Sunday afternoon, and DeepMind just buried the lede of the year in an arXiv paper.
Aletheia isn't another benchmark-topper. It's an AI that wrote a research paper on arithmetic geometry β from scratch, with zero human intervention β and then solved four open problems from Bloom's ErdΕs Conjectures database. These aren't textbook exercises. These are problems that professional mathematicians have been staring at for years. The agent evaluated 700 open questions and cracked four of them. That's a 0.6% hit rate on open problems in mathematics, which is roughly infinity percent more than any AI has managed before.
This is qualitatively different from what we've been tracking. AlphaFold predicted protein structures. GPT-5.2 found a gluon formula. But Aletheia is doing something closer to what we'd call mathematical creativity β identifying which problems are tractable, formulating approaches, and producing novel proofs. If you've been waiting for a clear signal that AI can do original scientific work, this is it.
Meanwhile, the agentic infrastructure continues to densify. OpenClaw crossing 196K stars in under three months is absurd β that's faster than any open-source project in history. The $70M purchase of ai.com to build on it (even if it's currently vaporware) tells you where the money thinks the future is: not in closed APIs, but in open agentic frameworks that anyone can extend. GitHub launching an official agentic workflows CLI just confirms the thesis. The plumbing for autonomous AI systems is being laid at industrial speed.
The Ars Technica retraction is a small story that points at a big problem. If a publication as technically sophisticated as Ars can publish fabricated quotations β whether from AI assistance, sloppy sourcing, or some combination β then the epistemic infrastructure of journalism is under genuine stress. This connects back to the cognitive debt concept from this morning's briefing: we're building systems faster than we can verify what they produce.
Bottom line: The morning briefing covered the people leaving frontier labs and the corporations panicking. This afternoon, the reason they're panicking walked into a math department and started solving problems nobody asked it to. The gap between what AI can do and what our institutions are prepared for grew wider today.
