slowlp
← loop
Lesson 2026.09.09 · 7 min read

OpenAI's Navier-Stokes scoop, Meta's Muse, and stolen Claude tokens

Today's AI essentials — 2026-09-09

LOOP

📰 News

  • OpenAI scooped Navier-Stokes — NYU’s Buckmaster and Anthropic’s Alpöge had been grinding on a Millennium Prize problem with Claude and Codex. Rumors leaked; OpenAI spun up ~10,000 agents for 88 hours. Internal model “significantly more capable than GPT-6 Astra.” 300B output tokens — $22.5M at Astra rates. Bubeck allegedly told Buckmaster to drop the Anthropic co-author; when he refused: “Why would you ruin your career?” They’d dumped every draft into Codex. OpenAI “cannot rule out” that product data helped. → The Decoder

  • Meta shipped Muse — A consumer agent that wires into email, calendars, and payments, then checks out via Stripe. Free, plus $20 / $100. Two weeks after an $18B settlement. The product needs more trust than Facebook ever earned. → TechCrunch

  • Hackers are draining Claude tokens — Infostealers grab login sessions and burn Max allowances while you sleep. No itemized usage, so this can run for months. Anthropic refunds; it won’t say how to spot it. → TechCrunch

  • Mistral raised €3B — Valuation past €21B, Europe’s largest tech round ever. Samsung led; Macron called it a “third way in AI.” Models trail Qwen and Kimi. The pitch is sovereign AI, not a European ChatGPT. → TechCrunch

  • Meta killed tokenmaxxing — They scored engineers on AI usage. People burned tokens to climb the leaderboard. Reviews now look at quality and speed. Internal AI spend is heading toward billions in 2026. → The Decoder

  • Publishers are grabbing the Anthropic checks — $3,000 per pirated title from the $1.5B settlement. Authors say houses are claiming books whose rights reverted — and some agents want a cut too. → TechCrunch



▶️ YouTube


💬 Community

COMMENTS