Ethan Mollick

Is the @METR_Evals long horizons measure effectively saturated? No updates to the most famous graph of AI progress since May, but other published work suggests you can effectively get 18+ weeks of work out of Fable (and I assume GPT-6) in harnesses.