François Chollet, a renowned AI researcher and former Google engineer, is co-founding a nonprofit organization to develop benchmarks that can assess AI systems for "human-level" intelligence. The ARC Prize Foundation, led by Greg Kamradt, an ex-Salesforce engineering director, aims to create a standardized way to measure AI's ability to acquire new skills and adapt to new problems, a crucial step towards achieving artificial general intelligence (AGI).

The nonprofit builds upon Chollet's earlier work on ARC-AGI, a test designed to evaluate an AI system's ability to efficiently acquire new skills outside the data it was trained on. Introduced in 2019, ARC-AGI consists of puzzle-like problems that require an AI to generate the correct "answer" grid from a collection of different-colored squares. The problems are designed to force an AI to adapt to new problems it hasn't seen before, mimicking the way humans approach complex tasks.

Chollet's vision for the ARC Prize Foundation is to inspire progress towards AGI by promoting the gap in basic human capability. He believes that current AI systems, despite their impressive performance in specific domains, still lack the ability to generalize and adapt to new situations like humans do. The foundation's benchmarks will focus on shrinking this gap towards zero, enabling AI systems to perform most tasks humans can.

Last year, Chollet and Zapier co-founder Mike Knoop launched a competition to build an AI capable of besting ARC-AGI. OpenAI's unreleased o3 model was the first to achieve a qualifying score, but only with an extraordinary amount of computing power. However, Chollet has emphasized that ARC-AGI has flaws and that o3 does not possess human-level intelligence. He has also hinted that the upcoming successor to ARC-AGI will pose a significant challenge to o3, potentially reducing its score to under 30% even at high compute.

The ARC Prize Foundation plans to launch a second-generation ARC-AGI benchmark this year, accompanied by a new competition. The nonprofit will also embark on designing the third edition of ARC-AGI, which will likely address some of the criticisms Chollet has faced for overselling ARC-AGI as a benchmark towards reaching AGI. The definition of AGI itself is currently being hotly contested, with some arguing that AGI has already been achieved if one defines it as AI "better than most humans at most tasks."

Interestingly, OpenAI CEO Sam Altman has expressed interest in partnering with the ARC-AGI team to build future benchmarks. While Chollet did not provide an update on possible partnerships in his announcement, the collaboration could have significant implications for the development of AGI. As the AI research community continues to grapple with the complexities of AGI, the ARC Prize Foundation's work will likely play a crucial role in shaping the direction of AI research and development.

In conclusion, the ARC Prize Foundation marks an important step towards creating a standardized way to measure AI's progress towards human-level intelligence. As the AI landscape continues to evolve, Chollet's nonprofit will provide a crucial north star for researchers and developers working towards the goal of achieving AGI.