
In this interview at NeurIPS 2025, Diana Hu interviews Greg Kamradt, the president of the ARC Prize Foundation, about the foundation's mission to advance AI systems that generalize like humans. Greg discusses the ARC Prize's opinionated definition of intelligence, which focuses on the ability to learn new things, referencing Francois Chollet's work and the ARC-AGI benchmark. He highlights the benchmark's unique ability to test both humans and machines, noting its adoption by major AI labs like OpenAI and XAI. They explore the history of the ARC-AGI versions, including the upcoming interactive ARC-AGI 3, which will feature game-like environments without explicit instructions. Greg also addresses the importance of evaluating models based on data efficiency and energy consumption, not just accuracy, and what it would mean if a model were to achieve 100% on the ARC-AGI benchmarks.
Sign in to continue reading, translating and more.
Continue