'Models beat 90% of humans at white-collar tasks but stall on untestable meta work'
Lux_Stella_ · x · 2026-11-06
X user LuxStella argues that not running agents 10-12 hours a day shows in these debates, and offers two spoilers: models are now better than 90% of humans at any white-collar work task, yet have barely moved on untestable meta tasks. teodorio replies that this is structural — models improve at what you can measure, and you may not know how to measure the rest.
Related event: AI debate: models soar on measurable tasks but stall on unmeasurable ones(2 posts)→