Lots of articles on AI this month. I held back on sending more. Enjoy!
(1) Capital, AGI, and Human Ambition
"Labour-replacing AI will shift the relative importance of human vs. non-human factors of production, which reduces incentives for society to care about humans while [entrenching] existing powers [...] the key economic effect of AI is that it makes capital [a] general substitute for labour. There is less need to pay humans for their time to perform work, because you can replace that with capital (e.g. datacenters running software replaces a human doing mental labour)." AI solves the two main constraints on capital substitution: implicit knowledge process automation and a scarcity of top talent. If all "intelligent" labor can be replaced by AI, then "the only [remaining] forms of human ambition [will be] playing local social and political games within your social network and class [...] AIs are better poets, artists, philosophers, everything; why would anyone care what some human does, unless that human is someone they personally know? [...] With AI, we could have the last and greatest opportunity for human ambition—followed shortly by its extinction for all of time. How can your reaction not be: "carpe diem"?"
(2) Training AI Models Might Not Need Enormous Datacenters
The Economist was ahead of the times with this article, which is especially relevant in the context of recent news on DeepSeek… Backpropagation is a method of training AI models that adjusts model weights based on the difference between predicted and actual text. Large datacenters are currently required for training frontier models because of "checkpointing" that synchronizes multiple training flows. But recent innovations in checkpointing, notably DiLoCo, are enabling training models in a distributed fashion by bundling the backpropagation steps in a way that reduces checkpointing frequency. The upshot is that, in the long run, these methods could reduce model training costs, enable training in a distributed fashion (iPhones could, in theory, be used as distributed clusters), and reduce current reliance on monolithic datacenters. I'm sure I butchered some of the nuances; this stuff is complicated. But if you're into AI, it’s worth reading the article. If you're not into AI, good luck.
This is a good (and technical) overview of why DeepSeek is in the news. They’ve made several innovations in AI development with cheaper training and inference costs (e.g., “mixture of experts” RAG to focus on subsets of the training data, inter-chip memory bandwidth optimization, DeepSeekMLA for more efficient memory usage within the context window, pure reinforcement learning with model training, etc.) that produced a frontier model at a fraction of the cost. The news is moving fast on everything in AI, but the more general implications are (1) AI capex might be going through the floor; (2) it’s even more unclear if the underlying models have competitive advantages; and (3) the cold war with China is heating up.
(4) Strategy Letter: Let Me Go Back!
A contrarian view on switching costs in the context of Excel beating Lotus 123 (from June 2000), which makes a useful distinction between moats around legacy and incremental users, and has a number of lessons for consumer and enterprise software companies… “The mature approach to strategy is not to try to force things on potential customers. If someone isn’t even your customer yet, trying to lock them in just isn’t a good idea. When you have 100% market share, then come talk to me about lock in. Until then, if you try to lock them in, it’s too early, and if any customer catches you in the act, you will just wind up locking them out. Nobody will want to switch to a product that is going to eliminate their freedom in the future.”
(5) Bill Gates: I Coded While I Hiked as a Teenager. Was I on the Spectrum? Probably.
Excerpt from the new Bill Gates biography on a hiking trip, writing BASIC formula evaluators, and positive reflection on the influence of his parents in his life: “If I were growing up today, I probably would be diagnosed on the autism spectrum. [My] parents had no guideposts or textbooks to help them grasp why their son became so obsessed with certain projects, missed social cues and could be rude [and not seem to] notice his effect on others. [They] gave me room to grow emotionally, and created opportunities for me to develop social skills. Instead of allowing me to turn inward, they pushed me out into the world—to the baseball team, the Cub Scouts and other families’ dinner tables. They gave me constant exposure to adults, immersing me in the language and ideas of their friends and colleagues, which fed my curiosity about the world beyond school. Even with their influence, my social side was slow to develop, as was my awareness of the impact I can have on other people. But that has come with age, with experience, with children, and I’m better for it.”
