AI news.
AI releases and research, with practical context and original sources.
More stories
News & analysis
Google tests when routing an AI task is worth the extra step
A new Google engineering study compares model routing with a simpler configuration change. The useful measure is what a verified result costs, including retries and checks.
Google runs an AI model on its prototype satellite
Google reports a successful AI response from an orbiting TPU. The next tests concern whether the hardware can keep working reliably in space.
OpenAI plans a shared signal for website AI training preferences
OpenAI plans a shared training-preference signal. Website owners can review that choice separately from search access and private content.
OpenAI updates its Model Spec for AI actions and oversight
OpenAI’s revised behavior guide adds detail on context, human oversight and the scope of an AI agent’s actions. Businesses still need to test and control the tools they connect.
Gemini adds in-chat editing for generated documents
Google’s October file-editing update, its format/platform boundaries and a worked change record for checking one revision through to the exported document.
Google’s new agent-testing guide checks what changes between versions
Google’s October 28 walkthrough shows how to rerun the same requests after an AI change, compare the results and inspect failures before a release.