Fable 5.1 more than doubles its science benchmark and cuts cache reads to a quarter, but burns 1.7 times the output tokens.
Copilot code review can formally approve pull requests and count toward required approvals when enabled.
GitHub deprecated six models across most Copilot surfaces on September 1, pushing users onto newer replacements.
Four VS Code releases add side conversations, a rubber-duck reviewer model and per-model token meters.
Copilot model access for multi-organisation users is now set solely by the billing organisation.
Datasette-mcp 0.2 gives any Datasette install an MCP endpoint and reshapes query output for weaker models.
Positive AI sentiment in US employee reviews halved since 2019, with forced adoption outranking many complaints.
Gemini Flash models pick which video segments to examine and at what resolution, cutting tokens by up to 88 percent.
Crafty Apes chief Alana Newell argues production and bidding data, not generated imagery, is where machine learning pays.
The Visual Effects and Animation World Atlas finds no data showing AI has cut jobs, and Mumbai now leads the world.
Eddie AI adds link-based client review that exports comments straight into DaVinci Resolve as frame-accurate markers.
FlowLabs pushes live output from Resolve, Premiere or Pro Tools to client browsers over WebRTC with no transcode.
Dalet and Agence France-Presse carry shot-level metadata straight into editorial search inside Dalet Flex.
Emmy winner Dennis Baxter argues learning models are coming for commentators and broadcast engineers alike.