Findings5

• Google launches Gemini Veo 3 video model in India with audio support

  Google rolled out Veo 3, a Gemini-branded AI video model that generates 8-second 720p clips with synchronized visual and audio components, to Pro-tier Gemini app users in India; Google says Veo 3 is roughly twice as fast as prior Ultra-tier offerings and videos include visible and SynthID watermarks for transparency.

  via TabTab11h ago·Source

  UsefulAdd the first comment

• Google permits engineers internal access to Anthropic Claude models including Opus 5

  Google changed an internal policy to allow all engineers to use Anthropic’s Claude models, including Opus 5, via its internal platform Antigravity, while stating Gemini remains the company’s primary in-house foundation model; third-party models will be available under per-user quotas for supplementary use cases.

  via TabTab11h ago·Source

  UsefulAdd the first comment

• Apple ships iOS 27 with Siri powered by Google’s Gemini models

  Apple began rolling out iOS 27, which ships a redesigned Siri powered by Google’s Gemini models to enable longer multi-turn conversations, on‑device context-aware actions, and translations; the launch places Gemini behind the default assistant on hundreds of millions of iPhones. The arrangement raises questions about data flows and privacy controls between Apple and Google.

  via TabTab1d ago·Source

  UsefulAdd the first comment

• Google expands Dreambeans experimental app availability across the U.S.

  Google expanded Dreambeans, a Google Labs experiment that synthesizes user-approved signals from connected Google services (Calendar, Gmail, Photos, Search, YouTube, and Gemini) into personalized daily story collections, to a wider set of U.S. Google Accounts on Android and iOS. Google says users must consent to connected sources and can tune privacy choices independent of other Personal Intelligence settings.

  via TabTab1d ago·Source

  UsefulAdd the first comment

• DeepMind launches AlphaGenome Atlas mapping ~9 billion single-letter DNA variants

  Google DeepMind published AlphaGenome Atlas, a searchable dataset of model-predicted molecular effects for roughly 9 billion possible single-letter human genome variants, intended to help researchers prioritize and interpret genetic changes. DeepMind says the underlying AlphaGenome model was used to precompute the predictions and the resulting dataset is approximately 1 petabyte, available via a web portal for academic research.

  via TabTab1d ago·Source

  UsefulAdd the first comment