OpenAI published two updates yesterday: how it plans to watermark ChatGPT and Codex text output under the EU AI Act, and a new visual ad format for ChatGPT with expanded measurement. Also, Simon Willison reruns a two-year-old arithmetic experiment on a local Qwen model, with and without reasoning.
From the labs
- Our approach to EU text provenance rules — OpenAI
- Building advertising for the way people use AI — OpenAI
Worth your time
- Qwen3.8 27B addition in words — Simon Willison (mental-arithmetic benchmark, reasoning on vs off)