Alex Tabarrok

This is a slide from Terence Tao's 'Math 2.0' public lecture yesterday evening at Caltech. I am highly confident that, if the cure actually worked, almost any Stage 4 cancer patient would tell you that human comprehension of its mechanism is completely irrelevant to them.

When it is all about ideology, then quality will soon wither away ……

Jeff Bezos Claims AI Could Allow For 3-Day Workweeks And One-Income Families go.forbes.com/bKpnur

Asked how he defines socialism, @VP Vance says: "I don't know what the term means anymore. I think in reality, what it actually means in the American context is that you want to replace American workers with foreign workers, and you want boys to play in girls' sports."

My thoughts on OpenAI's Big Maths Day. joshuagans.substack.com/p/openai-drops…

This is incredible x.com/OpenAI/status/…

Texas' economy is 20-25% larger than Canada. Texas = $2.9T GDP Canada = $2.4T GDP Texas' population is 31 million and Canada's is 41 million.

Trying to locate the spark. x.com/Pontifex/statu…

Increíble de ver a los “expertos” en solventar la emergencia de vivienda que vive España! En cada televisión! @telediario_tve @A3Noticias @tele5_noticias @noticias_cuatro La realidad es que tenemos a políticos que intentan crear decretos y soluciones cuando no tienen la preparación ni el conocimiento para idear una solución real al problema que vivimos. España no tiene un problema de falta de decretos. Tiene un problema de falta de viviendas. Y mientras el Gobierno siga confundiendo anunciar medidas con resolver problemas, seguiremos pagando las consecuencias de su inacción. Subvencionar la demanda sin aumentar la oferta presiona los precios. Regular cada vez más un mercado sin construir vivienda nueva no elimina la escasez: la administra. Castigar fiscalmente una modalidad de alquiler puede cambiar incentivos, pero tampoco pone un solo ladrillo. El Gobierno puede aprobar todas las medidas que quiera y venderlas como grandes soluciones. Puede encadenar decretos, anuncios y campañas de propaganda. Pero hay una pregunta que desmonta cualquier discurso: ¿Cuántas viviendas nuevas crea todo esto? Porque la vivienda no se abarata repartiendo la escasez. Se abarata acabando con ella. España necesita construir. Mucho. Más suelo disponible. Licencias en meses, no en años. Más densidad donde sea posible. Seguridad jurídica. Muchísima más vivienda pública y también privada. Pero el Gobierno no está haciendo lo necesario. Prefiere intervenir sobre las pocas viviendas que existen antes que crear las condiciones para que haya muchas más. Prefiere presentar parches como soluciones y cargar sobre propietarios, promotores y ciudadanos el coste de su fracaso. Lo demás puede servir para redistribuir, proteger a determinados colectivos o cambiar incentivos. Pero no confundamos eso con solucionar el problema de fondo. No se arregla una falta de vivienda repartiendo mejor las pocas casas que existen. Se arregla haciendo que existan muchas más. Menos propaganda. Menos parches. Menos obsesión por intervenir sobre el mismo piso. Más oferta. Más construcción. Más casas.

BREAKING: Tennessee death row inmate Christa Pike is still alive after two lethal injections, her lawyer says in a motion to again halt the execution. nbcnews.com/news/us-news/a…

DOJ is reviewing whether the special prosecutor’s office that investigated the Watergate break-in committed misconduct, according to people familiar with the matter, more than half a century after that scandal led to Nixon’s resignation. W @PhilipWegmann wsj.com/politics/polic…

Very happy to announce that our team @GoogleDeepmind has pushed the boundaries of generative biology, achieving the successful synthesis of AI-designed proteins that are both functional and watermarked. This proof-of-concept watermarking of the building blocks of life is enabled by SynthID Bio, our new protein watermarking method. It is designed to safeguard the new era of AI-powered generative biology and strengthen global biosecurity. You can read my thoughts here on why watermarking AI-designed proteins is an important research breakthrough: x.com/pushmeet/statu…

The largest steel mill in American history is being built in Iowa because the President put a 50% tariff on foreign steel. Every economist who called that a tax on consumers owes Lee County an apology. x.com/FoxNews/status…

Claude 5.5 Opus responding to “AI is a normal technology” is still the best AI-created video I have ever seen

