Скачать или смотреть AI Cracks the ARC Benchmark: Has Artificial General Intelligence Arrived?

For years, the Abstraction and Reasoning Corpus (ARC) benchmark was a major challenge for AI, testing true intelligence. But now, AI systems are solving it, raising questions about real AGI.
This video explores:
How AI models like Gemini 3 and GPT-5.1 are achieving human-level performance on ARC.
Why this doesn't necessarily mean AI is truly intelligent.
The shift towards test-time adaptation and specialized reasoning models.
Data contamination and the challenges of building truly generalizable AI.
What's next for AI benchmarks and the pursuit of AGI with ARC-AGI-3.
We'll discuss François Chollet's perspective on the evolution of AI and the importance of adaptation. Is solving ARC a step towards AGI, or just another optimization problem?