Keller Jordan: pausing algorithmic progress while compute piles up maximizes AI risk
kellerjordan0 · x · 2026-09-09
AI researcher Keller Jordan outlines a counterintuitive "maximize AI risk" strategy: pause lab-side algorithmic progress while hardware companies keep piling up compute exponentially, and focus on safety to avoid public-spooking accidents. This minimizes accidents over five years and keeps the public calm — but leaves massive hardware in place, so when algorithmic progress resumes, an uncontrolled FOOM becomes possible. Minimizing risk, he argues, means doing roughly the opposite.
Related event: Keller Jordan: Pausing Algorithmic Progress Could Maximize AI Risk(3 posts)→
More from AGI Musings
- FDA's GenAI medical device paper exposes the accountability gap in AI clinical practice — jonc101x · 2026-09-09
- Physician-Style Licensure for Clinical AI? A 14-Post Thread Argues Trust Can Be Manufactured — jonc101x · 2026-09-09
- tszzl: Superintelligence isn't a totally different thing, but pace should be regulated — tszzl · 2026-09-09
- Ethan Mollick: Labs' weird discourse signals everyone is in reaction mode — emollick · 2026-09-09
- Machine civs lack mortality's regularization — 'like training MoE without load balancing' — kellerjordan0 · 2026-09-09
- Mathematicians will 'rage' as AI solves their pet problems, Tao reportedly frustrated — ctjlewis · 2026-09-09