Mollick calls Fable 5.2 almost general intelligence, with jagged edges
Ethan Mollick, professor of entrepreneurship and author of One Useful Thing
Mollick argued that Fable 5.2 comes close to his working definition of artificial general intelligence, but remains uneven in taste and judgement rather than in simple errors.
Source published · In the Daily Brief of 10 November 2026
Mollick defined artificial general intelligence narrowly, as an AI as good as most human experts at most intellectual tasks done on a computer, excluding physical work and dealings with people. By that standard he judged Fable 5.2 to be almost there, a state he calls jagged AGI: expert-level across many domains, but still well behind people in places that are now harder to describe than hallucinations.
His tests ran across domains. From one prompt the model turned a small USB keyboard into a console with six games, working out the hardware from the code of its control apps. Asked to launch a business, it screened 22 ideas and read public agendas from 201 local governments to build a plan, giving its own estimate of a 20–40% chance of success; Mollick stopped before it went live. It also wrote a research paper on company filing fees that he placed ahead of a second-year doctoral student, though not publishable, and it did not carry out the pre-registration it was asked for.
The gaps, in his account, were of taste: a first course website he called joyless, readings of poems that ignored the author’s own commentary. The model fixed each in minutes once told but noticed none itself. In a thread on X he added that about 25,000 agents building more than 450 linked 3D worlds developed their own bureaucracy, which eventually slowed them down.