1. April 2026
  2. One benchmark number hides which jobs a model is actually good at (interconnects.ai)
    AI · · April 20, 2026 · Evaluation · Open Models
  3. Claude Opus 4.7: bigger vision input, steadier long-running coding (anthropic.com)
    AI · · April 16, 2026 · Anthropic · Claude · Models
  4. Nathan Lambert bets open models win on economics, not benchmarks (interconnects.ai)
    AI · · April 15, 2026 · Open Models · Forecasting
  5. Jack Clark sorts attacks on AI agents into six kinds (importai.substack.com)
    AI · · April 13, 2026 · Agents · Security
  6. Safetensors moves to the PyTorch Foundation, with neutral governance (huggingface.co)
    Engineering · · April 8, 2026 · Infrastructure · Security
  7. Project Glasswing: a 12-company push to fix critical bugs before attackers do (anthropic.com)
    Security · · April 7, 2026 · Anthropic · Vulnerabilities
  8. What actually makes a coding agent work: the six parts of the harness (magazine.sebastianraschka.com)
    Engineering · · April 4, 2026 · Agents
  9. Gemma 4: small open models that punch far above their parameter count (blog.google)
    AI · · April 2, 2026 · Open Models · Google DeepMind
  10. March 2026
  11. The chatbot window is the bottleneck, not the model (oneusefulthing.org)
    AI · · March 31, 2026 · Agents · Claude
  12. DeepMind built a way to measure when AI manipulates people (deepmind.google)
    AI · · March 26, 2026 · Google DeepMind · Safety