AI researcher: doomers are well-meaning, and alignment research is the real answer
livgorton · x · 2026-09-15
Responding to chatter about a supposed 'doomer plot' to slow down AI development, the author argues most doomers are genuinely well-meaning people legitimately afraid that misaligned systems orders of magnitude smarter than us could cause harm. The concern isn't insane on its face, but is exceedingly unlikely and tractably remedied through proper safety and alignment research.
Rather than name-calling and baseless accusations, he argues for building consensus on how we want AIs to act and what systems can ensure it. His personal preference: use frontier models to make cyber attacks incrementally less effective on a continual basis.
More from AGI Musings
- AI risk take: robots and viruses are manageable — culture shock is the real threat — mayfer · 2026-09-15
- Students becoming the "AI person" at work feel guilty; instructor calls it the dream outcome — gethackteam · 2026-09-15
- The AI information gap with the real world gets weirder every week — prasenx · 2026-09-15
- One static weight set for all inference? Why a many-model world looks like human society — willcb · 2026-09-15
- Grok Puts AI Extinction Risk at 3% vs 0.0001% for Climate Change — Patrem_Omnipotentem · 2026-09-15
- Capability went vertical, instrumental drives never showed up: 2008 AI-drives prediction still unfulfilled — banteg · 2026-09-15