March 24, 2025:
New AGI Test Challenges AI Models Intelligence - The Arc Prize Foundation, co-founded by AI researcher Francois Chollet, has introduced ARC-AGI-2, a challenging intelligence test for AI models. Major AI systems, such as OpenAI's o1-pro, scored only 1% to 1.3%, significantly below the human benchmark of 60%. ARC-AGI-2 is designed to assess adaptability and efficiency without relying on brute force, focusing on pattern recognition beyond data training.
This new test improves upon the previous ARC-AGI-1 by emphasizing cost-efficient problem-solving. The Arc Prize 2025 contest challenges participants to achieve 85% accuracy on ARC-AGI-2 while maintaining a minimal cost of $0.42 per task.
