Google AI and DeepMind News and Discussions

[firestar464]

[Powers]

while facing a single person

[spryfusion]

[weatheriscool]

Google Announces Experimental 'Thinking' Gemini AI Model
Gemini 2.0 Flash Thinking Experimental is Google's answer to Open AI o1.
By Ryan Whitwam December 20, 2024
https://www.extremetech.com/computing/g ... i-ai-model

Google has released yet another Gemini AI model, but this one isn't quite ready for prime time. The new, awkwardly named Gemini 2.0 Flash Thinking Experimental model is available in Google's AI Studio for testing. This bot uses additional computing cycles to check its work, allowing it to handle more complex queries. This could offer a hint of how Google hopes to evolve its Gemini models in 2025.

The new LLM is built on the same foundation as the recently released Gemini 2.0 Flash, which is the first version of Google's next-gen models. The "Flash" designation indicates it's a smaller model than the "Pro" variants, but Google has said the 2.0 update allows these large language models (LLMs) to perform similarly to the bulkier Gemini 1.5 Pro.

Gemini 2.0 Flash Thinking appears to be Google's answer to the OpenAI o1 model. This LLM is trained with reinforcement learning to check its work before providing an answer. This makes it slower and more computationally expensive than other models, but it can avoid some of the mistakes that other models make.

[weatheriscool]

Google's DeepMind to Develop 'World Models' for Artificial General Intelligence
AGI is the overarching goal, but improved chatbots, AI-generated media, and potentially robots are also a part of the equation.
By Adrianna Nine January 8, 2025
https://www.extremetech.com/computing/g ... telligence

DeepMind, the experimental branch of Google largely focused on artificial intelligence, is preparing to develop "world models" that mimic the real world. The models play a significant role in Google's goal to develop artificial general intelligence, or AGI, before other tech giants. But that's only the first step of the equation—once Google has its hands on AGI, it plans to use it to train robots, "cure" old AI training data, and generate backdrops for movies and video games.

Tim Brooks, a research scientist at DeepMind and a former Sora lead at OpenAI, shared via X (previously Twitter) that he was hiring AI researchers and engineers for the new project. The linked job listings say DeepMind is looking for someone to help "build generative models of the physical world," which will then be used to "train world simulators at massive scale." In turn, these simulators will be used to "power numerous domains, such as visual reasoning and simulation, planning for embodied agents, and real-time interactive entertainment," according to the listings.

[spryfusion]

Blog post: https://blog.google/technology/google-d ... uary-2025/

[weatheriscool]

Google Releases Gemini 2.0 AI Models for Everyone
Find your favorite Gemini flavor today.
By Ryan Whitwam February 7, 2025

Google has run through several LLM milestones since rolling out the first Gemini AI models, but its latest release is the biggest yet. After offering a limited preview of some Gemini 2.0 models in December, the company has now released a plethora of updated Gemini models, which are available in the Gemini app and web interface, as well as in the developer-oriented Google AI Studio and Vertex AI. The models are all still experimental, but honestly, every generative AI system feels prototype-y.

Gemini 2.0 comes in numerous flavors, from the compact and efficient Gemini 2.0 Flash to the weighty Gemini 2.0 Pro. The Flash model is particularly interesting. Google revealed the first Gemini Flash model at I/O last year, aiming to give users and developers a smaller model that can still handle big tasks. Increasingly, these efficient models offer enough functionality that you only need the larger context window of the Pro versions for analyzing large volumes of data. If you just need to have a chat with a robot, Gemini 2.0 Flash might be good enough.

https://www.extremetech.com/internet/go ... r-everyone

[weatheriscool]

Google Rolls Out File Upload Feature to Free Gemini AI Tier
The functionality is available on Android and the web... but not iPhones.
By Devesh Beri February 14, 2025

Google is rolling out its Gemini file upload and analysis feature for free users. Previously, only paying subscribers could feed data into Gemini, but now everyone can upload docs for summaries, insights, and even advanced analysis.

Once the feature is available, users will see a prompt stating, “Upload files to get quick summaries and actionable insights.” The upload can be done via the Android app or at gemini.google.com by tapping the ‘plus’ menu, where Files and Drive options now appear alongside Camera and Gallery.

https://www.extremetech.com/computing/g ... ni-ai-tier

[spryfusion]

[weatheriscool]

Google Announces Gemma 3: World’s Best Single-Accelerator Model
Gemma 3 packs an upgraded vision encoder that handles high-res and non-square images with ease.
By Devesh Beri March 12, 2025
https://www.extremetech.com/computing/g ... ator-model

Google has announced the release of Gemma 3, an update to its family of AI models. These models are made for developers to create AI applications that can run on different devices, from smartphones to workstations. Gemma 3 supports over 35 languages and can analyze text, images, and short videos.

Google claims that Gemma 3 is the "world's best single-accelerator model," outperforming competitors such as Facebook's Llama and OpenAI when running on a host with a single GPU. It is optimized for Nvidia GPUs and dedicated AI hardware, as reported by The Verge.

Gemma 3 packs an upgraded vision encoder that handles high-res and non-square images with ease. It also includes the ShieldGemma 2 image safety classifier, which filters out sexually explicit, dangerous, or violent content from both input and output images. Google found Gemma 3's improved STEM skills raised concerns about misuse of harmful substances, but tests show it's low-risk.

[spryfusion]

https://blog.google/technology/google-d ... arch-2025/

[wjfox]

Jesus...

