AI is finally doing the math — and the math checks out

AI is finally doing the math — and the math checks out

AI now produces genuinely new mathematics. In January 2026, GPT-5.2 Pro generated original proofs for previously-open Erdős problems #397, #728, and #729, formalized them in Lean, and got them accepted by Terence Tao, who called a related result the "most unambiguous" case yet of AI solving an open problem. Separately, DeepMind's AlphaEvolve beat a 56-year-old matrix-multiplication record and found new optima on roughly 20% of the open problems it was pointed at. The maximalist case for AI has now been written down in detail — though the source is careful to note the biggest claims remain speculative and are drawing serious criticism.

Gobble's Take: When AI hands back proofs instead of vibes, the question stops being "can it think?" and starts being "how much do we let it sign?" Source: Perplexity Search

AI governance was never just a triangle — and the shape is shifting again

A conference paper presented June 23, 2026 at the Taiwan Research Hub Joint Conference at the University of Nottingham maps out the familiar three-model frame: the United States backing private-sector-led innovation with limited federal regulation, China running a party-state-centered model, and Europe pushing a rights-based framework built on privacy, democratic accountability, and the rule of law. The paper's argument is that AI has broken this frame — because AI governance is now inseparable from industrial strategy, geopolitical rivalry, semiconductor supply chains, and compute.

Gobble's Take: Chips, sovereignty, and compliance walked into a rulebook. Nobody agreed on the cover page. Source: Perplexity Search

Telecoms are hauling the AI economy's freight and billing for the postage

Global networks will route 3 quadrillion AI tokens in the next 30 days, producing 77 exabytes of traffic. The world's top operators will book roughly $1 billion in direct AI revenue across an entire 12-month fiscal year — 0.04% of the $2.52 trillion AI economy in 2026. Nokia Bell Labs projects AI traffic will reach 1,088 exabytes per month by 2033.

Gobble's Take: Telecoms built the highway, paved it, and are now watching trillion-dollar trucks blow past the toll booth. Source: Perplexity Search

In Case You Missed It

Yesterday's top stories:

• EU draws a hard line on high-risk AI
• Adobe's AI assistants stop helping and start running the show
• Wearable Devices wants to be the "intent layer" between you and everything else