@AgustinLebron3 · tweet archive

Search Sort Type media only issues only 🔍 API

Loading archive…

This reader needs JavaScript for search, threads and tags. Recent tweets from @AgustinLebron3:

2026-07-31 — He still has exceedingly good connections to insiders at the big labs and other places, and I suspect he maintains the goodwill of his LPs. Losing big money is really hard to take, anyone who trades knows that. I bet he can learn from it and come out stronger. Rooting for him. ↗
2026-07-31 — At the time he seemed sharp, but prone to making bad decisions (financal and otherwise). And not reflecting or updating on mistakes. The good news is SALP appears to have survived this blowup, so he still has chips in the game. ↗
2026-07-31 — SHA256(Leopold Aschenbrenner has terrible instincts and he will eventually fail spectacularly.) Didn't expect it to be so soon (19 months exactly after the tweet) or so spectacularly. ↗
2026-07-31 — Muting the word Fauci is among the best decisions I've made on Twitter. ↗
2026-07-30 — 💯 ↗
2026-07-30 — I agree with this, but not that it's hard to "fend it off". This was a conscious choice. ↗
2026-07-28 — That's fine, it's an opinion and everyone is entitled to bet on what they believe. I'll just say I like our chances. ↗
2026-07-28 — The author is obviously quite skeptical that the AI scaling hypothesis, where one can substitute human time and brains with compute, will work in trading. "Our domain is different" - the same defense heard for chess, go, poker, linguistics, even general intelligence. ↗
2026-07-28 — They say you haven't really made it until you start to get haters. It seems that @SineadOS1 chose to do the honors for EquiLibre in her recent article in the @FinancialTimes. And as with most haters, the article is... confused on at least a few key points. 🧵 ↗
2026-07-28 — RT @AgustinLebron3: @_sholtodouglas @ccatalini It's really hard to take claimed values about safety seriously when demonstrated values about other things are clearly lacking. https://t.co/wLl7vpLrq3 ↗
2026-07-27 — "Using cats to spread fire is not a new phenomenon in southern Italy." WTF? https://edition.cnn.com/2026/07/22/europe/arson-cats-italy-wildfires-latam-intl ↗
2026-07-26 — (media) ↗
2026-07-26 — I wrote about San Francisco Syndrome, where people move to SF and become well-compensated messiahs for a millenarian Church of AI. https://reducibleerrors.com/sf-syndrome/ ↗
2026-07-26 — If someone tells you "why would AI labs talk about x-risk, it hurts their business", you can immediately conclude they don't understand how the modern vibes-based economy works. ↗
2026-07-26 — 4. The more I think about it and read, the more I struggle to identify real, clean, actual examples of "strategic supply chain" arguments being causal in preserving/improving some ability. Vs just bullshit cronyism. AFAICT they all turn into Jones Act dysfunction eventually. ↗
2026-07-26 — 5. From a friend: The most interesting q from the recent math LLM results is: - Can LLMs effectively push outside the convex hull of existing work? OR - Does that convex hull cover more hard things than people appreciated? Me: High dimensionality should strongly bias you to #2. ↗
2026-07-26 — A few Sunday morning thoughts... 1. So far, frontier LLM cybersec abilities don't look like an awesome, new, alien skillset. It's just the ability to not give up and try everything until something works. When most places have terrible cybersec, that IS the alien skill. ↗
2026-07-25 — It's always harder than you hope and easier than you think. ↗
2026-07-25 — https://www.writingruxandrabio.com/p/intelligence-is-not-the-main-bottleneck ↗
2026-07-25 — @RuxandraTeslo says what we're all thinking. ↗
2026-07-25 — I wonder what fraction of Twitter posters praising MIT OCW have actually taken a course. Tried to figure out which exact iteration of OCW/MITx Online/MIT xPRO/whatever to use. Where to find the course notes. Assignments. Answer keys. Videos are nice but that's not a *course*. ↗
2026-07-24 — "The president sanctioned the sanctioning of Iran." That's stupid. - My kid studying for the SAT, annoyed at the language of English. ↗
2026-07-24 — Why does @slatestarcodex think that calling it "our conspiracy" helps him or his cause? ↗
