Gemini describes what your camera sees
Google has launched Guided Vision in Gemini Live for conversational, real-time visual assistance. Share an Android phone's camera to hear descriptions, ask follow-up questions, and get spoken cues to reframe the view.
Google points to practical tasks including reading fine print, finding a dropped earbud, identifying clothing colors, and understanding an unfamiliar room. It is available on Android 9 and later wherever Gemini Live is supported.
Guided Vision is not intended for navigation, obstacle detection, or replacing a white cane or other mobility aid.
Copilot can operate desktop apps
Computer use is in public preview in GitHub Copilot CLI and the Copilot app on macOS and Windows. It can read visible app content, click controls, enter text, scroll, drag, and move through workflows in software without an API or command-line interface.
Copilot asks before taking control of an app, and organization-managed settings can disable the feature. GitHub suggests describing the desired outcome, the apps involved, and important constraints.
Faster building blocks for voice agents
Microsoft's MAI-Transcribe-2-Streaming provides real-time transcription in 60 languages with continuous automatic language detection. Microsoft says it begins producing partial transcripts a little over 100 milliseconds after receiving audio, letting captions appear while someone speaks or a voice agent begin work before the sentence ends.
The release also includes MAI-Voice-2.1 and a faster Flash variant for multilingual text-to-speech. Developers can access all three models through Microsoft Foundry, with the voice models also available through OpenRouter.
Also
- Turn existing assets into YouTube ads Google's Asset Studio can turn website URLs and photos into videos that stakeholders preview before publishing.
- Organize a season of interviews This workflow batches transcripts through Claude to build timelines, group quotes by theme, and flag contradictions before drafting.
- An argument that LLMs do not reason AlphaGo team member Thore Graepel argues that trustworthy reasoning needs an inspectable record of hypotheses, evidence, and unresolved questions.
- Coordinate a team of coding agents A developer shares a Claude Code setup where a coordinator creates small tickets, coders work on separate branches, and a reviewer checks each pull request.
Pass it on