Gemini AI News & updates this week

Google’s Gemini AI is coming to your TV

techcrunch.com • Sep 22, 2025

When fully rolled out, Google's AI will be on over 300 million active Google TV and other Android TV OS-powered devices....

TL;DR

Google's AI assistant, Gemini, is expanding to Google TV, enabling natural language conversations and AI-powered suggestions for TV content and daily tasks.

Key Takeaways:

Gemini will be available on over 300 million active Google TV and Android TV OS-powered devices.
The AI can be used for tasks such as suggesting TV shows and movies, providing homework help, and planning vacations.
Initially available on the TCL QM9K series, with rollout to other compatible devices to follow later in the year.

Gemini is coming to Google TV starting today

www.theverge.com • Sep 22, 2025

After Google removed the membership fee for Gemini in Chrome and introduced new browser features last week, Gemini is now being added to Google TV. It...

TL;DR

Google is introducing Gemini, an AI-powered assistant for Google TV, which allows users to have 'free-flowing conversations' and find shows based on specific details or vague descriptions.

Key Takeaways:

Gemini on Google TV will be available on TCL's QM9K series TVs first, with wider availability on other devices later this year.
The assistant can provide text responses to broader questions and suggestions for related YouTube videos.
Microsoft's Copilot AI assistant will also be integrated into Samsung and LG smart TVs debuting this year, marking increased competition in the AI assistant market.

Gemini Robotics 1.5 brings AI agents into the physical world

reddit.com • Sep 25, 2025

TL;DR

Google introduces Gemini Robotics 1.5 and Gemini Robotics-ER 1.5, two AI models that bring agentic experiences, embodied reasoning, and advanced capabilities to physical robots.

Key Takeaways:

Gemini Robotics-ER 1.5 achieves state-of-the-art performance on academic and internal benchmarks for embodied reasoning.
Gemini Robotics 1.5 enables robots to think before acting, generalizing to solve new tasks and be more robust to environmental changes.
The models learn across different embodiments, accelerating skill learning and allowing robots to adapt to new tasks and environments.

Improved Gemini 2.5 Flash and Flash-Lite

developers.googleblog.com • Sep 25, 2025

Article URL: https://developers.googleblog.com/en/continuing-to-bring-you-our-latest-models-with-an-improved-gemini-2-5-flash-and-flash-lite-release/ ...

TL;DR

Google releases updated Gemini 2.5 Flash and Flash-Lite models with improved performance and efficiency.

Key Takeaways:

Updated Gemini 2.5 Flash model sees a 5% gain on SWE-Bench Verified, and Gemini 2.5 Flash-Lite features stronger multimodal & translation capabilities.
Gemini 2.5 Flash-Lite now produces more concise answers, reducing token costs and latency for high-throughput applications.
Introduces a -latest alias for each model family, allowing easy access to the latest model versions and reducing the need for code updates.

How Google’s dev tools manager makes AI coding work

techcrunch.com • Sep 23, 2025

Google PM Ryan Salva is responsible for tools like Gemini CLI, giving him a front-row seat to the ways AI tools are changing software development....

TL;DR

Google's AI tool research shows most developers started using AI tools in April 2024, marking the beginning of the reasoning or thinking models era, and the future of development may shift towards more requirements-focused work.

Key Takeaways:

Most developers started using AI tools in April 2024, coinciding with the releases of Claude 3 and Gemini 2.5.
The use of AI tool-calling has enabled self-correction in models, enabling them to leverage external information for problem-solving.
In the future, development work may shift towards more focus on requirements, with developers potentially taking on more high-level tasks similar to architects, rather than just writing code.

From MCP to shell: MCP auth flaws enable RCE in Claude Code, Gemini CLI and more

verialabs.com • Sep 23, 2025

Article URL: https://verialabs.com/blog/from-mcp-to-shell/ Comments URL: https://news.ycombinator.com/item?id=45348183 Points: 75 # Comments: 24...

TL;DR

A group of researchers at Veria Labs discovered vulnerabilities in several popular AI tools, including Cloudflare's use-mcp library, Anthropic's MCP Inspector, and Claude Code, that can be exploited for Remote Code Execution (RCE) and account takeover.

Key Takeaways:

The vulnerabilities allow attackers to execute arbitrary code on user machines, potentially leading to session hijacking and account takeover.
The issues were found in the way MCP clients implement OAuth, specifically by trusting URLs supplied by malicious MCP servers.
Affected tools have since been patched, with Cloudflare, Anthropic, and Google implementing various security fixes and updates to prevent future exploits.

Gemini AI will help you play games on Google Play

www.theverge.com • Sep 23, 2025

Google will soon let you ask its Gemini AI assistant for help progressing in mobile games. On Tuesday, Google announced that a Gemini Live integration...

