Aligning One Model Isn't Aligning Them All: Preparing Society for Superintelligence
gandamu_ml · x · 2026-09-12
Replying to @tszzl's thread on free speech, ML practitioner gandamuml argues that aligning a single model does not align them all — the biggest piece of the puzzle is making everything and everyone (non-AI included) more robust in the face of superintelligence. Failing to prepare adequately, he warns, could end in draconian controls on intelligence itself.
Related event: From Free Speech Norms to Societal Alignment: Debate on AI's Foundations(3 posts)→
More from AGI Musings
- AI KOL: fly-model memes show people want to own models, and local runnability is why — beffjezos · 2026-09-12
- Paras Chopra: AI's real crisis is the deflation of human motivation — paraschopra · 2026-09-12
- Jeff Bezos on AI doomers: 'Those people are wrong' — AI will elevate everyone — beffjezos · 2026-09-12
- IISc professor warns AI is commoditizing intellectual labor — what's university for? — datawithsuman · 2026-09-12
- Frontier lab staff mostly reject p(doom)=0.1, but selection pushes doomers out, researchers argue — nabla_theta · 2026-09-12
- Dev predicts RL on org management will make AI-run companies work: 'it's around the corner' — willcb · 2026-09-12