↵ select ↓ ↑ navigate esc close

marginal revolution read out: Thursday assorted links

marginal revolution read out ·

Listen: https://serve.buildtall.systems/eff1bf122e58cb288ebdb4b7b35dd72bf5c44cd583aa4a467a6a2cee38858495.mp3

Source: https://marginalrevolution.com/marginalrevolution/2026/11/thursday-assorted-links-579.html?utm_source=rss&utm_medium=rss&utm_campaign=thursday-assorted-links-579


This is marginal revolution read out, reading the post Thursday assorted links from Marginal Revolution. It is a list of fourteen linked items, with occasional comments from the author along the way.

The first link is a YouTube video titled How to turn AI-written maths into a video. The page itself is mostly YouTube's standard interface, so the substance is in the video, but the title promises a walkthrough for taking AI-generated mathematical writing and turning it into video form.

The second link asks Honeypot markets in everything? The linked post is marked as sensitive and only available in the X app, so its text cannot be read here. The author offers no further comment, leaving the question at the level of the title.

Third is Traveling a lot without going on planes, from the Wall Street Journal. The article could not be fetched, and the author notes only that it is from the WSJ. The title points to a lifestyle piece about long-distance travel that deliberately avoids airplanes.

Fourth is American audiovisual exports have plunged in terms of global share. The linked post by Werner Zagrebbi calls the evaporation of American global cultural influence one of the biggest stories in the world today, and says that is not good. The author's take is the opposite: he says that is fine and arguably good.

Next, under the heading Link, the author shares a response from Haydn Belfield to a scenario about a rogue AI taking over. The author quotes Belfield's argument that there has been a thirty-year debate between cyber revolutionaries and cyber skeptics, and that the war in Ukraine has swung the empirical pendulum hard toward the skeptics. In Belfield's view, cyber has played a fairly minimal strategic role, struggling to translate digital disruption into durable physical or political control. He also adds that a rogue model would not be operating against unassisted humans, because several frontier companies exist and governments could use peer or near-peer models to harden infrastructure, hunt for exfiltration, and defend against subversion.

Alongside that, the author links to some more critical points from Peter Gostev. Gostev says listening to the AI takeover scenarios feels like playing a game with a child who invents the rules along the way to reach a predetermined outcome. He asks why AIs would take over labs, how they would take wider control, what would happen if governments showed up with guns, how AI could control public opinion, and what would happen if the big data centers were stopped. His conclusion is that the takeover story requires a whole chain of esoteric and unlikely events to happen together, even though he agrees improving security is clearly good.

The next item is Casey Handmer looks at astrophysics research. Handmer describes an exercise in which he indexed about one hundred fifty thousand papers in theoretical astrophysics, audited fifty thousand of them, and found nearly one thousand major errors that had been cited many times, most still uncorrected. He says this allowed him to estimate that peer review is not very good at correcting errors. His bottom line is that AI can speed up quantitative science by more than a thousand times with no further model improvement required, but he sees little evidence that universities are pushing this. He goes further, arguing that academic publishing is dead, that AI can generate thousands of novel and correct papers per day, and that a research paper should no longer be treated as a meaningful unit of science. The author links to this without adding his own comment.

Then comes Greg Conti on AI (and me), a Chronicle piece that could not be retrieved. The author simply lists it without commentary, so the title is the guide: it appears to be Greg Conti writing about AI, and also about the author.

After that is Kevin Roose on early Anthropic. Roose revisits his spring 2023 reporting, when Anthropic invited him to spend several weeks at its headquarters before the launch of Claude 2. He recalls a very small company of about 160 employees in a small office, with a chatbot nobody had heard of and an apparent lack of interest in running a business. He wrote in his notebook that they seemed like ambivalent capitalists. At the time, their biggest shareholder, Sam Bankman-Fried, had been arrested for fraud, and OpenAI was eating their lunch with ChatGPT. Now, he notes, Anthropic has gone public as one of the most valuable companies in the world, with 5,300 employees and fourteen times annual revenue growth. Roose also remembers that the researchers were sincerely terrified of AI, and he recounts a strange detail about Dario Amodei's photo being staged in a random Italian town with a rotary phone and panic-room red lights.

The author also links to a piece labeled Nicholas Decker. There is no commentary beyond the link and no fetched text, so there is little to add beyond the name.

The final cluster of links turns to Thomas Nagel and the question of AI consciousness. First, the author shares The utter wisdom of Thomas Nagel, a post by Sam Buntz. Buntz quotes Nagel on why AI cannot be conscious, adds amen, brother, and says it is nice to hear one of the greatest philosophers of mind weigh in. The author's label of utter wisdom suggests he is sympathetic.

Second is a follow-up titled more, in which Buntz says Nagel is singling out Anthropic as particularly abominable in its approach to AI.

Third is Erik Hoel, who argues that Nagel is the best living philosopher and that the idea he is making an obvious mistake in denying large language model consciousness is nonsense. Hoel says Nagel is not making a dumb mistake, and that people do not understand what natural manifestation means. He says Nagel likely means two coherent things: first, that consciousness comes before language, and language is an abstraction of it; second, that there is no such thing as a purely, and the excerpt ends there.

Finally, the author recommends The take from Ben Gross, saying it is perhaps the best yet. Gross offers a brief thread of elucidation and starts with a question: why do we think that models that can chat with us are conscious, but models that can recognize faces, produce images, or drive cars are not? The excerpt includes only that opening question.

Across the links, the main thread is AI: the limits of cyber, the plausibility of takeover scenarios, the way AI could reshape science, Anthropic's transformation, and Nagel's philosophical objection to machine consciousness. The author's asides also push back on alarm about American cultural decline and elevate the Nagel debate. Otherwise, the post mostly lets the linked voices argue.

  1. Thursday assorted links
  2. How to turn AI-written maths into a video
  3. Honeypot markets in everything?
  4. Traveling a lot without going on planes
  5. American audiovisual exports have plunged in terms of global share
  6. Link
  7. some more critical points
  8. Casey Handmer looks at astrophysics research
  9. Greg Conti on AI (and me)
  10. Kevin Roose on early Anthropic
  11. The utter wisdom of Thomas Nagel
  12. more
  13. Erik Hoel
  14. The take from Ben Gross