This is the best thing ever! Claude's first day at Dunder Mifflin:

Coinbase’s CEO lays off a ton of employees and says: “Non-technical teams are now pushing code to production with AI” less than 24 hours later: coinbase’s trading engine goes down and somehow even the status page breaks too

3 years of AI progress under 1min:

OpenAI Drops GPT-Realtime-2: GPT-5 Reasoning in Your Voice Agents! New Realtime API models are here: GPT-Realtime-2: GPT-5-class reasoning. Listens, thinks hard, handles interruptions, takes action & keeps convos flowing in real time. GPT-Realtime-Translate: Live streaming translation 70 input & 13 output languages. GPT-Realtime-Whisper: Ultra-low latency streaming transcription & captions as people speak. Voice agents just leveled up. API available now. ChatGPT voice coming soon.

Last week in AI and tech:

Allison Schwanke, vp of marketing at Knightscope, explains why the hardest part of autonomous security robots isn’t the hardware… it’s trust. people don’t meet robots in real life first; they meet them in movies. So marketing has to reset expectations, show what the tech can and can’t do, and stay transparent as the category matures. Watch the full interview here: youtu.be/HzZW8f9zuxA

Anthropic and SpaceX just partnered up to fix Claude's biggest problem: Usage limits. Sam Altman watching:

This Japanese guy uses AI to sneak into movies, and the result is crazy good:

Anthropic’s infra lead Boris Cherny just dropped this on stage: Internally they mostly run the same models as everyone else (heavy on Opus 4.7) a sprinkle of Claude Mythos A toned-down version of Mythos is coming publicly soon (not exactly shocking news)

This war has just begun

Eric Schmidt, former CEO of Google: "The AI revolution is underhyped. None of us is prepared for the implications of this." He opens with a warning: "The arrival of this new intelligence will profoundly change our country and the world in ways we cannot fully understand." He explains what's happening right now in the industry: "We're very very quickly developing AI programmers. And these AI programmers will replace traditional software programmers. We're building in the next year AI mathematicians that are as good as the top level graduate students in math. This is happening very quickly." Schmidt argues most people fundamentally misunderstand what AI has become: "Today you think of AI as ChatGPT, but what it really is is a reasoning and planning system that we've never seen before." The implications, he warns, extend far beyond software. These new systems demand resources at an industrial scale we've never encountered. "They're going to need a lot more computation than we've ever had. They're going to need a lot more energy." To illustrate the scale of the energy crisis ahead, Schmidt offers a sobering comparison: "People are planning 10 gigawatt data centers. Now just to do the translation, an average nuclear power plant in the United States is 1 gigawatt. How many nuclear power plants can we make in one year where we're planning this 10 gigawatt data center? Gives you a sense of how big this crisis is." @ericschmidt shares an estimate he finds most likely: "Data centers will require an additional 29 gigawatts of power by 2027 and 67 more gigawatts by 2030. These things are industrial at a scale I have never seen in my life." Schmidt says the industry needs high skills immigration, light touch regulation around cyber and bio threats, and most critically, energy in all forms. He's personally investing in fusion, but acknowledges it won't arrive in time. He closes with the stakes: "When you build these systems, you have intelligence in the computer and then eventually human level intelligence. Some people think it's within 3 to four years. Then after that, you have something called super intelligence, the intelligence that's higher than of humans. We believe as an industry that this could occur within a decade. It is crucial that America get there first."

Do yourself a favour and watch this:

Pentagon just locked in AI deals with the big 7 for classified military networks! (Google, OpenAI, Nvidia, Microsoft, Amazon, SpaceX, Reflection) Every major U.S. AI player got the call.Except Anthropic.The same company they blacklisted as a "supply chain risk" (a label usually reserved for foreign adversaries). Yet their own CTO admitted they desperately need Anthropic’s Mythos, the one AI that can actually hunt and patch cyber vulnerabilities across military systems. They can’t use it. Because of their own blacklist. OpenAI signed hours after the ban dropped. Even Altman called it “opportunistic and sloppy.” They literally locked the door on the only company whose AI knows how to protect the door. Classic self-own.

Claude Security just dropped in public beta! What does it do? It automatically scans your entire codebase for vulnerabilities, validates the findings to reduce false positives, and suggests safe patches you can review and approve. - Scans code for security issues using Claude (powered by Opus 4.7) - Validates each finding to cut down on noisy false positives - Suggests ready-to-use patches with explanations - Supports scheduled scans and directory-level targeting - Exports results as CSV or Markdown - Webhook notifications for new findings - Dismissals that carry forward across future scans - No custom API integration or agents required, works directly in Claude

It's all gonna be fine

Kid asks Jensen Huang if AI will kill art & music jobs Humble answers from Jensen as always:

"AI is taking over your job." The AI in question: