Draft v3 · · What changed
François Chollet
Chollet is the only major figure who built the benchmark used to evaluate AGI claims AND the company commercially betting on the alternative architecture; the concentration is structurally unique and currently held in equilibrium by the foundation's transparency, Ndea's nonresult, and the labs' continued willingness to report.
Positions
The test every frontier lab reports
ARC has become a de facto governance tool for the AGI claim because all four major frontier labs have reported scores on it, but the arrangement runs through one foundation and its co-chair, and a redesign process with…
Scaling isn't enough
The diagnosis that pure deep learning would not produce general intelligence has aged well; the prediction that scaling would plateau has not, and Ndea's existence is the bet that the diagnosis matters more than the…
The ARC residual gap
The standard criticism of ARC-AGI as a moving target gets the design backward; the benchmark family is meant to track the residual gap between human and machine learning, but its self-redesigning structure carries its…
Moving-target discipline
ARC-AGI changes what kind of test it is rather than raising the bar on the old one. A version that raised the passing score without changing the task would prove the goalpost charge. None has.
·Issue 1 · The Scaling Bet
deslop.media editorial. Issue 1, Draft v3, first published October 10, 2026. Reporting current through October 2026. Use the flags on this page to report an error or a disagreement. Changes to the issue are listed on the changes page.
Part of Issue 1 · The Scaling Bet