This paper argues that AI models’ thinking traces aren’t “real thinking” because swapped, noisy, and arbitrarily-truncated reasoning traces preserve or improve solution accuracy. When I examine my own train of thought when problem solving, I also wonder whether my “real thinking” is load bearing. Stop Anthropomorphizing Intermediate Tokens as Reasoning/Thinking Traces! — Subbarao Kambhampati et al.

Joan Tollifson writes a beautiful piece, steeped in paradox, about what it means to be ok with yourself, even with the parts of yourself that are not ok with you. I’ve definitely had the experience of accepting that something is unacceptable to me. Okay with Being Just As You Are? — Joan Tollifson

Chinese Cooking Demystified maps 63 distinct Chinese regional cuisines and substyles. There is a companion video if you don’t want to read all this, but it’s great. I found it fascinating to learn about how much variety there is in Chinese food. 63 Chinese Cuisines: the Complete Guide — Chinese Cooking Demystified

Kevin Munger on AI as an enabling condition for new forms of quantitative social science: a lot of the ideas in here should be familiar - decomposing the scientific PDF into structured, auto-updating artifacts organized around a scientific ontology. Great stuff. AI Allows More Diversity in the Forms of Social Science — Kevin Munger

Timothy Burke applies Tetlock’s Superforecasting findings to institutional decision-making: the most successful forecasting teams adopted rules that required the group to argue even when they happened to agree — “no shelter for yes.” No Shelter for Yes — Timothy Burke

Henrik Karlsson writing a beautiful piece last year on the biology of sustained attention. When attention holds, dopamine, hormones, and working memory synchronize and reinforce; jhanas are the inverse of a panic attack; cortisol’s 60–90 minute half-life is why frequent task-switching leaves the system decohered. Almost Anything You Give Sustained Attention to Will Begin to Loop on Itself and Bloom — Henrik Karlsson

A fake skin condition called bixonimania, seeded by Almira Osmanovic Thunström into two preprints in spring 2024, was cited as real by Bing, Gemini, Perplexity, and ChatGPT within weeks. This is a great example of why we need some kind of real grounded layer of scientific evidence. Relying on the foundation models by themselves to know what’s real doesn’t work. Scientists invented a fake disease. AI told people it was real — Chris Stokel-Walker

L.M. Sacasas’s piece provides a very compelling metaphor for the way social media affects our minds: digital platforms are doing to inner life what land enclosure did to the commons — turning the psyche into a resource to be managed and extracted. “Resist the enclosure of the human psyche.” Our attention is our most valuable resource. The Enclosure of the Human Psyche — L.M. Sacasas

Evan Erickson on a failure mode in opening-awareness meditation: if your basic orientation toward experience is one of pushing-away rather than welcoming, the whole practice quietly becomes a more refined form of avoidance. In other words “remain uninvolved” as a meditation instruction can lead to a kind of bypassing. Unlearning Default Awayness — Evan Erickson

Fascinating experiment which tests different organizational forms for AIs in solving complex problems - solo agent versus delegation to subagents versus markets. Adding an explicit error correction loop for markets seemed like putting the thumb on the scale a bit in favor of markets. Takeaway: AIs are bad at delegating and breaking up problems in subproblems intelligently. Why Coase needs Hayek — Rohit Krishnan

MacIver on C. Thi Nguyen’s The Score. Been meaning to read Nguyen’s book - his ideas seem very important in our metric driven era. MacIver’s long essay is worth reading for the account of the four horsemen of bureaucracy alone (those are: rules, replaceable parts, centralized control, and scale). How to be less box-shaped — David R. MacIver

Alison Roman: Spanish tortilla is more a feeling than a recipe (isn’t a lot of great home cooking?) — potatoes confit-cooked in olive oil, six or seven eggs, flipped (flipping is the hardest part!) or broiled to set, the reserved oil making the aioli. Spanish Tortilla: more a feeling than a recipe — Alison Roman

Sasha Chapin with a great little interview of Zen teacher Henry Shukman on the curriculum after koan training (mindfulness, support, absorption, awakening) and Shukman’s COVID-era head injury weakening his cognitive capacity and opening his heart: “when there’s no resistance to heartbreak, it’s actually not a problem. It becomes indistinguishable from just an open heart.” It goes without saying that Chapin’s writing is a gem, and my contribution is only linking to his older stuff which you maybe missed or forgot about. A Little Drop of Sweet Surrender (a conversation with Henry Shukman) — Sasha Chapin

Jared Henderson’s curated list of free philosophy lecture series on YouTube — Dreyfus on Heidegger, Kagan on death, Sandel on justice, Brandom on Frege, Sadler’s paragraph-by-paragraph Hegel. Sometimes I watch these imagining David Chapman chastising me for indulging in philosophy. Guilty pleasure. The Best Philosophy Lectures on YouTube — Jared Henderson

