Economists in the top 10% of AI use see no step change from GPT 5.6 to 6

aniketapanjwani · x · 2026-09-08

aniketapanjwani polled a group of economists — all in the 90th+ percentile of AI use — and none noticed the step change from GPT 5.6 to 6 (Astra).

Quoting nicbstme's analysis: everyday tasks are already saturated — AI does babysitter emails and bike-tire PSI at effectively 100%, and there's no 120%. The capability jump is only visible at the frontier of coding, math, or science.

His deeper take: the bottleneck for economics research isn't what models are RL'ed on — either econ research isn't economically valuable, or labs won't pay the high price for those RL environments versus, say, financial-analyst ones.

Original post →

More from Models

Models channel →