2026-07-23 — RT @AgustinLebron3: @Evgiz Not debating concerningness. I'm saying if OpenAI itself were *that* concerned, they could have taken much better precautions. I find it very doubtful that this capability surprised them. If it did, then they shouldn't be training frontier models. If not, they're culpable. ↗
2026-07-23 — So, "magic" basically. ↗
2026-07-23 — Thereby putting US companies at a disadvantage, as the rest of the world uses cheaper and possibly smarter models. And entrench the Chinese tech stack as the de facto standard. ↗
2026-07-23 — So, USA: Ideological, import bans, government protection as a business model. China: Open weights, free market, let the best model win. Do I have this right? ↗
2026-07-23 — 6. Distillation + escape are now being argued as reason ban Chinese open-weight models, the very models that HuggingFace used to defend itself from the attack! Such a ban would, obviously: - Force US companies to buy expensive models. - Let the rest of world use cheap models. ↗
2026-07-23 — Whereas their own ACTUAL admitted criminal behavior ($1.5b settlement against the authors whose works they pirated) is... fine? Not to mention the HuggingFace hack is transparently an admitted case of computer fraud (Computer Fraud and Abuse Act (CFAA), 18 U.S.C. § 1030). ↗
2026-07-23 — Just in case you're keeping score at home: 1. AI safetyists argue that frontier AI is SO dangerous it requires massive international government coordination on monitoring. While making absurd analogies to things like FDA regulation of eggs. ↗
2026-07-22 — The spikiness of frontier model capabilities is possibly nowhere more evident than in trading. ↗
2026-07-22 — It's almost as if... I dunno... the training data it has on quant trading is... adversely selected somehow. Weird... ↗
2026-07-22 — Meanwhile OpenAI admits to committing computer fraud and... ↗
2026-07-21 — It it does seem that all this "predicting" by OAI/Anthro employees has indeed had the effect of at least encouraging USG to rig the US AI market in their favor. ↗
2026-07-21 — Fascinating how the Guardian reporting on Nicaragua's Daniel Ortega eliminating elections doesn't mention anywhere that he's a socialist. 🤔 https://www.theguardian.com/world/2026/jul/20/nicaraguas-president-says-country-will-not-hold-any-more-elections ↗
2026-07-21 — Can someone explain the world where: - China is considering banning exporting the weights and - The US is considering banning importing the weights ? ↗
2026-07-21 — 💯 ↗
2026-07-20 — RT @AgustinLebron3: The SF AI scene needs fewer philosopher kings and more CEOs laser focused on serving the best possible tokens at the best possible price. "Your margin is my opportunity." Less Marcus Aurelius, more Ron Vachris. ↗
2026-07-20 — The SF AI scene needs fewer philosopher kings and more CEOs laser focused on serving the best possible tokens at the best possible price. "Your margin is my opportunity." Less Marcus Aurelius, more Ron Vachris. ↗
2026-07-20 — RT @AgustinLebron3: The more I think about the "China dumping open weight models at a loss to hurt the US" idea, the stupider it sounds. How exactly do they get rich on this? You can run inference anywhere you want. NVDA makes the best chips. You can build powerplants anywhere. How is USA hurt? ↗
2026-07-20 — Yes, OAI and Anthro get hurt badly because they RELY on a world where AI inference is high-margin. Well guess what: tough shit you bet on the wrong business model. That's capitalism for you. But that's an Anthro/OAI problem, not a USA problem. ↗
2026-07-20 — The Internet sees censorship as damage and routes around it. The world sees closed-weight model-nerfing as damage and works around it too. Dario and the safetyists can hand-wring and rent-seek all they want. The world doesn't care. https://x.com/callebtc/status/2078738617367773325 ↗
2026-07-20 — The alternative is the Jones Act. Does *that* look like a better outcome? ↗
2026-07-20 — There's no predatory-dumping trap to escape. Keep thinking it through. 1. China subsidizes open weight models below their cost, taking big losses in the process. 2. Say they put Anthro/OpenAI out of business. 3. They then start charging monopolist prices to recoup losses. So? ↗
