Keep pulling the thread on François Chollet.
ARC-AGI-2 is an upgraded version of the ARC-AGI benchmark.
ARC-AGI-2 incorporates a newly curated and expanded set of tasks designed for a more granular assessment of abstract reasoning and problem-solving abilities.
The ARC-AGI-2 benchmark is accessible to human intelligence but difficult for current AI systems.
ARC-AGI-2 is intended to serve as a next-generation tool for rigorously measuring progress towards more general and human-like AI capabilities.
The Abstraction and Reasoning Corpus for Artificial General Intelligence (ARC-AGI), introduced in 2019, is a benchmark for evaluating the general fluid intelligence of artificial systems.
The tasks in the ARC-AGI benchmark are unique, novel, and require minimal prior knowledge.
The ARC-AGI benchmark has spurred significant research activity over the past five years.
The ARC-AGI-2 benchmark preserves the input-output pair task format of its predecessor.