TranslateGemma shows how targeted training helps Google squeeze more performance out of smaller models: the 12B version translates better than a base model twice its size and runs on a regular laptop. With the growing Gemma family, Google is staking its claim in the race for open AI models.
Deutsche Telekom is soon using AI voice agents from Elevenlabs in its customer service. Customers can talk to realistic-sounding AI voices around the clock through the app or by phone. The partnership between Europe's largest telecom company and the AI audio startup goes back a while. Since October 2025, Magenta customers have been able to convert text into podcasts up to 25 times a month for free in the MeinMagenta app. Deutsche Telekom also invested in Elevenlabs' Series C funding round.
According to Elevenlabs' internal data, the AI support agent successfully resolves about 80 percent of user queries, particularly when it comes to specific documentation questions. For more complex issues like troubleshooting or pricing inquiries, though, the system still hits its limits and needs to hand off to human support.
Elevenlabs recently launched a marketplace for licensed voices of famous people like John Wayne and Judy Garland. Last year, the company introduced the Eleven v3 voice model with expanded expression options.
Google has updated its open-source medical AI with MedGemma 1.5, a model capable of analyzing 3D medical scans like CTs and MRIs. The release also includes a specialized speech tool that reportedly outperforms OpenAI’s Whisper in medical dictation tasks, though strict licensing conditions apply for clinical use for both models.
Researchers at Princeton University, UCLA, and the University of Pennsylvania have developed an approach that gives AI agents persistent worlds to explore. Standard web code defines the rules, while a language model fills these worlds with stories and descriptions.
Chinese open-weight AI is conquering the world: According to a Stanford analysis, models from China have already overtaken their US counterparts in distribution and adoption. But with success come growing geopolitical and security risks.
Benchmarks are supposed to measure AI model performance objectively. But according to an analysis by Epoch AI, results depend heavily on how the test is run. The research organization identifies numerous variables that are rarely disclosed but significantly affect outcomes.
Harry Potter, Game of Thrones, 1984: researchers pulled nearly complete books from commercial language models. Two of the four systems tested didn’t even put up a fight. The findings could shape ongoing copyright lawsuits against AI companies.