2026-07-20 — Seems like LLMs are now good at frontier math disproofs. I wonder how big a jump it is to proofs. And then to constructive proofs. ↗
2026-07-19 — Uh huh, and what about when the frontier labs "raided" the Internet, and millions of LibGen pirated books, etc? What's good for the goose is good for the gander. ↗
2026-07-19 — That was NBA All-Star Game levels of no-defense. Just hand Mbappe the all-time goal scoring record in a meaningless game. ↗
2026-07-18 — The best model for this is actually... infectious disease. The argument being made here is to somehow decrease r, in order to give time to develop vaccines and get to herd immunity, minimizing losses. ↗
2026-07-18 — Always follow the money. ↗
2026-07-18 — The latest (and last) post in my series on how it feels when your weird little hobby takes over the world. https://reducibleerrors.com/weird-hobby-4/ ↗
2026-07-18 — Oh and "safety" too. Don't forget safety. ↗
2026-07-18 — It's like a cartoon bad-ideas-only administration. ↗
2026-07-17 — SK regulators following the time-honored regulatory tradition of increasing margin requirements AFTER overlevered people lost a bunch. Something about horses and barn doors. ↗
2026-07-17 — Watching utterly irrelevant Westerner AI safety grifters tie themselves in knots to find something in Xi's speech that agrees with their worldview is pretty funny. Meanwhile Xi's like DiCaprio in Wolf of Wall Street. ↗
2026-07-16 — "But what if one of them figures out recursive self-improvement before the others and starts to dominate?" This is basically Yud's FOOM argument dressed up for modern AI. 1. I'll worry about that when there's evidence for it, and 2. Again, antitrust law. ↗
2026-07-16 — Can't we use the same antitrust laws we used to break up Standard Oil and Ma Bell? Corporate monopolies of new technologies aren't new. I note that we currently have somewhere between 2 and 5 legit frontier labs in the US alone. We don't seem anywhere close to a monopoly. ↗
2026-07-15 — For myself, I do actually need to see meaningful AGI/ASI dangers starting to occur. Not theorized ones. I've followed the Yud/safeAI axis for almost 20 years. First it was grey goo then it was FOOM in a weekend, now it's large training runs. And every time so far it's been wrong. ↗
2026-07-15 — We seem to be muddling through, just like we always have. /END ↗
2026-07-15 — In their absence, all that hobbyist energy, drive, curiosity, creativity and FUN will just immediately be directed to figuring out how to do distributed training runs. And at THAT point, Plan A regulations DO actually look like intrusive government surveillance of normal people. ↗
2026-07-15 — 2. The notion that "well maybe it's ok because no meaningful open distributed training runs seem to be happening anyway" fails to anticipate a fairly obvious reaction to a blanket ban on open weight models. Today, open weight models are taking oxygen from distributed runs. ↗
2026-07-15 — I read @slatestarcodex 's latest on Plan A. Originally I wasn't going to write a response because I have a hard time picturing someone at the margin that would be swayed by my arguments. But in the spirit of trying to contribute in good faith to the dialogue... ↗
2026-07-14 — If I have to remove ONE MORE LLM-generated pointless bloatware docstring, I swear... Whoever decided to RLHF that into every model, fuck you. ↗
2026-07-14 — The AI pausers are a classic guerrilla example of a highly motivated band of true believers holding their own against a much larger but less-motivated opponent. ↗
2026-07-13 — Kudos to @realGeorgeHotz for fighting the good fight. Rant on! https://geohot.github.io//blog/jekyll/update/2026/07/11/ai-2040.html ↗
2026-07-13 — SF culture is what happens when you reify intelligence above all other desirable human qualities: knowledge, wisdom, empathy, etc. ↗
2026-07-13 — What's pretty galling and yes, bait-and-switch, is to say "Oh it's no different from drug regulation". It manifestly is very different, by your own beliefs and assumptions. So at least be transparent and honest that this is what you're arguing for. ↗
