This week, a researcher who literally wrote the code the AI industry uses to train models declared that AGI will never happen. GPUs peaked in 2018, he argued. Scaling is over. We have “maybe one, maybe two more years” left.
The same week, a humanoid robot started sorting packages in a factory. Eighteen months from stiff walking to dynamic balance and coordinated motion. UBS projects 2 million workplace units by 2035.
Here’s the thing: Dettmers can be completely right about hardware limits—and it won’t matter for the next decade. The capability that’s already baked in is disruptive enough to transform every industry. The scaling debate is a distraction from what’s actually deploying.
The discourse zigzags. Deployments don’t. OpenAI shipped GPT-5.2 in a “code red” scramble—weeks, not months, after 5.1—because Gemini 3 threatened their benchmark lead. The White House signed an executive order to preempt state AI regulation because competitiveness narratives drive policy. AI agents discovered zero-day vulnerabilities in live DeFi contracts at break-even API costs. Andrej Karpathy told everyone to stop anthropomorphizing models and start treating them as simulators.
Chaos if you’re watching the takes. Coherent if you’re watching the incentives.
Top Stories this week:
OpenAI shipped GPT-5.2 with 400K token context after internal “code red” over Gemini 3 benchmarks
Trump signed an executive order creating an AI Litigation Task Force to challenge state AI laws
Tim Dettmers argued GPUs peaked in 2018 and AGI “will not happen” due to physical constraints
Anthropic agents exploited smart contracts for $4.6M in simulated theft and found live zero-days
Andrej Karpathy’s “LLMs as simulators” thread reframed prompting strategy for 27,000+ builders
Elon Musk proposed orbital AI compute via solar-powered satellites and lunar factories
DeepSeek reportedly using smuggled Nvidia Blackwell chips despite export bans
Humanoid robots shifted from lab demos to factory deployment, costs dropping toward $13K by 2030
Plus, this week’s prompt: your personal news interpreter.
Eight stories. Maybe two actually matter for your work. The prompt below has all the context from this week’s coverage baked in—just tell it your role and what you’re trying to figure out. It’ll cut through the noise: which stories affect your decisions, what they mean for your specific situation, what to do differently, and what you can safely ignore. Not a summary. A filter calibrated to you.
Listen to this episode with a 7-day free trial
Subscribe to Nate’s Substack to listen to this post and get 7 days of free access to the full post archives.