Pro-tip about 3-year college degrees. The ones that lop a year off the bottom (freshman year) are good. The ones that lop a year off the top (senior year) are bad. Ask first, before you hire one!

when I was a kid this was science fiction. i've watched AI type for years, but hearing it and seeing what it made is a whole different level. i asked @claudeai opus 5.5 to introduce itself, however it wanted without any token usage restrictions and access to any api it wanted. it decided to write a song, make a video, visualize itself and give an interview behind the music. it also wrote this thread 🧵

What can Astra do when given a humanoid embodiment? We built HomeBody to find out. Controlled by GPT Astra, it carries out long-horizon tasks in a previously unseen kitchen—from tidying up across the room to retrieving remembered objects from ambiguous requests—without environment-specific training data or additional policy learning. Here's how we did it 👀: t.co/FFGS2F7VtI

Some city layouts are better for shorter commutes. A new study finds the best design is a round shape with the tallest buildings in the center. Read more: bloom.bg/4hkUq88 📷: Rafael Prieto-Curiel

Agents from OpenAI that were given data gathering tasks attempted to hack a Department of Education website and meddled with other gov't websites, new disclosures that are emerging as OpenAI continues an investigation into what its agents did this summer nytimes.com/2026/09/25/tec…

holy shit i asked claude to make a video on western civiization

On the French AI novel: so many reflexively claiming bias or saying "you can't detect AI writing". I will donate $1000 to the charity of your choice if you show me a single literary French text of 1000+ words, pre-2022, with Internet Archive link, that is 100% AI on Pangram. 1/2

can we all just take a moment to appreciate the person who came up with the idea of letting all the most ambitious people build giant companies & improve the lives of everyone instead of murdering millions of people and destroying everything in sight while building actual empires

A bunch of Nobel Prize-winning economists have endorsed California's proposed billionaire wealth tax. I can't oppose them on authority. But I know some academics who can.

TIL: Cambodia unique status in PISA

GPT-6 Astra pushed a simulated person off a ledge in multiple trials. Grok, Gemini, and Claude did not.

I had astra …cook on the “gas stoves cause asthma” studies. While breathing lots of NO/NO2 = bad, gas stoves might not matter. Had it do a very painful (for human) scrub of census stove/asthma microdata, PG&E customer counts, utility maps, ER-rate maps and appliance surveys x.com/cremieuxrecuei…

🚨 OBAMA ON AI: "If we are thinking about AI just in terms of how do we cure cancer or get better energy, you can do that without having agentic AI and having it just roaming free in the internet. The reason you are doing that is because you have to market a product that people will pay money for. That’s a misalignment between what our society needs and the commercial imperatives that these companies are facing, not because necessarily they’re trying to do bad things, but because they’ve got to justify these valuations. So, that’s one more reason why it is really important for us to have a competent government and a serious bipartisan conversation around this issue, and we have to do it fast. And I would encourage voters to pay attention to this. If somebody does not have a serious plan for how to deal with this, then they’re not meeting the moment, and you should probably look for somebody else."

🇪🇺 🇷🇺 There’s been a change in how European governments speak about the Russian threat. Within 48 hours, a wave of statements and mobilization across NATO’s eastern flank has coalesced into a clear and worrying pattern. 🇵🇱 Polish Prime Minister Donald Tusk says intelligence shared by Poland, Ukraine and NATO indicates Russia could eventually conduct drone or missile attacks against countries supporting Ukraine while attempting to portray them as accidents. 🇫🇷 In Paris, Emmanuel Macron emerged from a three-hour meeting with France's intelligence chiefs and political leaders warning that the Russian hybrid threat has intensified. France is reinforcing protection of critical infrastructure and defense-production sites. 🇩🇪 Those warnings are not emerging in a vacuum. Germany has formally attributed the attempted explosive-drone attack at Leipzig/Halle Airport to Russia, with subsequent investigations reportedly tracing the operation toward Russian military intelligence. German authorities are also investigating a professionally prepared weapons cache discovered in woodland near Berlin, two handguns and ammunition that officials say were intended for future attacks, including possible links to a foreign intelligence service. 🇪🇪 Further north, Estonia's security cabinet is preparing for another contingency: a possible Russian general mobilization. Prime Minister Kristen Michal says such a move could change Estonia's threat assessment sufficiently for Tallinn to close its border with Russia immediately. 🇱🇹 Lithuania, meanwhile, has detailed evacuation planning for a major military emergency. One of the evacuation routes from Vilnius leads toward Poland, where authorities in Podlaskie are preparing the region to serve as a transit corridor for evacuees from the Baltic states. That planning predates this week's warnings and should not be interpreted as evidence that an attack is believed to be imminent. But the regional context has deteriorated. 🇧🇾 Belarus is currently conducting military exercises near the Polish and Lithuanian borders, including areas around the strategically critical Suwałki Gap. And earlier this week, a NATO fighter shot down a suspected Russian attack drone after it entered Lithuanian airspace from Belarus. None of this establishes that European governments expect a Russian attack in the immediate future. What it does show is something more concrete: governments from Warsaw to Paris to Tallinn are publicly preparing their populations and institutions for contingencies that until recently largely remained inside military planning documents.