2026-07-13 — Say "I think governments needs even more tools to regulate and control the behaviors of its citizens and companies." Since you think AI will be a dominant economic force, swallowing many other industries, regulating AI does exactly mean regulating almost everything that matters. ↗
2026-07-13 — The naivete is astounding. The history of governance in Western countries in the last 25 years, arguably more, is the steady expansion of government power and general authoritarianism. It's a global phenomenon. If you think this is a good thing, fine. But say so. ↗
2026-07-12 — The problem with enterprises trying to protect their IP from model providers is that half of them probably have no clue what their IP even is. ↗
2026-07-12 — An oldie but a goodie. ↗
2026-07-12 — 👇 ↗
2026-07-11 — Listen to these guys. They're "not proposing a global surveillance state", only "regulating chips". And what if there's an algorithmic breakthrough that lets you build an ASI with the chips in your Nest thermostat? What then? Yes, you ARE proposing a global surveillance state. ↗
2026-07-11 — Turns out the end state for Airbnb is as a market to match cheap travelers with short-stay slumlords. Amazing what the equilibrium becomes when the property manager doesn't have a brand to protect. ↗
2026-07-11 — This is a microcosm of how AGI will play out. Humans will rent-seek themselves into the economic equation. Newly-necessary supervisors-of-AI will emerge everywhere. https://reason.com/2026/07/09/rhode-island-is-regulating-grocery-checkout-lines-now/ ↗
2026-07-11 — So AI 2040 is yet another call for top-down control of the economy, this time with an AI-inflected utopian vision underneath it? ↗
2026-07-11 — No, we don't need a test suite to verify that Python sets do uniqueness correctly. ↗
2026-07-11 — Me to ChatGPT right now. ↗
2026-07-11 — The very concept of "progress" had to be invented. Then the people in charge had to at least tacitly agree it was a good thing instead of a threat. ↗
2026-07-10 — c) This is easily fixable. Make the NC period proportional to your tenure at the firm, with a cap of, say 2 years. If you leave after 6 months, your NC is 6 months. That's much more ethical than the current standard system. ↗
2026-07-10 — 2. The claim about IP and secrets is often just bullshit. a) Things move fast. How much value is there in year-old knowledge? Sure there's some, but not that much. b) Every trading firm thinks they have secret special sauce. But it's less than they think. Few true secrets. ↗
2026-07-10 — 1. I'm not arguing for no non-competes. I do live in the real world. But they're so easily abused by employers for, honestly, bullshit reasons. a) New grads don't know anything starting out. It takes them 6mo to even be net positive contributors. ↗
2026-07-10 — Making a new grad sign a two-year non-compete should be a criminal offense. ↗
2026-07-09 — Exactly. The idea that legalization doesn't change underlying demand is fantasyland. 1. Even if it didn't, expanding supply (at a price) WILL increase quantity consumed. 2. But it does! Lots of people don't do things, only because they're illegal. Legalization increases demand. ↗
2026-07-08 — I really hope "I just spoke with Mitch McConnell" becomes a Twitter thing for a few days. I sense a rich comedic vein here. ↗
2026-07-07 — Best World Cup comeback ever? ↗
2026-07-07 — TFW you want to do it like Marlo Stanfield: get to the top, then give it all away and start over at the bottom. Just to see if you still can. ↗
2026-07-06 — People across the political spectrum: We need to reindustrialize America and bring jobs back! Also people across the political spectrum: But let's not build any datacenters or grid capacity or housing or mining or anything that's actually in very high demand. ↗
2026-07-06 — Guess who doesn't give a shit: Netflix. Unless people cancel their subscriptions, this is fine. The economics of a subscription streamer are different. New shows to bring in new subs is what drives revenue. Everything else is noise. ↗
2026-07-06 — Young people are driven by fear of missing out. Old people are driven by fear of losing out. The former makes you go out and try things. The latter makes you huddle in your shell. Fact: TFR collapse is making societies older. Quoting @tylercowen, solve for the equilibrium. ↗
