On the measure of Intelligence↩︎ 27 Feb 2025 (François Chollet, Nov 2019)

Fundamental paper for AI research. Chollet tries to find a definition of intelligence that is not naive, not something that shifts every time there is a new model that beats humans at some narrow task (like chess, at go, or that writes like humans). He aims for a definition that lets us discuss AGI in a practical way, not the holy grail that solves all the problems on earth, but a concrete goal we can actually work toward. Brutally summarized: For Chollet, intelligence is the ability to generalize quickly beyond what is known; in his words: "skill-acquisition efficiency".