Google DeepMind’s “Aletheia” just solved 6 open research-level math problems. Is this the AGI moment we've been waiting for?

Google DeepMind has unveiled Aletheia, a new AI research agent that autonomously solved six open, research-level mathematical problems, marking a shift from models focused on competition benchmarks like the IMO to those addressing genuine academic frontiers. Unlike its predecessors, Aletheia tackled problems that stump PhD-level researchers, demonstrating an ability for long-horizon reasoning. Its success is underpinned by a novel 'generator-verifier' loop, where the AI essentially argues with itself to produce and rigorously check proofs, ensuring logical soundness. A standout achievement was solving a persistent problem in algebraic topology using two different proof methods, one leveraging Lefschetz numbers.

This capability is part of a broader infrastructure push, with Google investing in a massive 1.4 GW clean energy data center in Minnesota, hinting at the computational demands of such 'Mathematical AGI.' The development suggests a move beyond AI as a tool for known tasks toward AI as an autonomous researcher capable of novel discovery. While not AGI, Aletheia represents a critical step in creating systems that can navigate complex, open-ended problem spaces without human intervention, potentially accelerating progress in fields from pure mathematics to materials science. The strategic investment in energy infrastructure, including iron-air batteries, highlights the scale of compute required for this next phase of AI development.