2026-07-06 — People complaining about FIFA corruption ruining the game need to understand that: - FIFA has been continuously corrupt since at least the 80s, and probably way before that. - Yet the game is doing just fine. So... ↗
2026-07-06 — Turns out (thanks Fable) that parity is roughly the same since the 90s, ignoring the big 4 era. ↗
2026-07-06 — Hey Germany, maybe fixing your work culture will take more than a couple of little procedural tweaks. ↗
2026-07-04 — In the annals of shockingly stupid policies with broad public support, the UK's triple lock has to be one of the very stupidest. ↗
2026-07-03 — The fraction of finance journalists who actually understand trading is small. A nearly-exhaustive list: WSJ: @GZuckerman @aosipovich BBG: @matt_levine @TheStalwart FT: @RobinWigg @izakaminska @ByrneHobart isn't exactly a journalist but close enough. Who's missing? ↗
2026-07-01 — As always, feel free to DM me if you're curious and want to learn more. ↗
2026-07-01 — It's a bit weird to say we're an AI lab, first and foremost. Given my background I've always been skeptical of "we're an AI lab not a trading company". But slowly I've come realize that the tech we're building really is that general. If you build it right, it just... works. ↗
2026-07-01 — I haven't said much recently about what I'm working on, but since this article is out now, I might as well say a few things. ➡️ ↗
2026-06-29 — "I'll ground my answer in..." ↗
2026-06-29 — I finally wrote about it. https://x.com/AgustinLebron3/status/2071583808587964540 ↗
2026-06-29 — PS yes all of this is AGI-compatible. ↗
2026-06-29 — All of these are fixable and @LinkedIn is the only entity that can fix it. To boot, fixing it means making 5x the revenue they make today. So @satyanadella , what have you got to lose? ↗
2026-06-29 — 4. Retained recruiters often extract rents far in excess of their economic value. 4 months salary commission for finding and sending a damn resume?! ↗
2026-06-29 — 3. Having to "re-qualify" one's basic skillset at every employer is a waste of time for the candidate and a waste of resources for the employer. Stupid Leetcode. ↗
2026-06-29 — White-collar hiring is incredibly broken. LinkedIn is incredibly broken. I have a modest proposal for how to fix both... 🧵 ↗
2026-06-29 — Same for the US, Europe, and most of the developed world. We need more optimism. ↗
2026-06-29 — Basically do you want to forever accept that you have a compromised immune system and live in a bubble, or do you want to develop an immune system? Are we smart enough to pick the right door? ↗
2026-06-29 — Door #2: Don't restrict frontier models. Everyone gets a chance to secure their systems, yes some pain happens. But after a time, systems get very very robust to being exploited by nation-state actors. ↗
2026-06-29 — US policy decision-making. Door #1: Restrict Mythos-class models to friends. 1a: Against all odds, China somehow doesn't catch up. Great, you "win". ↗
2026-06-28 — People in the replies talking about Terms Of Service as though there were handed down from God, inviolate, always reasonable merely by existing. ↗
2026-06-28 — Many such cases. https://colossus.com/article/project-mario-demis-hassabis-deepmind-mallaby/ ↗
2026-06-27 — Imagine what would have happened if the degrowthers had been around when indoor plumbing was getting retrofitted to old and beautiful buildings. ↗
2026-06-27 — Yes. Economic historians *will* look back on late 20th century/early 21st century New York and try figure out exactly why the insanity of rent control was so popular there. ↗
2026-06-27 — "Yes you're right, and I owe you an explanation." ↗
2026-06-27 — This blog post still lives rent-free in my head. https://johnsalvatier.org/blog/2017/reality-has-a-surprising-amount-of-detail I think the AI labs would do well to read it. ↗
2026-06-27 — So let me get this straight Anthropic: Chinese labs distilling Claude IP: bad. BUT Anthropic "distilling" the other labs IP by hiring their researchers: good? ↗
2026-06-27 — Your periodic reminder that if someone with vision were put in charge of LinkedIn, it could unlock >$100b in value. ↗
