for many of those people who were primarily focused on those tasks to figure out new, higher productive uses of their time. Still, when push comes to shove, there will be some things that just go away. The question is how we deal with that. Nick is thinking about political implications if we've got this whole set of disenfranchised people. And I think it's good to spend time thinking about those things. Of course, from my standpoint with super intelligent you, you might be realizing it's not an accident that I'm thinking about how to educate and give people access to this new technology as soon as humanly possible and at a scale that involves everyone. I don't think it's a silver bullet, but I think giving as many people as humanly possible the tools to actually go out and try and understand how to use this new set of platforms is likely to have a significant impact on how many people get to enjoy the benefit of this transition anyways. All really interesting stuff, things that we will be talking about a lot more in the years to come. For now, one more big thank you to Nick for writing an excellent piece. And of course a big thank you to you for listening or watching wherever you are. Until next time, peace.