Google AI and DeepMind News and Discussions

[weatheriscool]

Google Releases Full-Screen Gemini App for iPad
There is a new ability for Gemini to connect with Google Photos, letting users ask questions about their images and search for photos.
By Devesh Beri May 8, 2025
https://www.extremetech.com/mobile/goog ... p-for-ipad

Google has brought a full-screen, optimized Gemini app to the iPad, which has been the subject of complaints from users. The updated app now uses the entire iPad display in both landscape and portrait orientations, making it similar to the experience found on Android tablets and the Gemini website.

Users can now read chats in a less cramped layout, and the app supports Split View multitasking, letting them use Gemini alongside other apps, as reported by 9to5Google.

The new release also adds Gemini as a home screen widget, so users can now quickly access its tools directly from the iPad's home screen. Another update is the ability for Gemini to connect with Google Photos, letting users ask questions about their images and search for photos.

[spryfusion]

Blog post: "AlphaEvolve: A Gemini-powered coding agent for designing advanced algorithms"

[wjfox]

[Cyber_Rebel]

That Google IO was literally like watching something 20 minutes into the future. Imagine being shown this conference 10 or even 5 years ago. Safe to say they most certainly are the current leaders at least for the moment. All of the agentic advancements demoed were something that both has real world seamless applications and works in sync with their own infrastructure. We quite literally are about to have "proactive" AI which you don't have to prompt for beforehand while learning all about a person's lifestyle. Yes, just about like the "Her" movie.

https://blog.google/products/gemini/gem ... emini-live

I'm not certain what I was most excited about overall, especially the mention of World Models, near perfect AI video with sound and voice, or the interesting Diffusion architecture approach to Gemini; but perhaps the fact that they will finally FIX Google search.

Here's more relating to Gemini Diffusion: https://deepmind.google/models/gemini-diffusion/

Could Diffusion perhaps be an alternative path towards AGI? Or something which could help LLMs scaffold on the way there? Google so far to my knowledge are the only major AGI company who have discussed alternative architecture.

[spryfusion]

[spryfusion]

[spryfusion]

[wjfox]

Incredible. But also kind of eerie. Knowing that none of these people are real.

[spryfusion]

[zzz]

The bedridden woman is another level.

[weatheriscool]

Google Rolls Out Gemini AI Summaries in Gmail Side Panel, Will Be the Default on Phones
The mobile rollout for both groups begins June 24, 2024, and may take up to 15 days for all users to see the new options.
By Devesh Beri May 30, 2025
https://www.extremetech.com/mobile/goog ... nel-phones

Google has announced the general availability of Gemini in the Gmail side panel, which brings AI-powered tools to help users manage their inboxes. This update comes after Gemini was added to the side panels of Google Docs, Sheets, Slides, and Drive, and is now available to both web and mobile versions of Gmail. The most stand-out feature is the default email summarization. But let's also see what else is there.

Gemini in Gmail uses Gemini 1.5 Pro. Users can now ask Gemini to summarize email threads, say responses to ongoing conversations, help draft new emails, and find specific information from their inbox or Google Drive files. For example, users can ask Gemini questions like "What was the PO number for my agency?" or "When is the next team meeting?" and receive quick answers without leaving Gmail.

On iOS and Android mobile devices, Gemini gives a summarized view of email threads by default, making it easier to catch up on conversations on smaller screens. This is especially useful for long email threads.

[weatheriscool]

[weatheriscool]

[firestar464]

gemini 0605 is out. (you can find it on AI Studio)

[spryfusion]

[spryfusion]

[weatheriscool]

Google Releases Upgraded Gemini 2.5 Pro AI Model in Preview
The company says the model is now more creative and better at formatting responses.
By Devesh Beri June 9, 2025

Google has released a preview version of Gemini 2.5 Pro, the new version of its AI model that the company claims is its smartest yet. The update builds on the model shown at Google I/O in May and will be available to everyone in the coming weeks.

Gemini 2.5 Pro is made for enterprise-scale applications and is now accessible to developers through the Gemini API, Google AI Studio, Vertex AI, and the Gemini app.

Google says the model continues to do well at coding, leading on difficult coding benchmarks such as Aider Polyglot. It also shows top-level performance on GPQA and Humanity’s Last Exam (HLE), which tests math, science, knowledge, and reasoning abilities.

https://www.extremetech.com/computing/g ... o-ai-model

[firestar464]

[spryfusion]

Blog post: "Gemini Robotics On-Device brings AI to local robotic devices"

[weatheriscool]

Google Launches New Imagen 4 Text-to-Image Models for Super Realistic Results
All images generated by Imagen 4 models will have a non-visible digital watermark.
By Devesh Beri June 26, 2025

https://www.extremetech.com/computing/g ... ic-results
Google has brought its latest text-to-image model, Imagen 4, to paid preview in the Gemini API and for limited free testing in Google AI Studio. The company says this new model is its best text-to-image system so far, with improvements in how well it renders text within images—which hasn't been the best thing about Generative AIs ever since its beginning.

The Imagen 4 family includes two main options: Imagen 4 and Imagen 4 Ultra. Imagen 4 is intended as the standard model for most needs, letting users make a wide variety of images with better quality, especially for text inside images. Each image made with Imagen 4 costs $0.04. Imagen 4 Ultra, priced at $0.06 per image, is made for those who want their images to go well with complex or detailed prompts.