1. Yann LeCun Raised $1 Billion to Prove Large Language Models Are a Dead End For years, Yann LeCun told anyone who would listen that large language models were headed nowhere interesting.

2. Redox OS Bans AI-Generated Code While Debian Can't Even Vote on It Redox OS updated its contributor guidelines this week with a flat prohibition: no code produced by large language models will be accepted.

3. Amazon and a Federal Court Both Put Humans Back in the AI Loop Amazon now requires senior engineers to approve AI-assisted code changes before deployment.

In Brief

• OpenAI Drops Oracle from Stargate Data Center Expansion OpenAI is walking away from expanding its Stargate data center project with Oracle. The split raises questions about Oracle's debt-fueled buildout strategy as AI infrastructure demands shift.

• Iran Conflict Threatens Data Center Electricity Costs Rising oil and gas prices from the spiraling Iran conflict are pushing up electricity costs for data centers. The Atlantic Council's Reed Blakemore warns the energy price impact could tighten margins across AI infrastructure operators.

• OpenAI Publishes IH-Challenge to Harden Models Against Prompt Injection OpenAI released IH-Challenge, a training method that teaches models to prioritize trusted instructions over injected ones. The approach improves instruction hierarchy compliance, safety steerability, and resistance to prompt injection attacks across frontier models.

• Google Claims State-of-the-Art Performance for Gemini in Sheets Google launched new beta features putting Gemini directly into Google Sheets for creating, organizing, and editing spreadsheets via natural language. Users can describe tasks from basic formatting to multi-step data analysis.

• Niantic's Pokémon Go Map Data Now Guides Delivery Robots Niantic is licensing the spatial mapping data collected by millions of Pokémon Go players to give delivery robots precise, street-level navigation. The dataset offers inch-perfect 3D views of sidewalks, curbs, and obstacles that standard maps lack.

• Ford Launches AI-Powered Fleet Management Service Ford announced Ford Pro AI, a service that analyzes commercial vehicle data — speed, seat belt use, engine health — and converts it into action items for fleet managers. The system layers generative AI on top of Ford's existing telematics platform.

• ChatGPT Adds Interactive Visual Explanations for Math and Science OpenAI shipped a feature letting ChatGPT render interactive diagrams for math and science topics. Students can manipulate formulas and variables in real time rather than reading static text explanations.

• Analysis Debunks $5,000-Per-User Cost Claim for Claude Code A detailed breakdown by Martin Alderson challenges widely circulated estimates that Anthropic spends $5,000 per Claude Code user. The analysis reexamines the assumptions behind the viral figure.

• Anthropic Opens Sydney Office, Its Fourth in Asia-Pacific Anthropic is establishing a Sydney office, expanding its Asia-Pacific presence to four locations. The move follows the company's recent expansions across the region.

• HuggingFace Paper Explores How Far Unsupervised RLVR Can Scale LLM Training A new study provides a taxonomy and experimental analysis of unsupervised reinforcement learning with verifiable rewards (URLVR), which derives training signals without ground-truth labels. The method uses model-intrinsic signals to push past the supervision bottleneck.

Read the full edition →