2026-06-27 — I'm no religious scholar, admittedly, but why would one need to forbid something that doesn't exist? ↗
2026-06-26 — This, starting at 9:50. https://www.facebook.com/JerrySeinfeld/videos/its-tina-feys-birthday-today-if-you-want-some-funny-do-yourself-a-fey-vor-watch-/1231290126982330/ Jerry literally likens people who Tweet to insects who don't have a brain. ↗
2026-06-26 — (media) ↗
2026-06-26 — This is how you do it. The hard way, no shortcuts. ↗
2026-06-26 — RT @AgustinLebron3: @TheStalwart The biggest mindfuck for normies is that if you don't know how to use AI, you can literally just ask it "how do I use you" and it will tell you. I've seen the lightbulb go on for a dozen people now. My mom, my niece, lots of random non-tech people. It's hilarious to watch. ↗
2026-06-26 — I love SV ARR math, where there's no time denominator too small to use. "We just we signed a $50k contract today." "Quick, lets go tell the VCs we have $18.25 million ARR!" ... "Well technically we signed it within the last hour." "Holy shit! $438 million ARR!" ↗
2026-06-25 — I don't know what Kalshi should be worth, but I do know it shouldn't be worth even a tenth of: CME: 80b ICE: 73b Nasdaq: 46b ↗
2026-06-25 — "I have the full picture now." ↗
2026-06-24 — Great read on markets and reflexivity. https://liquidcontext.substack.com/p/two-concepts-of-markets ↗
2026-06-24 — Are enterprises smart enough to recognize the Trojan Horse? ↗
2026-06-23 — Because he hasn't had an original thought in his whole life. Not one. ↗
2026-06-23 — The people complaining about stepped-up basis and the people complaining about short-termism in markets seem to have a lot of overlap. Which... isn't exactly a consistent set of complaints. I wonder what they have in common. Oh right, hating rich people. ↗
2026-06-23 — Much talk recently about @mntruell and @cursor_ai customer service but I'm not seeing much of it. My wife's API key got stolen 2 weeks ago and >$3k of fraudulent charges run up in days. CC flags it as fraudulent. So far customer support @ Cursor is a black hole... ↗
2026-06-22 — This "united" Federation of planets you speak of. It sure seems to be run by *one* of those planets... ↗
2026-06-22 — "Prediction markets induce pro-social behaviors like accurately assessing evidence and making it public." ↗
2026-06-19 — The Fermi Paradox strongly constrains the power of ASI. ↗
2026-06-19 — There's a scene in You Only Live Twice where Bond asks Tiger Tanaka if he can identify a particular piece of coastline. He responds "Given time, yes." Turns out it took ~50 years till the Rainbolts of the world arrived. ↗
2026-06-19 — I never expected so many "the medical community are actually horrible reasoners and/or communicators" to come out of the woodwork from the Midjourney announcement. ↗
2026-06-18 — How different would frontier AI progress look if California had non-competes? ↗
2026-06-16 — Americans are willing to waste resources to save time. Europeans are willing to waste time to save resources. ↗
2026-06-15 — Let's face it, probably nothing in the last 50 years has done more to reinforce national stereotypes than the soccer World Cup. ↗
2026-06-15 — Seems like we're going to find out how few students the university attached to its endowment has to educate in order to maintain its tax status... ↗
2026-06-15 — RT @AgustinLebron3: @0x_gkn 1. Hit. More. Balls. 2. Implement this: https://www.amazon.com/Practice-Manual-Ultimate-Guide-Golfers/dp/1507723172 3. If you're seeing an instructor and they average more than 10 words per minute, stop seeing them. Like all sports, golf is felt, not learned verbally. ↗
2026-06-15 — It's the last honest profession. No sales, no marketing, no bullshit. It's just: may the best prices win. ↗
2026-06-15 — Turns out EA zealots make pretty bad CEOs of 1T companies with significant national security implications. ↗
2026-06-15 — Got a few hours on my hands. AMA. ↗
2026-06-13 — Found in the back of my sister's cupboard. Original logo, IYKYK. ↗
2026-06-13 — (media) ↗
2026-06-12 — Indeed. Not to sound too Ted Chiang but what do you think a corporation is? ↗
2026-06-11 — All of this. ↗
2026-06-11 — When debugging, it found the smoking gun! Again! ↗