Mi’sen (another spiritual practitioner with a culinary passion) on Lyon as a “power place” in French cooking, the Mothers of Lyon, and a chicken-in-vinegar recipe sitting between Alain Chapel and Simon Hopkinson — plus the historical claim that French cooks were the ones who saved post-war British food culture in the 60s–80s. Despite a number of visits to France, I haven’t been to Lyon yet, which is criminal. Lyon: t’art & soul — Mi’sen

Victoria Lynn Carroll on dying as the answer to a long riddle, via a Death Cafe and the Bill Hicks roller-coaster bit. “Won’t it be fun to finally know?” A different kind of memento mori. I’ve read much of Andrew Holocek’s book on dying, Tibetan style, and didn’t see anything in there about fun! What If Dying Was Fun? — Victoria Lynn Carroll

@rohit4verse argues against getting on the latest tooling bandwagon for AI and instead focus on learning the fundamental primitives -- evals, context engineering etc. I have approximately 100 of these best practices type tweets saved. Each one is like a little time capsule of AI at a particular moment. I like this one since it goes against the grain of “use this specific tool” But I just spared you most of the need to read it. What to Learn, Build, and Skip in AI Agents (2026) — @rohit4verse

The best Caesar dressing has dill and caraway in it, courtesy of Jeremy Salamon’s Second Generation by way of Emily Nunn. I just noticed this is behind a paywall. Saving you a click again - just put dill and caraway in your Caesar. Caraway is a super underrated seed - I typically only see it on rye bread. Another underrated spice? Fenugreek. Emily Nunn is cool though - even just her post titles give me cooking ideas. Worth clicking around to her free stuff. The Best Caesar Salad You’ll Ever Eat — Emily Nunn

Andy Hall (another must read) reports on a fascinating experiment in AI governance where AI tries to learn students’ preferences and then represent them in a virtual legislature. How do you get an AI to understand your worldview? Training AI to Govern for Us — Andy Hall

Henry Farrell persuasively argues that the harms from social media don’t stem from disinformation, but rather from internalizing incorrect assumptions about the prevalence of certain views. Like for instance you might think that racist beliefs are much more common than they really are based on social media alone. We’re getting the social media crisis wrong — Henry Farrell

The Countess Caroline von Keyserlingk was Kant’s closest friend for thirty years, probably his lover, and the likely source of the French turn in his work around 1760. I loved this essay - it totally blew up my assumptions about Kant’s boring conventional life. The Most Important Woman in Kant’s Life — Daniel Andreas

Richard Danzig’s 2018 CNAS report, written before LLMs and aging scarily: “superiority is not synonymous with security,” the human-in-the-loop reassurance is too weak and steadily eroding, and “early is imperative, late is too late” for control of complex opaque autonomous systems. Just because we can doesn’t mean we should. Technology Roulette — Richard Danzig

Kevin Kelly bets against the AGI-resolves-by-2029 vibe: AI keeps advancing in ways that expand our ignorance rather than answering it, and US-China duopoly, post-globalization social chaos, and AI-induced media-trust collapse compound for a 10–15 year stretch of uncertainty about uncertainty itself. This seems plausible to me. It feels like the world is more uncertain than it ever has been in my life. Our Uncertain Uncertainties — Kevin Kelly

Or is that perception of everything becoming more uncertain a kind of illusion? Adam Mastroianni in this article from 2024 (ah, a simpler time): apocalyptic beliefs are so common across cultures because they feel reasonable, not because they feel good — extending his and Dan Gilbert’s “illusion of moral decline” work to explain why people persistently expect the world to end soon. Re-reading this piece, I am struck that they don’t really ever show that their proposed mechanisms (biased attention and biased memory) actually cause the illusion. The end is nigh and here’s why — Adam Mastroianni

New paper which scores 6,957 Organization Science submissions and 10,389 reviews with Pangram and find a 42% post-ChatGPT volume surge driven entirely by AI-flagged work, with editorial outcomes deteriorating sharply above a 30% Pangram threshold and the volume surge tracing to publication-count incentives at business schools. Fun fact: I know the last author (Lamar Pierce) from grad school and have a fun memory of him standing on one foot in a kitchen at a dinner party exclaiming “I’m really good at balancing.” More Versus Better: AI, Incentives, and the Emerging Crisis in Peer Review — Gartenberg, Hasan, Murray, Pierce

Dan Davies applies cybernetics and public choice to outsourced state functions and identifies “management capacity” — the ability to process information and respond to feedback — as the load-bearing missing quantity in modern bureaucracy. I’m a big fan of Davies’ work. Taming the unaccountability machine — Dan Davies

John Psmith’s review of F.W. Mote’s Imperial China: 900-1800 argues that barbarism and civilization aren’t a binary but a dial humans turn in response to incentives. Fun read, though I can’t remember why I saved this article a year ago… REVIEW: Imperial China, by F.W. Mote — John Psmith