Alibaba has open-sourced the AI code-review assistant it says served tens of thousands of developers and flagged millions of defects internally. Open Code Review combines an agent with deterministic checks, line-level comments, codebase search, and full-file scans, then targets fewer false alarms: its benchmark covers 200 real pull requests, 10 languages, and 1,505 engineer-annotated issues. The trade-off is explicit: higher precision and lower token use, but lower recall than general-purpose coding agents.
YuE2’s biggest barrier was not the model’s ambition, but the GPU needed to run it locally. audio.cpp now ships Q4_0 and Q8_0 GGUF weights, matching demos, and a native runtime path that brings testing into roughly the 8 to 9 GB VRAM range. It is still on the dev branch, so expect rough edges, but this makes serious open music generation practical on consumer cards.
A free, open-source app scans Bluetooth Low Energy broadcasts for manufacturer identifiers associated with smart glasses, then warns you when one is nearby. That makes it a useful privacy signal, not proof of recording: the author warns it can produce false positives, miss devices, or flag another product from the same company. The code runs without ads or telemetry, giving users a more inspectable countermeasure than a closed commercial detector.