Debate: does superintelligence need teleoperation data, or can sims close the gap?
herbiebradley · x · 2026-10-13
Herbie Bradley, Aashish Reddy and others debated two camps of beliefs about superintelligence, focused on teleoperating general-purpose robots:
- Bradley distinguishes category 1: models as in-domain performers bounded by the long tail of failure data (no teleoperation data → poor performance), vs category 2: taking the "datacenter of geniuses" literally, imagining what von Neumann could do trapped in a datacenter — implying they could teleoperate robots in unseen environments.
- Bradley argues category 2 is wrong: cognitive superset of von Neumann ≠ mastery of specific skills he never learned.
- Reddy counters that the distinction collapses if simulations get good enough — Waymo already has sims, and hard computational-complexity barriers making synthetic data infeasible seem unlikely in most domains.
Core disagreement: is the ceiling set by data long tails, or bypassable by general intelligence plus simulation?
More from AGI Musings
- Super-Worms Warning: AI-Powered Self-Propagating Malware and a Fragmenting Internet — JensHonack · 2026-10-23
- Wolfram Opens Rule 30 Prizes to AIs: $30,000 for Three Open Problems — cephaloform · 2026-10-23
- Mathematicians and Biomedical Researchers React to AI in Strikingly Different Ways — manorlaboratory · 2026-10-23
- Stanford and Databricks use AI agents to reproduce medical findings and test analytical robustness — ChengleiSi · 2026-10-23
- ISMS Fellow Avi Goldfarb: AI disruption raises returns to exploring research directions — avicgoldfarb · 2026-10-23
- signull on AI-found PHP exploits: economic liability is the forcing function legacy software needs — signulll · 2026-10-23