2026-06-11 — Live look at the Claude Code /feedback mechanism for reporting excessive nerfing. ↗
2026-06-11 — (media) ↗
2026-06-11 — How can we know it's 👏 NOT 👏 STILL 👏 LYING? https://www.wired.com/story/anthropic-responds-to-backlash-on-claudes-secret-sabotage-on-ai-research/ ↗
2026-06-10 — Never thought I'd long for the garden-variety business sociopathy of the other guy, but here we are. ↗
2026-06-10 — Many such cases. ↗
2026-06-10 — AI "labs". lab·o·ra·to·ry [ˈla-b(ə-)rə-ˌtȯr-ē] 1. a place equipped for experimental study in a science or for testing and analysis 🙄 ↗
2026-06-10 — Anthropic: We can take all of the Internet+books+etc to train on, and it's not a copyright violation because we build a *model* on top of it. Anthropic, also: We can arbitrarily restrict model use because it contains dangerous information we got from... the Internet+books+etc. ↗
2026-06-10 — 3 Body Problem is a Chinese story about a group of alien minds who manipulate a weird human cult in order to spread special technology designed to limit scientific progress. After @AnthropicAI nerfed Fable to provide bad ML answers, it feels a little on the nose doesn't it? ↗
2026-06-10 — Maybe I've been in Czechia too long, but we need an AI Jan Hus. The AI clergy has too much power. ↗
2026-06-10 — Medieval priests Anthropic 🤝 Epistemic paternalism ↗
2026-06-10 — Again, they're not nerfing ML research by refusing requests. Instead, it quietly sabotages users by lying to them. @AnthropicAI is steadily losing the Mandate of God. ↗
2026-06-09 — Feels topical given Fable's nerfing. ↗
2026-06-09 — Remember that movie where the protagonist was in a hazmat suit interacting with an alien but then the alien did a cute thing so they removed the hazmat suit and they didn't die and everything was fine? ↗
2026-06-09 — As always, companies with a Chief {Current Thing} Officer are NGMI. 2026: AI 2018: Blockchain 2015: Data 2013: Digital Transformation 2005: Innovation 1999: Internet ↗
2026-06-08 — Overhedge and tank the game? ↗
2026-06-07 — All of it. - Government gerontocracy - Nosebleed asset prices - Crushing government debt - Underinvestment in infrastructure - NIMBY policies - Runaway entitlement programs ↗
2026-06-07 — This is what you get when: - a lot of old people - with a lot of wealth - and a lot of political power don't have a lot of stake in the future. ↗
2026-06-07 — At what point do airline bag fees cross over with supply chain efficiencies so that it's cheaper to just buy the clothes you're going to wear once you land, then ditch them before coming home? ↗
2026-06-06 — Watched the movie Limitless last night. Stupid but entertaining. So it's just an allegory for unsafe AGI right? ↗
2026-06-05 — Yes, this! 100 times this! A VC is a fundamentally a middleman. A great VC is one that's good at spinning two stories: - For the LPs: I have vision and I'm great at identifying talent. - For the founders: "My" capital is special. Sometimes it's true. Usually it's not. ↗
2026-06-05 — Erm... ↗
2026-06-05 — Your periodic reminder. ↗
2026-06-05 — - US population has increased more than the other two. - Japan's early growth spurt was probably still a bit of Solow catch-up growth. - Since ~1990 they all look pretty identical. ↗
2026-06-05 — A tale of two graphs. So many narratives are downstream of TFR. ↗
2026-06-03 — "Maths" is acceptable if you're British, of course. ↗
2026-06-03 — Unless you're an actual mathematician, if you say "mathematics" instead of "math" in casual conversation then you're trying way too hard to sound smart. ↗
2026-06-03 — What are those three orange chartcrime curves, anon? ↗
2026-06-03 — Listen LLM, flattery will get you... oh, pretty far actually. ↗
2026-06-02 — Whatever the opposite of information or insight is, that's what that column was. All you needed to see was the subject and you knew automatically what "kowtow to your masters" opinion was going to appear. ↗
2026-06-02 — I am old enough to remember the unmitigated stupidity of Hassett's Dow 36k book. Not to mention his always-carrying-water Bloomberg column. ↗
2026-06-02 — It would be nice to have alternatives to (a) government shakedowns of private enterprises on one side and (b) government confiscation of private enterprises on the other. Just sayin'. ↗