(6) Treading Alert: Greenland LLP
Trump loves doing deals… “We revisit the investment case for Greenland [...] the territory is in long-term decline. It has shrunk for 28 consecutive years, decreasing in size over the period by 11,000 square miles. Earnings growth since 2015 has been reliant on fishing [...while] Greenland has significant LBO potential, carrying virtually no gross debt, elevated leverage at subsidiaries [and] labor shortages [have caused its] maintenance expenditure [to trend] below replacement rates. Given the scale of these challenges and the illiquidity of its citizenship, M&A offers value realization for Greenland’s 56,000 independent stakeholders, who can effect a change of control via a simple majority vote. [...] Greenland is uniquely positioned [to] benefit from acceleration of cyclical climate change [and] near-term deregulation of fossil fuel extraction presents a virtuous circle: put simply, more drilling means more to drill."
(7) Lorne Michaels Is the Real Star of “Saturday Night Live”
SNL is celebrating its 50th birthday, and The New Yorker has a profile on the man behind it all: Lorne Michaels. SNL is the single most influential force in comedy—it established the careers of Chevy Chase, Bill Murray, Al Franken, Adam Sandler, Chris Rock, and basically every famous comedian of the last half-century. Lorne is the man behind the scenes pulling the strings, and while it sounds like working for him is traumatic, his methods got results: “[Lorne has] imbued “S.N.L.” with a clockwork stability. Every week, the sketches are written, the index cards shuffled, the vases of flowers replenished. New employees are still routinely terrified of him; when the office feels too friendly, whispers circulate that Michaels doesn’t like “the tree-house vibe.” Will Ferrell thinks that Michaels’s emotional withholding is part of a baseball-derived management style. “Baseball players keep the highs not too high and the lows not too low,” he said.”
(8) What “The Power Broker” Gets Wrong About Robert Moses and Ambition
Does Robert Moses embody the spirit of “building” things that is resurgent in the US? Maybe! “Others have over-learned the lessons of the “The Power Broker” and [over-simplified the story of] a deeply complex, unsettling and extraordinarily accomplished historical figure. […We] have lost faith in government and failed to comprehend that some of the Robert Moses spirit must be recaptured today if the United States is going to be a great builder again. As megalomaniacal as Moses could be, he showed that the public sector could be a force for civic improvement in America, delivering monumental public works that could stand for a century and beyond.”
“In the 11th century, the King of England, William the Conqueror, [commissioned a survey] to measure the capacity of his kingdom to pay taxes. The product was the Domesday Book. […] Specifically, the Domesday Book was used to collect a tax called the geld, levied annually on landholders. The tax [originated] in the Danegeld, a tax used to pay tribute to the Vikings in exchange for pausing their pillaging. This tax was levied on all subjects, the first direct tax of this breadth in Europe. The new geld was based on a landholder’s ability to pay, [which was] based on the value of a landholder’s estates, [which was imputed from] their annual income.”
(10) Sam Altman on ChatGPT’s First Two Years, Elon Musk, and AI Under Trump
An in-depth interview with the founder of OpenAI, Sam Altman. It’s surprisingly candid, and some details stand out: (1) it has clearly been stressful and they're flying by the seat of their pants; (2) they’re clearly focused on being research-first (because winning seems to require attracting the best talent in AI research), (3) the consumer-oriented aspects of the business, notably with ChatGPT, came about mostly as a surprising accidental success, (4) it’s crazy to think about how much has happened over the past two years, and (5) they should have used Superwall to optimize their pricing: “We launched ChatGPT for free, and then people started using it a lot, and we had to have some way to pay for it. [We] tested two prices, $20 and $42. People thought $42 was a little too much. They were happy to pay $20. We picked $20. It was late December of 2022. [This] was not a rigorous “hire someone and do a pricing study” thing.”
(0) Miscellaneous
Social media and declining birthrates. Estimating the valuation of Greenland. The rise and fall of corporate bonds. Melatonin reduces jet lag. The origins of non-alcoholic beer brand Athletic Brewing. The growth of the UFC. VCs love accounting rollups. The early history of financial markets data. Detailed guide to the QSBS tax break. The problems with being data-driven. Confusing opacity of modern gender norms. Buying discounted assets in Russia. It’s easier to imagine the end of the world than the end of capitalism. Getting tricked by an AI deepfake of Brad Pitt. Personal library design inspiration. Patrick Mahomes “has turned the Super Bowl into his personal man cave.” Trump’s immigration fight with Colombia. Trump’s immigration team is pulling every lever in the administrative state.