TL;DR

Google will integrate Gemini AI chatbot into the Google Play Store, enabling users to ask for in-game hints and recommendations while playing mobile games.

Key Takeaways:

Google's Gemini AI assistant will provide in-game hints and recommendations to users playing select games over the coming months.
The integration aims to offer a more interactive gaming experience and improve user engagement.
Over 200,000 Google Play games will be available to play across mobile and PC, with more games adopting the new in-game overlay.

Gemini now explains why your Sheets formula failed

www.theverge.com • Sep 25, 2025

Nine months after Google infused Gemini AI into Google Sheets, the AI system graduated from text and charts to taking on formulas. (Even those plopped...

TL;DR

Gemini AI in Google Sheets now provides step-by-step instructions on how to calculate and debug formulas, and can even suggest alternative functions and corrections for errors.

Key Takeaways:

Gemini AI can assist with complex calculations and offer explanations for formulas in Google Sheets.
The AI system can also suggest step-by-step instructions for manual calculations or scripting solutions when direct formula support is not available.
Users can leverage Gemini's capabilities to increase productivity and accuracy in Google Sheets, reducing the need for extensive Google searches or trial and error.

TCL’s Google TVs with an mmWave presence sensor are out now

www.theverge.com • Sep 22, 2025

You can now buy TCL’s new QM9K series QD-Mini LED TVs, and they run Google TV, are the first to come with Gemini AI, and have a built-in mmWave presen...

TL;DR

TCL has released its new QM9K series QD-Mini LED TVs, running Google TV with Gemini AI and a built-in mmWave presence sensor.

Key Takeaways:

The QM9K series TVs start at $2,999 and go up to $99,999 for the largest models.
The mmWave presence sensor allows for automatic TV on/off and contextual screen savers based on proximity.
Gemini AI support will be extended to other 2025 TCL TVs, Hisense TVs, Google TV Streamer, and Walmart Onn 4K Pro devices.

Sam Altman’s AI empire will devour as much power as New York City and San Diego combined. Experts say it’s ‘scary’ - Fortune

fortune.com • Sep 24, 2025

Sam Altman’s AI empire will devour as much power as New York City and San Diego combined. Experts say it’s ‘scary’ FortuneEnergy-aware machine learnin...

TL;DR

Sam Altman's AI company plans to build data centers consuming 10-17 gigawatts of power, comparable to the electricity demand of two US cities or a small country, raising concerns about environmental costs and grid impact.

Key Takeaways:

OpenAI's data centers could consume up to 10-17 gigawatts of power, equivalent to powering two US cities or a small country.
Experts warn that the environmental costs of AI growth, including carbon emissions, water scarcity, and e-waste, are becoming significant concerns.
Nuclear power is not a viable solution in the short term, with most experts suggesting reliance on renewables, natural gas, and new storage technologies to meet AI's electricity demands.

Google’s Mixboard is an AI moodboard builder

www.theverge.com • Sep 23, 2025

Google is launching a new AI concepting tool that lets you create moodboards to explore design ideas. The Mixboard Google Labs experiment gives users ...

TL;DR

Google launches Mixboard, an AI moodboard builder that uses generative AI to create Pinterest-like design boards.

Key Takeaways:

The Mixboard uses Google's Gemini 2.5 Flash model and allows users to create projects from pre-populated board templates or text prompts.
Users can edit boards with natural language prompts and adjust or combine images, or reference uploaded images to generate new visuals.
The tool is available through public beta in the US and is visually similar to platforms like FigJam or Adobe's Firefly Boards.

Microsoft’s AI CEO on the future of the browser

www.theverge.com • Sep 23, 2025

The AI browser wars are heating up. Google has Gemini in Chrome, Perplexity is building its Comet AI browser, and The Browser Company just got acquire...

TL;DR

Microsoft's AI CEO plans to integrate Copilot into the Edge browser, enabling AI to control tabs, navigate pages, and read content, while users will still be in control.

Key Takeaways:

Microsoft's Edge browser will be evolving to become an agentic browser, controlled by AI.
Copilot will allow users to see AI browsing in real-time, and transparency will create trust.
Windows user base and competition with OpenAI's rumored AI web browser are factors driving Microsoft's Copilot integration.

Google Photos expands conversational editing to more Android phones

www.theverge.com • Sep 23, 2025

After launching on Pixel 10 devices, the Google Photos feature that allows users to edit images by describing the changes they want to make is now com...

TL;DR

Google Photos' conversational editing feature is expanding to more Android phones, enabling users to edit images by describing the changes they want.

Key Takeaways:

More Google Photos users can now edit images by describing their edits using voice or text instead of manually using adjustment sliders.
The feature aims to make photo editing quicker and more accessible, providing suggested edits and support for vague commands.
Currently rolling out to eligible Android users in the US, accessible by tapping 'Help me edit' in the Google Photos editor.