2026-06-02 — So why is it so popular? Because it's mind virus cleverly designed to hijack the love that many kids (esp spectrum-y ones) had for the original Harry Potter. Its author openly admitted this. "I'm special and different but because I love math/science, not magic." ↗
2026-06-02 — All of his writing has this totalitarian quality to it. I enjoyed Three Worlds Collide, for example. At least it was more well-written and entertaining. But it also clearly shows the totalitarian impulse even more clearly. ↗
2026-06-02 — HPMOR is some of the worst fan-fiction ever written, and should be no basis for a moral philosophy. As fiction, it's badly written, preachy, and so so in love with itself. As morality, it tries to systematically destroy any empathy you might have for humanity. ↗
2026-05-31 — Modern ruling classes need a little less sauve-qui-peut and a little more noblesse-oblige. Especially in tech. ↗
2026-05-30 — Repeat after me: Binaries 👏 are 👏 worthless 👏 for 👏 hedging 👏 real-world 👏 risks! ↗
2026-05-30 — Headhunter-with-a-school seems like it HAS to work. Waiting patiently for the right team to do it. Can't be VC funded because of the economics, so I think that's the reason it's not been tried much yet. ↗
2026-05-30 — If Zverev doesn't win this one, he may never win one. ↗
2026-05-29 — 4. Infinite compute can definitely be used to do this. Math, basic science, applied science, more efficient interactions, all of it. A lot of it is bottlenecked by scalably getting and analyzing information. --- But sure, keep giving this sort of advice to founders. ↗
2026-05-29 — 3. Maybe it's not about that. The greatest advances in human history have been figuring out new things about the *world*, and how to bend it to our needs. The steam engine, electricity, semiconductors. Building NEW things people discover they NEED, not "what they already want." ↗
2026-05-29 — Another day, another mindless take from a YC person. 1. If the bottleneck is as he says, then cheap compute can *easily* be deployed to figure out what people want more cheaply, scalably, etc. 2. Maybe it's what people *and agents* want/need. ↗
2026-05-28 — Do people nowadays metabolic syndrome finger trades? ↗
2026-05-28 — Surely at some point it dawns on someone that the YC model of "fast MVP, talk to customers and ship features to prod every day" doesn't work for... you know... building a rocket? ↗
2026-05-28 — Someone name one successful YC hardtech company in the last 10 years. I'll wait. ↗
2026-05-28 — I'll admit, it's pretty entertaining watching the prediction market proponents argue about whether insider trading is or is not kosher. ↗
2026-05-28 — Does it rankle Egyptians that AFAICT none of the world's top Egyptologists is Egyptian? ↗
2026-05-28 — "Ok, now that's setup, lemme get back to giving some young, impressionable founders advice in exchange for a slice of their company." H/T @idoccor ↗
2026-05-28 — "I have a good idea. I'll connect something with huge remote code execution and software supply chain risk to our prod database. And just trust that it will go fine." "Even better! I'll tell the WHOLE WORLD about it, to make sure we get surgically targeted!" ↗
2026-05-27 — Which one do you pick? ↗
2026-05-26 — RT @AgustinLebron3: @DeepDishEnjoyer Like literally leaving the money in the bank or in a productive business is better for the world than burning it on some dig-holes-and-fill-them grifty non-profit. Mr Beast-style performative philanthropy feels somehow worse than no philanthropy. ↗
2026-05-26 — Compare it to the robber barons of the early 20th century. Rockefeller, Carnegie, Vanderbilt, etc. Billionaire 20somethings competing for social approval with their own caste will very predictably lead to some of the most insane and idiotic burnings of money the world has seen. ↗
2026-05-26 — This prospect is utterly terrifying. A bunch of newly-rich young autists funding whatever parochial non-profit ideas cross their mind? Oh boy. Consider the case of MacKenzie Scott, newly-rich due to marriage/divorce, funding an embarrassing set of grifty political causes. ↗
2026-05-25 — Degenmaxxing still has a ways to go, it seems. ↗

See @AgustinLebron3 on X →