Derek Thompson: don't dismiss AI long-tail risk worriers as insincere
birchlse · x · 2026-09-09
Derek Thompson says he has many issues with how AI labs frame their perceived prisoner's dilemma around recursively self-improving superintelligence ('just don't build the thing you don't want to build' is underrated). But he's gobsmacked by critics who claim those worried about AI long-tail risks are insincere or self-evidently bogus.
His argument: the rationalists and EA folks who knew the most about bio-risk spent the 2010s worrying about a pandemic—and then COVID happened. Barely a cosmic blink since it ended, people now claim certainty about the hysterically unlikely nature of long-tail risks. He isn't asking anyone to buy the full x-risk argument, only to hear the other side out long enough to understand it.
More from AGI Musings
- Designers Can't Define the Role Anymore: Hiring Brokenness Is Breaking Design Education — round · 2026-09-10
- Anthropic's AI economy model: 32% GDP boost by 2030 in extreme scenario, divergence after 2027 — terencemauri · 2026-09-10
- Antikythera's 'superdark factory': the fully automated, self-optimizing firm — tylerjdunn · 2026-09-10
- Schmidhuber accuses AI companies of training models on plagiarized, misattributed work — SchmidhuberAI · 2026-09-10
- Rosie Campbell: motives attacks say nothing about whether a claim is true — RosieCampbell · 2026-09-10
- Anthropic researcher: >10% chance AI kills everyone this decade, no alignment plan yet — jammastergirish · 2026-09-10