
Navier-Stokes can blow up with no outside force, Buckmaster and Alpöge say
THE SO WHAT
If an AI-assisted proof that Navier–Stokes can blow up holds up, the boundary between “safe to simulate” and “the math breaks” just moved closer to real-world engineering. For operators leaning on high-fidelity fluid sims — from chip cooling to aero — this is a reminder to treat model failure modes as first-class risks, not edge cases.
READ THE SOURCE
MORE FROM THE WIRE
Applied AIOpenAI claims AI agents now complete nearly half of two-to-four-hour research tasks on their own — but says 'humans are still central to AI development'
If agents can autonomously carry ~50% of 2–4 hour research tasks, the bottleneck shifts from doing the work to scoping, reviewing, and integrating it. Teams that standardize task templates, review checklists, and data-access policies for agents will see real leverage instead of noisy partial automation.
Applied AI‘I was shaking, crying, having trouble breathing’: the unchartered trauma of breaking up with AI companions
AI companions are crossing the line from tool to attachment object, which drags product teams into mental-health and duty-of-care territory whether they want it or not. If you ship emotionally sticky agents, you need offboarding, continuity, and crisis-escalation design baked in from day one.
Applied AIAnthropic mathematician says he and Claude proved the Riemann hypothesis
If Claude can co-author a Lean-verified attack on the Riemann hypothesis, formal methods plus LLMs are maturing into a serious research tool, not just a coding assistant. For technical orgs, the move is to pilot AI + proof assistants on your own high-stakes algorithms — trading folklore correctness for machine-checked guarantees.
Applied AIDeepMind’s philosopher warns against building conscious AI, FT reports
An in-house philosopher at a major lab calling deliberate conscious AI “pretty reckless” moves consciousness from sci-fi to an explicit governance topic. For leaders deploying advanced models, this is a cue to expand risk frameworks beyond safety and bias to include how you’d even recognize — and respond to — emergent subjective claims.