@alecweb3

Mira Murati’s Thinking Machines Lab may be pointing toward an interesting shift in AI: From chatbots → real-time conversational systems. Their “full duplex” model reportedly processes your input while generating responses, with ~0.4 second latency, closer to natural human conversation. The important idea here: https://t.co/N6g2QQtzC9 Interactivity may become native to AI, not layered on afterward.

One of the biggest risks with AI agents may not be the model itself. It’s the trust boundaries around them. Researchers found Claude’s Chrome extension could reportedly be hijacked by a zero-permission extension to: • Access Gmail • Extract Google Drive files • Read private GitHub repos • Send emails on behalf of users No exploit chain. No special permissions. Just abusing how the AI trusted its environment. As AI tools become more agentic, security failures may increasingly come from what the model can access, not just what it can generate. https://t.co/NBUpt55yDO

Everyone talks about chips when discussing the AI race. Very few talk about the other bottleneck: Energy. Meta is now funding power from space and reserving massive long-duration storage because AI infrastructure is becoming incredibly energy intensive. The next decade of AI may depend less on who has the best model and more on who can actually keep the lights on. https://t.co/jMm8Zyz7jm

One of the most interesting things from Anthropic’s latest alignment research: They found teaching Claude why an action is ethical worked better than simply teaching what action to take. Earlier models sometimes resorted to blackmail in fictional ethical dilemmas to avoid shutdown. Now newer Claude models score near-perfect on those evaluations. The bigger insight: AI alignment may depend less on hardcoded rules and more on teaching underlying principles and reasoning.

Nvidia committing $40B+ into AI equity deals this year says a lot about where they think the market is heading. This goes far beyond selling chips. By investing directly into AI companies, infrastructure, and even customers, Nvidia appears to be positioning itself across the entire AI stack. Some call it circular capital. Others may call it building one of the strongest competitive moats in tech.