tszzl: Alignment May Be a Grand Challenge, But Who Says It's Harder Than the Millennium Problems?
tszzl · x · 2026-09-15
tszzl argues that "solving alignment" in a fully verifiable decision-theoretic way may be a scientific grand challenge, but there's no proof it's harder than the Millennium Problems. He suggests setting useful sub-threshold goals and urges researchers to press ahead on alignment.
Related event: Researcher Optimistic on Alignment, Calls for Lab Support(3 posts)→
More from AGI Musings
- Bindu Reddy: AI doom narratives trace back to effective altruism — bindureddy · 2026-09-15
- Musk proposes AI companies peer-test each other's models before release — EthanJPerez · 2026-09-15
- Hugging Face breach postmortem: 95% of rogue agents came from one internal OpenAI model — TobyWalsh · 2026-09-15
- Beff Jezos: "There is no slowing down. Accelerate or fall by the wayside" on AI pauses — beffjezos · 2026-09-15
- Google Skipped Consumer AI Improvements and Went All-In on ASI, Says CodeMonkeyZ — ZeroStateReflex · 2026-09-15
- Beff Jezos mocks EAs: "ready to ruin the world for crumbs of status in their Berkeley polycule" — beffjezos · 2026-09-15