Robotics AI can't copy the LLM playbook: 5% error rates kill the product
carlosdponx · x · 2026-09-19
The author argues robotics AI differs fundamentally from LLMs: chat and coding can tolerate 30% success rates because retries in a loop still extract value, but nobody accepts a robot that throws your smartphone in the trash 5% of the time while "tidying up" — the lazy default of not cleaning beats that risk. His conclusion: nail specific tasks first and expand from there, unlike the broad-bet approach that worked for software AI.
More from AGI Musings
- Minerva vs Casado: air-gapped model coordination still infeasible at 0.002 bps covert bandwidth — BlancheMinerva · 2026-09-19
- Economists Ben Moll and Alex Imas: don't expect double-digit GDP growth from AI within 10-15 years — binarybits · 2026-09-19
- AI Growth Bet: Per-Capita GDP Growing 15% Above Peak Has Just 0.9% Historical Odds — RishiBommasani · 2026-09-19
- AI Makes Old Expertise Cheap—and Creates More Expert Work Finishing What It Starts — every · 2026-09-19
- "Think AI Safety Is Bad? Wait Until the Public Writes AI Policy" — NathanpmYoung · 2026-09-19
- Many New Projects Are Being Built on Chinese Open-Weight Models — signulll · 2026-09-19