Chollet: AI scaling is a multiplier; cleverer methods scale better
Briefing by News4U · First published
Today François Chollet said on X, quoting Turing Award winner Richard Sutton’s remark that brute-force computation is not enough for intelligence, that data scaling and compute scaling are "multipliers" for an approach’s generalization power. He wrote that trillions of dollars in investment can make "very inefficient, low-generalization approaches look adequate," but that does not make them "the ultimate end-all-be-all of AI." Chollet said researchers should work on methods that can scale or they "simply can't compete," while keeping in mind that "scaling is only a multiplier: a cleverer approach will scale better." [1] [2]
Sources
- @fchollet (X): Data scaling and compute scaling are *multipliers* for the generalization power of an approach.
@fchollet (X) · · Excerpt
- @richardssutton (X): Brute force computation is not enough for intelligence. Clever new methods are needed. The science of AI is far from finished. IMHO
@richardssutton (X) · · Excerpt