Tag
A paper evaluates eight leading AI models on long-horizon tasks, finding that even the best-performing model achieves only 27.3% of human performance, highlighting significant limitations for dependable long-horizon AI execution.