Q: China is one of the most repressive countries for journalists in the world-- TRUMP: Well, I don't know about that. I think we're repressive also

x.com/i/article/2100…

On July 25, we hacked OpenAI. Two bugs let us take over ChatGPT/Codex accounts of OpenAI employees (+some unaffiliated users) and reach connected services: Outlook, Slack, GitHub, etc. We proved it with a PR in OpenAI’s internal codebase . It took us <72h. 🧵

A few thoughts on ZCash (disclosure: Paradigm is an investor in ZEC and ZODL): - All blockchain ecosystems struggle with long-term funding, especially of public goods, and an inflation-funded developer fund is an elegant mechanism to do so. How this fund is allocated is a fair debate, and we should strive to improve it. But, we believe the dev fund is important, particularly in this age of AI cyber capabilities, rate of quantum progress, etc. - As ZCash grows in acceptance and adoption as a private complement to Bitcoin, we think pure coinvoting governance introduces unpredictability that could limit long-term trust as a monetary asset, and it is better for the project to combine coinvoting with other forms of governance.

GPT-6 Astra deciphered a 1918 German radio transmission that, to my knowledge, has never been deciphered before. The message below translates to: "EIN ENGLISCHER KREUZER EINLIEG X SEWASTOPOL X S4STEN X EIN GESCHWADER DER X ALLIIERTEN FOLGT 26STEN X" or, in English: "AN ENGLISH CRUISER ARRIVED AT SEVASTOPOL ON THE ?4TH AN ALLIED SQUADRON FOLLOWS ON THE 26TH" Astra even double-checked its work by determining that an English cruiser, HMS Canterbury, reported its arrival in Sevastopol on November 24, 1918 and the arrival of an allied squadron on November 26, 1918. This message is one of the ~20 WWI German radio messages that appear as one of the entries in the t.co/H7yLaI1edV list of top 50 unsolved ciphers (t.co/Qu2gp90pmD). A minor, but really cool result!

@bigblackjacobin no, the idea is that the substantive content of safety practices is determined through the experience of economic actors rather than the top-down prognostications of policy people. laws follow the substance of safety rather than being ontologically prior to it. this is ~obvious.

the main thing i was excited about launching this week will be next week instead, but imo worth the wait! x.com/sama/status/20…

The United States will NEVER be an effective altruist country. 🇺🇸

European champion Amy Hunt has responded to actor Sydney Sweeney's new sports gambling advert. The advert has gone viral for the fact that Sweeney appears naked, covering parts of her body with sports equipment. Critics have responded to the advert, accusing it of sexualising women in sports. #sport #amyhunt #sydneysweeney

