Mistral AI

The Leverage SF Build Night; AI, Nvidia, LinkedIn Updates

This Substack newsletter announces The Leverage Launch, a one-night build event in San Francisco sponsored by Cursor, and covers several technology and AI developments: OpenAI’s new model (codename Astra) reportedly solved open mathematics problems using about $2,000 in API tokens; Ilya Sutskever’s Safe Superintelligence disclosed a partnership with Nvidia including a reported $5 billion investment; Nvidia’s investment strategy across AI labs and neoclouds (CoreWeave, Nebius, etc.) is highlighted; LinkedIn introduced a “Seems like AI slop” button and removed an AI post-enhancement feature amid findings that ~41% of long-form posts may be AI-generated. The newsletter also mentions cultural notes and consumer product endorsements.

Last Week in AI: Open Models, Robots, Markets

This newsletter summarizes a week of AI developments spanning policy, open models, robotics, and markets. Nvidia-backed signatories published a letter defending open-weight models while Moonshot AI released Kimi K3 weights, signaling open models' growing competitiveness. Google DeepMind released Gemini Robotics 2, extending model capabilities into whole-body robot control. Financial stress surfaced when the AI-focused hedge fund Situational Awareness liquidated public equities to Citadel after concentrated losses, and multiple large technology companies disclosed substantial data-center lease commitments. The update also highlights technical papers (e.g., Chain-of-Evidence / ScientistOne, Shieldstral) and a string of corporate moves: acquisitions, partnerships, funding rounds, and model releases that together emphasize the industry's shift from proof-of-concept to questions of distribution, embodiment, ownership, and durable monetization.

Microsoft openly competes with OpenAI, Anthropic

Microsoft signalled a more explicit competitive posture toward OpenAI and Anthropic during its fiscal-quarter earnings call, arguing enterprises should use multiple models and keep the agentic "harness" separate from models. The company reported a blockbuster quarter — $90 billion in revenue and $35.8 billion in net income, and $331.8 billion revenue with $133.7 billion net income for the fiscal year ending June 30 — and framed that success as leverage to sell its own models, agents and silicon. CEO Satya Nadella promoted Microsoft’s MAI model family and Maya chips (citing a 40% performance-per-watt gain on Maya 200) and announced MAI Cyber One Flash as a competitor to larger frontier models. Nadella referenced a recent Hugging Face security incident to argue against relying on a single frontier model.