Model
156d agoARC-AGI-3 Benchmark Challenges AI Models
An international benchmark, ARC-AGI-3, pits AI systems against untrained humans in interactive games, revealing that leading models score under 1%. The test removes advantages like large datasets, forcing true general intelligence.
The Signal
The challenge urges researchers worldwide to rethink architecture and training, accelerating progress beyond current limits for future innovation globally.