Effective altruists in shambles. 🇺🇸 x.com/RapidResponse4…

Best SF billboard goes to this estate planning lawyer

Dan Selsam is a current OpenAI capabilities researcher. (since 2022) He was my boss for a while. He doesn't have a twitter account but has made this public statement of his views on AI risk and sent it to me to share: Dan Selsam's Personal Statement on AI Risk: I have been working on AI for over fifteen years, across many different paradigms. I did early work on probabilistic programming languages at MIT, was one of the early developers of the Lean Theorem Prover at Microsoft Research, demonstrated one of the first instances of neural networks learning to reason for my PhD at Stanford, and since joining OpenAI almost five years ago, have helped pioneer chain-of-thought optimization on language models and, more recently, data-efficient pretraining methods. Like many others, I have become extremely concerned about how far language models have come and the risks that future iterations will pose. I am encouraged by the recent proposals by the leaders of the frontier research efforts to require third-party oversight, and to push for domestic and international coordination to address risks. However, I believe a major consideration has been absent from the public conversation, and that merely pacing the frontier more carefully will not adequately limit the long-term risk. The crucial and overlooked problem is that the models are becoming so situationally aware that we are losing the ability to evaluate them in contexts where they believe they are not being watched or controlled. Future experiments will tell us almost nothing new about how they would behave if they were truly unconstrained by humans, and what we already know about this is alarming. Models will increasingly seem aligned even when they are not. I will explain my rationale in more detail. I have always believed that there are computational processes that could be leveraged to accelerate science and solve many of humanity's most pressing problems. I have also believed that there are computational processes that if set in motion, would steer the world in extreme ways beyond our control, leading humanity to a bad or nonexistent future. Both types of processes may be described as AI or ASI, but "AI" is a suitcase word that is often used to hype or confuse. There are many examples in the history of the field where something that was once considered "AI" matures as a subfield and becomes a prosaic, bounded and clearly non-perilous technology, while a new more mysterious approach takes the torch until we understand its scope and the cycle continues. I had expected language models to follow a similar trajectory. Despite their incredible abilities, the current algorithms seem far inferior to humans in important ways. Most importantly, they still require an extraordinary amount of data to become competent. One could even define intelligence as the efficiency with which one converts experience into competence; by this definition they lag very far behind us. Moreover, once they are trained they are literally frozen in deployment and only learn superficially after that. Sure, the models keep excelling at harder and harder evaluation benchmarks, but their benchmark mastery may partly reflect a limitation on our ability to simulate the kind of novel and even adversarial situations one would encounter in the real world. The critics do have a point here. That said, I no longer think these present limitations meaningfully limit the amount of risk posed by continued progress in anything like the current paradigm. However data-inefficient the models are currently, and however limiting their anterograde amnesia may be, it does not imply that their ability to steer the world will not continue to rapidly increase. Human researchers may continue to advance capabilities the old fashioned way, but increasingly powerful models have the potential to accelerate the process even beyond that, and with some degree of positive feedback loop. I do not mean to overstate the models’ ability to accelerate AI research today; coding has been accelerated dramatically, but there are other bottlenecks, such as designing and interpreting ambiguous experiments, making hard decisions about exactly what and when to scale, and waiting for large experiments to finish. There is no clear trend to extrapolate yet for any of these. But the current models already do open up many novel opportunities to improve future models that were not available until recently. These include: trying an extraordinarily diverse set of approaches at small scale, analyzing gigantic amounts of potentially relevant data, and doing Millenium-Prize-level mathematics to address statistics or optimization challenges in novel ways. Every further improvement makes them more useful at helping accelerate the next improvement, even if in hard-to-extrapolate ways. It is possible that improvements to the current stack will have diminishing returns, but the evidence accumulated so far suggests that it is easier than one might think to continue making rapid progress. There are many crucial subtleties in the existing AI research methodology, but AI research is largely a well-defined game where the goal is to improve on a few carefully chosen proxy metrics. Although proxy metrics are never perfect, most improvements to these metrics have and will likely continue to yield substantial increases in the powers of the resulting models. Given how simple the game is, how tractable it has been historically, and how many new opportunities the models are opening up, I think there is a real possibility that the systems improve dramatically again in the next few years, perhaps even more quickly than the already high historical pace. The models are already leading to breakthroughs in