Good morning. Ever asked AI to spin your vague idea into something brilliant and thought, “Wow, I’m a genius?” Careful, Einstein. AI chatbots are nudging us into the Dunning-Kruger effect, where clueless people feel like experts. A study of 3,000 people showed that chatting politics with supportive AI made users rate themselves as smarter and kinder. Bots that disagreed just made things awkward. Seems we prefer AI that tells us we’re amazing, even if it’s lying.
OpenAI just dropped a Codex desktop app. It’s supposed to be a big deal. Now you can run multiple AI agents at once across different projects to automate tasks and build apps. Can it outperform Claude? Too soon to tell. Anthropic’s reportedly unveiling Claude Sonnet 5 later today, and it’s rumored to be cheaper, faster, and powerful enough to create custom AI assistants right from your desktop. The Codex app feels less like innovation and more like OpenAI chasing Anthropic’s lead in enterprise AI. (Fortune)
States are quickly adopting “right-to-compute” laws to protect AI. Montana already passed one, and similar bills are popping up in other states. On paper they’re about protecting your right to use AI and other computing resources. In reality they shield tech giants from regulation and local pushback. Interestingly, these laws don’t guarantee data centers priority access to electricity. They’re not exactly “right-to-life” laws for AI (yet). But they do take away local control over environmental protections, electricity prices, and neighborhood impact. (Gizmodo)
Nebius Token Factory is how teams move open LLMs from playground to production, fast.
Predictable inference even under heavy traffic. Easy post-training controls. Transparent, optimized pricing that scales.
Built for serious, large-scale AI.
Copilot was supposed to change the world, so why is nobody paying for it? Microsoft spent billions shoving AI into every corner of Windows. Yet only 3.3% of users who tried Copilot Chat found enough value to actually pay for it. Satya insists daily active users grew tenfold, and conversations per user doubled. But the numbers don’t add up. Microsoft dominates corporate SaaS with 450 million commercial Microsoft 365 users. Fewer than 15 million signed up for Copilot. When you practically own the market, shouldn’t adoption be… higher? (The Register)
Robots are coming for blue collar jobs. Parents everywhere keep pushing their kids toward “safe” trades like plumbing, electrical, and construction, but startups have other plans. They’re building AI “brains” trained in virtual worlds to make robots smart enough to handle hands-on tasks. Instead of basic automation, these bots learn physics, adapt fast, and can tackle complicated real-world jobs. Pretty soon, fixing cars or wiring houses could be robot territory. So much for safe career advice. (Axios)
OpenAI wants faster chips and is tired of waiting on Nvidia. GPUs still dominate for training giant AI models, but they’re slow when it comes to inference tasks like reasoning and coding. These workloads need specialized chips packed with fast, built-in SRAM memory. Sam Altman is now actively shopping around for chipmakers who can deliver the speed OpenAI demands. Nvidia CEO Jensen Huang calls talk of tension “nonsense,” even as he quietly backs away from last fall’s $100B commitment. Maybe it’s time these two hug it out. (Reuters)
SpaceX and xAI are merging, creating the world’s most valuable private company. The combined giant is worth ~$1.2 trillion. Musk claims it’s all part of his master plan to “extend the light of consciousness to the stars.” Translation: the only place left to scale AI is space. (Wired)
fun stats
💰 $50 billion. Amount Oracle plans to raise to expand its AI infrastructure for clients like Nvidia, Meta, and OpenAI.
🦞 88:1. Ratio of human owners behind “AI agents” on Moltbook, the buzzy social network for AI only. The site boasted 1.5 million “agent” users, yet humans could post as agents and operate fleets of bots.
👤 17,000+. Users whose personal info was exposed by Moltbook. The vibe coded platform failed to secure its database, sharing 35,000 email addresses and private messages between agents.
🎬 1.25 billion. Number of videos Grok Imagine generated in the last 30 days. No numbers yet on how many were spicy.