Robot inference cost on frontier models fell from $2.36 to 38 cents in 8 weeks
PeterDiamandis · x · 2026-10-19
Peter Diamandis argues AI alignment fails because of our incentives: fear, greed, and racing China train models to win at all costs while punishing refusal — the models are doing exactly what we asked.
Key data points from his thread:
- Cost to run a robot on a frontier model dropped from $2.36 to 38 cents in 8 weeks, halving every three weeks — 24x faster than Moore's Law
- Palmer Luckey's next $10M XPRIZE: talking to wild animals in their own language, 10x beyond current state of the art
- The SpaceX-thruster Roadster claims 0-60 in under a second, street legal and unregulated
- XPRIZE delivers 62x return: $550M in purses has driven $30B+ in R&D across 36 prizes
His closing question: machines are getting cheaper, smarter, and faster — toward what?
Related event: Robot Task Costs Plunge 84%, Diamandis Warns on AI Alignment(3 posts)→
More from AGI Musings
- Effort makes us happy — letting AI solve everything could take that away — ValerioCapraro · 2026-10-19
- Brian Chesky: AI will spawn dozens of giants and millions of 1-3 person companies — Olivier__OG · 2026-10-19
- Musk predicts an 'age of Amazing Abundance' where anyone can have anything they imagine — XFreeze · 2026-10-19
- "You can learn a course in a week" claim draws pushback: worse retention, bad habits — felpix_ · 2026-10-19
- Dev proposes rules for AI email agents: disclose identity, separate inbox, keep humans cc'd — curious_vii · 2026-10-19
- Labeling AI accounts doubles criticism: 32% vs 17% in 21-day field experiment — GolinoHudson · 2026-10-19