Weekly Top Picks #107

Apr 21, 2025

∙ Paid

Google is targeting AI game-making for virality: When you target a measure, it stops being a good measure—unless the measure is a moving target.
“AI as Normal Technology” by the AI Snake Oil authors: The most important article you will read in 2025 so far: the singularity is not near.
People cost OpenAI millions of dollars saying “thank you”: “Dollars well spent,” says Sam Altman, as the robots cleanse the world of unapologetic twats.
Nvidia’s role in the US vs. China semiconductor war: What is arguably the most important American company in AI is market-first, not nation-first.
Sampling can only prove the presence of knowledge: Another instance of an evergreen lesson that AI skeptics have yet to learn.
A clue about the future: No comments.

(PAID) Weekly Top Picks #106: AI is for personal support / A market for AI companions / OpenAI's incoming releases / The "Era of Experience" / How GPT-4.5 was made / My "Google is winning" post on Hacker News
(FREE) AI Doomers Got the Superintelligence Thing All Wrong: Human ambition is not a virtue but a bug. A sufficiently intelligent and capable AI wouldn’t fall victim to an endless quest but turn inward into endless bliss instead.
(PAID) Why AI Isn't Getting Better at Writing: Writing well is much harder than people think. In AI, what’s easy for us is often hard for them, and vice versa.
(PAID) OpenAI o3 and o4-Mini Are More Impressive Than I Expected: The AI race is not about evals but about usefulness in the real world. o3 and o4-mini integrate reasoning, perception, and action (tool use), topping this ranking.
(FREE) I Wish I Believed What Edward Norton Says About AI: Critics and skeptics are in denial, but I don’t resent them, I envy them. Why aren’t more people who believe AI will make us obsolete mourning this eventual tragedy?

The Algorithmic Bridge