Underrated Ways to Change the World

The most consequential work often happens in places nobody is looking. Three pieces this week make that case in different domains.

The overlooked paths

Adam Mastroianni published Underrated Ways to Change the World, presenting seven paths to meaningful impact that get almost no attention: amplifying whistleblowers, building communities, doing unglamorous government work, teaching in under-resourced schools, running local nonprofits, doing maintenance work on existing systems, and being a reliable person in unreliable institutions. Mastroianni's argument is that the choice isn't just "gain power" or "make extreme sacrifices" — there's a vast middle ground of high-impact, low-glamour work that most ambitious people overlook because it doesn't have a narrative arc.

The five-hour conversation

Lex Fridman's interview with Dario Amodei — plus Anthropic researchers Amanda Askell and Chris Olah — runs over five hours and earns most of them. The most substantive sections: Amodei's framework for thinking about AI safety that doesn't require knowing the exact timeline, Olah's work on mechanistic interpretability (actually understanding what's happening inside neural networks), and Askell's description of how Claude's "character" is designed. Amodei's prediction that AGI breakthroughs could happen by 2026-2027 is notable, but the more interesting claim is his argument about why the transition period — the years between "AI is useful" and "AI is transformative" — is when the important decisions will be made.

What just happened

Ethan Mollick's What Just Happened, What Is Happening Next is a useful snapshot of the current moment — a wave of capability leaps that arrived faster than most observers expected and what they signal about the near-term trajectory. Mollick's strength is specificity: instead of vague predictions, he describes concrete experiments he's run with the latest models and reports what they can and can't do. The post's most useful contribution is a framework for evaluating AI claims: look at what it does on your tasks, not on benchmarks.

The thread

Overlooked paths to impact, a CEO's safety framework, and a professor's empirical tests. The connective tissue: the people doing the most useful work right now aren't the ones making the biggest claims. They're the ones doing the most careful testing.