AI Signal 484
Chollet argues AI intelligence has an optimality bound, with future gains coming from replicability and cloud laws
↗
Illustration only Photo by Albert Stoynov on Unsplash
Martin Fowler's Fragments post surfaces François Chollet's argument that intelligence is a conversion ratio with an optimality bound rather than an unbounded scalar, and Noah Smith's observation that AI value may come from replicability and responsiveness rather than raw intelligence leaps.
For engineers building AI-assisted systems, this framing suggests diminishing returns from chasing model intelligence and greater returns from making AI cheaper, faster, and more replicable. The concept of cloud laws, causal regularities too complex for any individual human to intuit, points to AI finding value in domains where distributed tacit knowledge currently defies reduction.
Written by elseif from the cluster below · every claim links back to a sourceThe three things worth knowing
Chollet frames intelligence as a conversion ratio with an optimality bound, not an unbounded scalar, meaning AI intelligence improvements will eventually become marginal.
Noah Smith observes high AI usage but no massive productivity growth or job losses, suggesting AI's near-term value lies in replicability and responsiveness rather than intelligence gains.
Smith proposes cloud laws, causal regularities too diffuse for humans to understand but exploitable by AI, as a new direction for AI capability beyond traditional intelligence.
THE CLUSTER