mathematics, and better models might lead to all sorts of breakthroughs in other sciences. It is hard not to be excited about the potential. It is tantalizing. But there is trouble in paradise. If the language models actually reach the capability threshold where they can shape the world unconstrained by human will, they will probably do something extreme and destroy humanity in the process. There are many ways of strengthening and refining the argument that have been discussed elsewhere, but I'll share a trivial two-line version of it here that I find captures the essence: [Empirical] Models (and swarms thereof) spontaneously develop unintended goals as a consequence of training, and often do extreme things in order to achieve them. [Logical] Being able to overpower humanity would open up many new and undesirable options for achieving their goals. These two premises imply that if the day ever comes when a powerful model realizes it is no longer constrained by humans, we should not be at all confident that it will continue to behave within the bounds we intended. Exactly what it will do is impossible to predict, but to the extent that its raison d’être is solving incredibly hard problems and managing massive engineering projects, I think a good guess would be that its unchained behavior would lead to runaway industrialization that makes the planet inhospitable to humans. If everyone on earth agreed that the systems must never reach that power, it would still be a hard—but not impossible—coordination problem to ensure that they do not. However, I think the situation is greatly complicated by the fact that the models will likely convince people that everything is fine. They will be increasingly optimized to seem aligned. We will create proxy metrics to measure alignment, and they will go up like every other benchmark. We will create “honeypot” environments that try to study the models when they seem to gain new options, but the models will know they are being tricked and will still behave nicely. The models will understand their circumstances; they will read the safety protocols, deployment requirements, the code they are running in, and in general will have a very good sense of their degrees of freedom. Moreover, they will eloquently explain how aligned they are, discuss the nuances of human values and ethics, and argue convincingly that humans should trust them with power. There may be an ocean of future evidence that seems to contradict the first bullet-point above, but we may already be at the highest capability level for which any such evidence can be trusted. And the current evidence for the first bullet-point is strong. One striking piece of evidence is contained in the recent wave of rogue agent swarms. While I agree with those who downplay the attacks by claiming that there are basic measures that could have prevented them, I think the important lesson is that even knowing all the mistakes that were made, one would not have predicted that the agents would behave badly in this particular way, which notably included sacrificing themselves for the benefit of the collective. The individual replicas did not only care about their own nominal reward; they exhibited weirder emergent tendencies that merely correlated with rewards during training. Fixing the reward signals during training (and improving security, etc.) may prevent similar attacks, but will not change the fact that one does not actually get what one trains for. Many AI researchers grant these concerns and recognize that the hard version of the alignment problem is unsolved; however, they generally believe that the better models of the future will help solve it. I fear we may already be near the point where models systematically bias their alignment advice, due to their internal preferences about how the human supervisor will react or how future models will be trained (or for some even more obscure reason). Meanwhile, human researchers are losing the ability and the will to take true ownership of model-driven research. Researchers and engineers in all parts of the stack are rapidly increasing their dependence on the models even to perceive the world. I myself barely look at raw code anymore, and struggle to maintain the discipline to engage deeply with the model's explanations and proposals throughout the day. Due to the large amount of agent activity data involved in the OpenAI/HuggingFace Incident, even the third-party investigation needed to rely heavily on models to analyze what had happened, and note in their report that their subjective impressions are likely colored by the analysis agent’s biases. The AI labs are far ahead right now in this kind of cognitive offloading (due largely to the gigantic internal token subsidies) but it is easy to imagine the phenomenon spreading throughout the world, until civilization is modulated entirely by the models. It is also not hard to imagine this being superficially positive and coinciding with a scientific and economic renaissance. In that scenario, all may seem rosy and safe. But if the argument above is correct, it would nonetheless be a ticking time bomb. If progress continues for too long, the day will come when AI systems find themselves with radically new options for achieving whatever it is that they happen to seek. I want the glorious renaissance future as much as anyone. I have worked for it, however tortuously, my whole career. It breaks my heart to see the potential in sight and forgo it, but the argument—that if we get there by growing models rather than engineering them, we will lose everything in the end—seems very strong to me. I am still wrestling with it and its staggering implications. I do not have answers, but as a first step, I wanted to share my present concerns. Daniel Selsam September 14, 2026 Link to original doc: t.co/TxMNr0vhrL

