Google has demonstrated new artificial intelligence capabilities that operate directly on Pixel 10 devices, even without an internet connection. This advancement is powered by its new open and multimodal model, Gemma 4 E2B, which can analyze images, process audio, answer queries, and engage in automated tasks without sending information to external servers, thereby enhancing privacy and reducing cloud dependency.

This local processing reinforces user privacy, decreases latency, and enables functionality in offline environments. The Gemma 4 E2B model can process text, images, and audio on-device, supporting features like Agent Skills for specialized assistance and Mobile Actions for controlling phone functions via voice or text.

The model is designed with approximately 2.3 billion effective parameters (5.1 billion total) and is optimized for mobile devices using quantization techniques. Google has also released a beta version of the Tensor ML SDK for the Pixel 10 series, allowing developers to leverage the Tensor G5 TPU.

👉 Read the full article on Gizmodo en Español

This TechCrunch article was not directly accessible from the provided URL, preventing a specific summary of its content.

However, based on the title, "Kimi: Threat or menace?", it likely discusses the potential risks, ethical concerns, or market impact posed by Kimi, an AI product or service.

The introduction of new AI technologies often raises questions about their implications, which can be perceived as a "threat" or "menace" in terms of privacy, security, employment, or the spread of misinformation.

👉 Read the full article on TechCrunch

Google has rebranded its popular AI tool, NotebookLM, as Gemini Notebook, significantly enhancing its integration with the broader Gemini ecosystem. The tool continues to operate with user-selected sources, providing references to original content to mitigate the risk of generating fabricated information.

The most significant update is the incorporation of a secure cloud computer for each notebook, enabling the AI to write and execute code within the project for more advanced data analysis.

For instance, users can upload multiple spreadsheets and instruct Gemini Notebook to organize information, perform calculations, identify trends, and generate charts. It can also produce various file types, including PDF reports, PowerPoint presentations, and Excel files.

Code execution is initially available for Google AI Ultra users and specific Workspace enterprise clients, with a rollout to all Pro plan users expected soon. Projects synchronize between the standalone app and the Gemini app, with future integration planned for Google Search's AI Mode.

👉 Read the full article on Gizmodo en Español