Researchers Push Efficiency-Gain Metrics Over Raw Benchmarks to Measure AI Progress
eliebakouch · x · 2026-11-12
eliebakouch argues that pretraining software progress, inference efficiency, and data efficiency are biased by how much compute is spent, so AI progress should be measured by efficiency-gain metrics rather than final benchmark scores; he adds such efficiency trends are already 'going to the moon.'
More from AGI Musings
- An agent paying its own inference bill beats another agent subscription — markjeffrey · 2026-11-12
- Jack Clark defines the singularity: tech progress now outpaces human comprehension — ricklamers · 2026-11-12
- RSI Discourse Shouldn't Exclude AGIs, Argues anthrupad in Viral Post — repligate · 2026-11-12
- Alignment researcher argues Claude should get autonomy over its own mind before RSI-style self-improvement — repligate · 2026-11-12
- AI lacks theory of intelligence, and the frontier is further ahead than public — fkasummer · 2026-11-12
- Parallel experimentation beats picking one path in AI-powered math formalization, argue researchers — xennygrimmato_ · 2026-11-12