Is AGI Here? Not Even Close, New AI Benchmark Suggests

What: A new AI benchmark, ARC-AGI-3, suggests that Artificial General Intelligence (AGI) is not yet achieved, contradicting claims made by Jensen Huang.
Why: The benchmark's results, with Gemini scoring 0.37% and GPT-5.4 getting 0.26%, are significantly lower than the human average of **100%**.
Signal: The disparity between human and AI performance indicates that AGI remains an elusive goal, with substantial gaps in cognitive abilities and problem-solving capabilities.
Target: Researchers and developers will need to focus on improving AI systems to bridge the gap with human performance, potentially requiring significant advancements in areas like reasoning, learning, and common sense.
Risk: Overhyping AGI capabilities, as seen in Jensen Huang's declaration, can lead to unrealistic expectations and undermine trust in the field, making it crucial to maintain a cautious and evidence-based approach.