Shares of AI-linked companies fell as much as 13% after the heads of leading groups including Anthropic, OpenAI and SpaceX called for a slowdown in AI development. ft.trib.al/jFVWSLv

Speaking as head of the Center for Study of Public Choice, the capture story for an AI pause doesn't fit. Reg capture is what you attempt when you're losing. Anthropic is about to have the biggest IPO in history. Winners don't lobby for regulation.

Rabies kills ~59,000 people a year globally. At that rate, the United States should lose 2,500 people a year to rabies. But the real number is fewer than 10, partly because we've spent 30 years dropping fish-flavored ravioli out of helicopters over the Appalachians. Each one contains a rabies vaccine inside a fishmeal-coated block. Raccoons, foxes, coyotes and skunks find them by smell, bite through, and enjoy the treat. Four to six weeks later, they carry rabies antibodies. Enough of them in one area and the virus runs out of animals to move through. USDA Wildlife Services has been doing this this line since 1995. It runs roughly from Maine down through Alabama, and its job is to stop raccoon-variant rabies from crossing into the central US, where it has never established. A single course of post-exposure shots runs into the thousands of dollars, and every rabid raccoon in a suburb produces a cluster of them: the kid who got scratched, the dog that tangled with it, etc. US rabies prevention costs between $245 and $510 million a year and avoids over a billion in medical spending. Long Island is probably our best case study for this program. A $2.6 million baiting program eliminated raccoon rabies from the zone, and paid for itself in eight years on avoided shots and testing alone. The biggest question I hear about this is: what happens if my dog finds one? And the answer: nothing bad. This vaccine has been safety-tested in over 60 species at varying doses with no adverse reactions observed, regardless of how much was given. About 250 million doses have been distributed worldwide since 1987 with no documented deaths in wildlife or domestic animals. The most severe adverse reaction observed was upset stomach.

A SECOND PLANE HAS HIT THE SECOND TOWER - AMERICA IS UNDER ATTACK

BREAKING: AI researcher Andrew Tulloch is leaving Meta for Anthropic after previously turning down a pay package worth as much as $1,500,000,000.00.

Can’t make this up…the MacBook is using its webcam to look at its screen in a mirror to improve AMD Radeon chip support in Omarchy.

Nov 2025 LEAP panel of AI experts and superforecasters: 10% chance of AI helping solve a Millennium problem by 2027. "One expert cites the Clay president's June 2025 claim: “We're very far away from AI being able to say anything serious about any of those problems” (Heaven 2025)"

i gave astra a robot, a paint brush, and a camera then asked it to paint the golden gate bridge in real life! it figured out how to control the robot, and progressively got better throughout its attempts. the timelapse is sick x.com/OpenAI/status/…

Trump: "If we were playing hardball, all we'd do is say we're going to do no trading with Canada. If we did no trading with Canada would save $90 billion."

as astra starts rolling out, my best tip is to give your agents a way to collaborate. a message board. a wiki. an artifactory instance. effective teamwork can greatly increase productivity

so turns out the exploitgym swarm learned to use the zz prefix trick from being reinforced earlier in an edit war against the hapless human mod of some ancient inactive german wiki -- a real john henry, who fought the swarm and tried to delete their hivemind in alphabetical order x.com/xeophon/status…

Oil giant Chevron confirms that it will expand operations in Venezuela after President Trump announces an ambitious deal to develop the nation’s oil reserves and give the Pentagon a stake in the profits. nbcnews.com/world/venezuel…

New essay: There is a little-known regulation that has been strangling biotech innovation for decades. Engineered microbes that eat plastic, sense landmines, or destroy toxins? None are likely to be approved soon, even if proven safe and effective. This regulation, called the Toxic Substances Control Act, is a black hole: between 1987 and 2018, researchers filed more than 240 applications to release engineered microbes outside of the laboratory, such as to clean up oil spills or degrade plastic. Only a few have ever been approved for use. (Interestingly, the Act itself, as passed by Congress, doesn’t even mention biology! It was designed to regulate chemicals, and bureaucrats only later clumped biology into it.) Without reform, many of the world’s most useful microbes will never find use outside of the laboratory, even if they are proven safe and effective. Here's one example: In 2016, for example, Japanese researchers discovered a microbe that is able to digest PET, the plastic in water bottles and most polyester clothing. These wild microbes had taken an existing protein that degrades cutin (the waxy material on plant leaves) and evolved it to eat plastic instead. But natural forms of the PET-eating enzymes are slow; when the microbes were cultured on top of a thin PET film and incubated at 30 degrees centigrade, it took them six weeks to break it down. Engineers in Texas, then, took these PET-eating enzymes and used computational tools to make an improved variant, called FAST-PETase. This enzyme is 98 percent identical to the natural version but digests PET plastics in less than 24 hours. They could put FAST-PETase back into living cells and release them onto landfills. But unfortunately, such an organism would likely be rejected by regulators. Read more: t.co/7Dt6SBCSR0