"If recreating ancient Rome is the eval, just declare AGI": dev mocks demo-chasing
vivekhaldar · x · 2026-11-12
Following up on his call for model reviews focused on daily cognitive work, vivekhaldar quips that if recreating a dazzling ancient Rome is the only way to tell model n+1 from model n, we might as well declare AGI and call it a day — a jab at demo-chasing AI evaluations.
Related event: Critics Call for AI Benchmarks to Focus on Everyday Cognitive Tasks(2 posts)→