1. Google puts Gemini on robots, no cloud needed
Google DeepMind released Gemini Robotics On-Device, a stripped-down model that runs directly on robots. Early demos showed machines unzipping bags, folding clothes and tackling simple assembly without an internet link. An open-source SDK lets developers train robots with just 50–100 task demonstrations. (techcrunch.com)
2. Landmark ruling backs Anthropic’s training data
A US federal judge decided that Anthropic’s use of published books to train its language models counts as fair use. While the ruling doesn’t close every copyright case, it is the first big legal win for an AI firm on this issue. A separate trial will decide any damages for pirated copies. (techcrunch.com)
3. ElevenLabs launches mobile voice-generation app
Creators can now craft studio-quality voice clips on iOS and Android. The new app taps ElevenLabs’ latest “v3 alpha” model, offering about 10 minutes of free generation time and shared credits with the web version. (techcrunch.com)
4. Synthflow raises $20 million for enterprise voice bots
Berlin startup Synthflow, which helps companies build no-code, voice-AI customer-service agents, secured a Series A led by Accel. The platform already handles five million calls a month for more than 1 000 clients and plans a US office next. (techcrunch.com)
5. Abridge doubles valuation to $5.3 billion in four months
Medical-scribe specialist Abridge closed a $300 million Series E round led by Andreessen Horowitz. The cash will fuel expansion from transcribing doctor visits into automated medical coding, a move aimed at easing clinicians’ paperwork. (techcrunch.com)
6. Wispr Flow picks up $30 million to rethink dictation
Wispr Flow’s AI-powered dictation app—already popular with venture capitalists—won new backing from Menlo Ventures and others. The company supports 104 languages and plans Android and enterprise releases later this year. (techcrunch.com)