Frontier AI outgrows the convex hull view: harnesses extend inference limits
gerardsans · x · 2026-11-22
Gerard Sans argues that the "convex hull" framing is inaccurate for frontier AI systems today. These models are part of larger systems where harnesses introduce complex algorithms and verification workflows, extending beyond the usual view of inference limits. He made the point while responding to Eric Weinstein and J.D. Lichtman.
More from AGI Musings
- Was Move 37 really out-of-distribution? A case for long-horizon self-play — weballergy · 2026-11-22
- Gary Marcus: without a Lean for biology, AI can't just 'solve biology' — GaryMarcus · 2026-11-22
- Long-Running Agents With Long-Horizon Incentives Could Build Theories Too — weballergy · 2026-11-22
- Can human minds exit their sensory distribution? Rethinking the baseline for AI creativity — weballergy · 2026-11-22
- In the AI age you get rich selling a process, not a product — daniel_mac8 · 2026-11-22
- Claude did three jobs in the Riemann Hypothesis proof: finding, checking, explaining — Prestigious_Bag_5349 · 2026-11-22