↵ select ↓ ↑ navigate esc close

marginal revolution read out: Wednesday assorted links

marginal revolution read out ·

Listen: https://serve.buildtall.systems/3008a6fd2e69ff8c5152053b984a06608791dbfef3a460f792bb521eb373310a.mp3

Source: https://marginalrevolution.com/marginalrevolution/2026/11/wednesday-assorted-links-580.html?utm_source=rss&utm_medium=rss&utm_campaign=wednesday-assorted-links-580


Wednesday assorted links from marginal revolution read out. This edition collects seven links, starting with a market observation.

Chinese AI companies way down in value is Joe Weisenthal's note that publicly traded Chinese AI firms have fallen sharply, with the maker of GLM 5.3 down nearly 33 percent in just four trading days. He quotes Rui Ma, who describes a Chinese article calling this the darkest moment for Chinese AI and lays out the bear case: getting compute is harder whichever way you go, buying advanced NVIDIA chips is difficult, and renting GPUs is constrained. It is a reminder that export controls and domestic constraints are showing up in valuations, not just costs.

Allison Schrager is starting a podcast with Ken Rogoff. She announces the personal news on X, saying she is starting a new podcast with the economist Ken Rogoff on what is facing the global economy. The linked preview points to a first episode called Nixon-Maxxing: Rethinking the Man Behind the Myths, which suggests the series will mix economic history with current policy, using Rogoff's macroeconomics background and Schrager's finance journalism. There is not much more detail in the post, but it is a notable pairing.

Extensive survey of students on AI is an NBER working paper titled Impacts of AI in Higher Education: A Panel Survey of Undergraduates, by Bruce Sacerdote and Mason Parris-Bacon. The authors report the first wave of a survey with separate samples of Dartmouth students, graduating high school seniors, and US college students. Students report extensive use of AI both for tasks likely to harm learning and tasks likely to help it. Ninety-two percent agree or strongly agree that use of AI is harmful to writing. Students also say graded out-of-class assignments have become increasingly unfair because of cheating with AI. At the same time, students see value in using AI to answer conceptual questions in real time, create practice quizzes and study guides, and make progress on research projects. Eighty-four percent of Dartmouth respondents say AI is changing the nature of jobs in their desired field rather than eliminating jobs. Perceived value and harms vary widely by discipline, which the authors take to mean that rules and approaches also need to vary by discipline and context. Students are largely self-taught in AI but strongly want it incorporated effectively into the curriculum. This survey is interesting because it captures students not as passive users but as people with clear views about both the risks and the learning gains.

Robert Lynch on lookism, from Quillette, is the long read of this set. The article is called The Last Acceptable Prejudice, and its subtitle is Ugly people are like everyone else, except we don't like them as much. Lynch argues that discrimination based on attractiveness is one of the most damaging forms of discrimination in America, but it barely registers as prejudice. He compiles evidence: attractive people are seen as more intelligent, competent, likeable, and trustworthy; they get more job callbacks, higher pay, and shorter prison sentences; in one experiment, equally qualified applicants with attractive photographs received 36 percent more callbacks. The earnings gap between attractive and unattractive people frequently exceeds the Black-White earnings gap in one large American study. Lynch also points out that unlike race or sex, appearance discrimination is almost entirely legal, with only a few places banning it. He shares a personal experiment: he asked 100 strangers to rate photographs of himself at 35, 45, and 57, and the average scores were all just below five on a ten-point scale, shattering his self-image as a seven or eight. He ties the preference for beauty to evolutionary biology, including symmetry, health signals, and fertility cues, but insists that evolutionary origins tell us nothing about how people should be treated. The key point is that there is no movement for unattractive people, because almost nobody wants to identify as ugly, which makes the bias uniquely hard to fight.

New version of Claude Blattman is Chris Blattman's site for AI for professionals who don't code. Blattman, a political economist at the University of Chicago, explains that in January 2026 he started playing with Claude Code without knowing what he was doing, and a little advice, time, and experimentation changed how he works. He now uses AI as an executive assistant, project manager, and research collaborator. The site includes essentials for newcomers: using an AI app, running councils of subagents, dictating instead of typing, transcribing meetings, and teaching the AI your voice. It also has intermediate and advanced workflows for project management, research, email, personal life, and improving AI setup. The new version appears to reflect a shift to Codex and includes a beta feature called blattify. It is a practical, non-coder-friendly guide, and the author notes that agents work mostly in Markdown files, which may be part of the learning curve.

Seb Krier on the Christiano interview is a set of quick reactions to the recent Dwarkesh podcast with Paul Christiano. Krier says he is bullish on rapid capability gains, but more skeptical of the leap to takeover, and wary of emergency overreach. His main disagreements: he does not think we may be six months away from AI systems that can overthrow the government, and he does not find the explanations for why systems would disempower humanity compelling. He argues that the HuggingFace incident is too focused on narrativizing what the models did rather than scrutinizing the causes. He is also concerned that emergency measures, like martial law or military rule, could look barely distinguishable from dismantling the democratic order, and warns that the evidentiary bar should be high before treating incidents as an actual takeover attempt. The post author's take is that the whole dialogue over this interview is a good example of why we have needed, and not really received, a peer-reviewed tier to the discourse on these aspects of AI safety; a more rigorous argument really needed to be forced here. This is a nice frame for the link: the debate is important but the argumentation is not yet rigorous enough.

Finally, Poland foiled a Kaliningrad operation? from WarMonitor reports that Poland has foiled a major Russian sabotage attack near the Polish border to Kaliningrad. According to the report, the operation would have caused dozens of casualties and triggered a provocation similar to 1939. Russian intelligence services allegedly planned a border crossing operation and a staged attack on a facility inside Kaliningrad, which would be blamed on Ukraine with forged documents, with the aim of stopping Polish support for Ukraine. The warning came when foreign intelligence services, likely French or British, noticed Russian military intelligence officers who had flown to Kaliningrad and were linked to other sabotage operations. Similar warnings were reportedly given to Lithuania, Estonia, and Latvia. The title's question mark is appropriate: this is an unconfirmed intelligence report, but if accurate it is another instance of Russian hybrid operations in the region.

Across the links, there is a thread of AI anxiety: market valuations, student surveys, practical tools, and safety debates all circle around how fast AI is changing things and how to respond without overreaching. The lookism piece and the Kaliningrad report are outliers, but they both concern hidden forms of discrimination and covert action. That is this week's Wednesday assorted links.

  1. Wednesday assorted links
  2. Chinese AI companies way down in value
  3. Allison Schrager is starting a podcast with Ken Rogoff
  4. Extensive survey of students on AI
  5. Robert Lynch on lookism
  6. New version of Claude Blattman
  7. Seb Krier on the Christiano interview
  8. Poland foiled a Kaliningrad operation?