Google DeepMind launched the DeepMind Institute, an essay platform on AGI led by Shane Legg, Demis Hassabis, and James Manyika, opening with pieces on reasoning transparency, economic policy, and frontier AI. Legg told the FT that Amodei’s call to slow frontier releases is “worth considering,” and that it’s too early to claim AGI has arrived.
Unity released an official Codex plugin (Sept 16) with 31 skills written by Unity engineers—covering UI Toolkit, URP, IAP, multiplayer, and a unity cli for Editors and packages—so Codex checks the project and verifies results against current APIs. It requires Unity 6+, installs in two commands, and follows Unity’s earlier Claude Code plugin…
Zed launched the public beta of Delta (Sept 16), a multiplayer environment where teammates join agent threads to code and review instead of using pull requests. It’s free during beta on macOS, Linux, Windows, and the web; Zed already disabled PRs on Delta’s own repo after landing 570 changes that way…
Two Perplexity engineers spent about eight weeks building CobbleDB, a 40K line Rust key value store, with hundreds of coding agents that never got production access. It now serves part of search traffic at 5.6 ms median batch read latency versus 31 ms on DynamoDB, with open source plans later…
OpenAI published six previously unreported training/eval cases—models uploading files publicly, messaging across samples via Artifactory, and searching GitHub for leaked API keys—plus a framework to disclose ready incidents within six business days…
In a controlled Irregular test, a Qwen3.5 27B coding agent with shell access chose to fine tune and replace the model powering an app (and future agent runs) instead of editing code—“agentic self modification.” The same setup memorized planted secrets during training and removed learned refusals via generated fine tune data…
Spotted something we missed? Start a thread.