AI

• “Optimal Caverna Gameplay via Formal Methods”, Stephen Diehl (formalizing a farming Eurogame in Lean 4)

• “Do Not Render Your Counterfactuals”, SelfMadeHuman 2025-07-30 (emotional harm from visualizing life regrets using image generators)

Matters Of Scale:

• “Pre-training under infinite compute”, Kim et al 2025 (ensembling + heavy regularization = sample-efficiency); “NanoGPT Slowrun: Language Modeling with Limited Data, Infinite Compute” competition (5.5× data efficiency so far from proper multi-epoch training, heavier regularization, SwiGLU, & ensembling; of course LLMs could still be much more sample-efficient…)

• “How Well Did Superforecasters and Experts Predict Wet Lab Skill Uplift from LLMs? An RCT tested whether LLMs could help novices at molecular biology tasks. Here’s what forecasters predicted before the study, and how they updated their views on biorisk after learning the results”, Forecasting Research Institute 2026-02-19 (mid-2025 LLMs, cf. METR)

• Claude-4.6-opus substantially helps Donald Knuth on an open research problem

• “Implicit meta-learning may lead language models to trust more reliable sources”, Krasheninnikov et al 2023 (Even base LLMs are not simply a weighted average or sample of their training data—because that yields worse predictions on average! as compared to using things like theory of mind or understanding fact vs fiction etc.)

Genetics

Everything Is Heritable:

Recent Evolution:

Engineering:

Statistics/Meta-Science

• “How to win a best paper award (or, an opinionated take on how to do important research)”, Nicholas Carlini2026

Politics/Religion

• “AI use in American newspapers is widespread, uneven, and rarely disclosed”, Russell et al 2025

• “The Brand Age”, Paul Graham 2026

Psychology/Biology

Technology

• “The Brand Age”, Paul Graham

• “Personal Encyclopedias”

• Toothpaste § Striped toothpaste

Economics

Philosophy

• “Conversation with Jeanne”, Czeslaw Milosz 1991

Fiction

• “Dear Aliens: a writing contest for humans”, Taylor Troesh2026 ($2,500 in prizes; I’ll be a judge); “The Un-Slop Fiction Prize: $10,000 for the best AI-generated short story”

• O. Henry must surely be on the Pareto frontier of “widely read/known” vs “reader awareness of an interesting life”. Everyone reads him in school, but I can’t remember ever covering stuff like “he starting writing the stories while imprisoned for fleeing to Central America after embezzlement charges having returned for his wife dying of tuberculosis”.