Google AI Solves Decades-Old Math Problems; AGI Still Far

Google DeepMind's AlphaProof Nexus AI system has solved several long-standing mathematical problems, including some that remained unsolved for over 50 years. This includes nine problems related to mathematician Paul Erdős and a 15-year-old problem in algebraic geometry. The system combines AI reasoning with a formal verification tool called Lean. It generates step-by-step proofs, which are then automatically checked for accuracy. This method ensures every logical step is verified and helps reject incorrect assumptions. The cost for solving each problem was only a few hundred dollars in computing power. Unlike some other AI developments that rely on human verification, DeepMind's system uses automated verification to reduce errors, like "hallucinations" where AI can produce seemingly correct but flawed results. However, despite these breakthroughs, DeepMind CEO Demis Hassabis states that current AI systems are still far from achieving artificial general intelligence, or AGI. He explains that solving specific problems doesn't mean AI can match human intelligence across all areas, which would require creativity and independent reasoning. This development shows AI's growing role as a powerful assistant in scientific research, helping to explore complex problems and check proofs more efficiently.