[weatheriscool]

Google’s 'Most Intelligent AI Model' Now Available for All Users
Gemini 2.5 Pro is called a 'thinking model' due to its ability to analyze information, draw logical conclusions, and add context and nuance to its responses.
By Devesh Beri April 1, 2025

Google announced that its experimental Gemini 2.5 Pro AI model is now free for all users just a week after unveiling it. This new version, which the company says is its "most intelligent AI model" yet, was initially only available to Gemini Advanced subscribers but has now been made accessible through the Gemini app and Google AI Studio. Non-paying users will have tighter rate limits.

Gemini 2.5 Pro is called a "thinking model" because it can analyze information, draw logical conclusions, and add context and nuance to its responses. The company says this helps the model give more accurate results in tasks that require reasoning.

It also supports text, audio, image, and video inputs, making it useful for a lot of things. It can handle up to 1 million tokens in a single session.

https://www.extremetech.com/computing/g ... -all-users

[weatheriscool]

Gemini Now Lets Samsung Users Search Their Google Photos With Simple Commands
Users can type "@Google Photos" or "my photos" in the Gemini app and describe what they are looking for.
By Devesh Beri April 4, 2025

Google's AI assistant, Gemini, is now adding Google Photos to Samsung Galaxy phones and tablets. Thanks to this, users will be able to find their memories quite easily. This will let users find photos and videos from their Google Photos library using natural language commands. As a Pixel user, I'll admit that I am jealous, but I am also sure that this functionality will be introduced to Pixels sooner or later.

Users can type "@Google Photos" or "my photos" in the Gemini app and describe what they are looking for. For example, they can ask Gemini to find photos of specific people, show pictures from last summer, or locate recent selfies. The assistant can also analyze photos for details, such as recalling what was eaten during a trip or identifying a passport number.

https://www.extremetech.com/mobile/gemi ... e-commands

[weatheriscool]

Google's Veo 2 Video Generator Is Now Available for Google One AI Premium Customers
This subscription also gives users 2TB of storage, Gemini integration in Gmail, Docs, Sheets, and other Google services, and NotebookLM Plus with higher usage limits.
By Devesh Beri April 16, 2025

Google has announced the launch of Veo 2, Google's text-to-video generator, which is now available to users through Gemini as part of the Google One AI Premium plan. This lets users create eight-second video clips at 720p resolution in a 16:9 format from simple or detailed text prompts. Users can specify details such as cinematic styles, camera angles, and lens effects.

Access to Veo 2 is limited to those subscribed to the Google One AI Premium plan, which costs $20 per month. This subscription also gives users 2TB of storage, Gemini integration in Gmail, Docs, Sheets, and other Google services, as well as NotebookLM Plus with higher usage limits. Google says there is a monthly cap on the number of videos users can generate, but will send notifications as users approach this limit, as reported by ZDNet.

https://www.extremetech.com/computing/g ... ai-premium

[weatheriscool]

Google Expands Gemini Live’s Real-Time Video and Screen Sharing to All Android Users
It's no longer limited to Gemini Advanced subscribers.
By Devesh Beri April 21, 2025
https://www.extremetech.com/mobile/goog ... ll-android

Google has expanded Gemini Live's real-time video and screen sharing to all Android users, so it's no longer limited to Gemini Advanced subscribers: No more paying to use it.

To use Gemini Live, Android users must activate Gemini as they normally would. When the "Ask Gemini" window appears, a new clickable short window called "Share screen with Live" appears above it. Selecting this lets Gemini view the user's screen in real time and respond to questions based on what is shown.

Users can also use their device's camera or scroll through any app while talking to Gemini Live, which gives answers and advice about what's visible on the screen. There is a notification for Gemini Live screen sharing, and users can stop sharing their screen whenever they wish.

For those concerned about privacy, Google has included options to manage data sharing. Users can set up auto-delete for audio, video, and screen shares through Gemini Apps Activity in their Google account settings, as reported by Android Central.

[weatheriscool]

Gemini Adds Image Editing, Intensifying Competition With ChatGPT
The rollout will happen gradually over the coming weeks.
By Devesh Beri May 2, 2025

Google is rolling out new image editing tools to Gemini, accessible via the Gemini app. It can now edit and modify AI-generated images and photos uploaded by users via prompts.

Users can now change backgrounds, replace objects, add elements, and experiment with style adjustments, such as different hairstyles or hair colors. Google says users can continue refining their edits through a series of prompts, with the app preserving the context of previous changes throughout the editing session.
Gemini app

https://www.extremetech.com/computing/g ... th-chatgpt

[Yuli Ban]

Jesus...

In a time where there's a new king every other week, Gemini 2.5 has remained on top for a while.
It feels like it's the truest "GPT-4.5." If Gemini 2.5 Pro came out back in December 2023, when everyone thought GPT-4.5 was going to come out initially, or honestly if it came out at any point before o1 was first teased and the deluge of new models started, we'd have said that it's the truest successor to GPT-4 and first of a true new class of model.

[weatheriscool]

Google Plans to Roll Out Its A.I. Chatbot to Children Under 13

Source: NYT

-snip-

The chatbot will be available to children whose parents use Family Link, a Google service that enables families to set up Gmail and opt into services like YouTube for their child. To sign up for a child account, parents provide the tech company with personal data like their child’s name and birth date.

-snip-

Google and other A.I. chatbot developers are locked in a fierce competition to capture young users. President Trump recently urged schools to adopt the tools for teaching and learning. Millions of teenagers are already using chatbots as study aids, writing coaches and virtual companions. Children’s groups warn the chatbots could pose serious risks to child safety. The bots also sometimes make stuff up.

-snip-

Google acknowledged some risks in its email to families this week, alerting parents that “Gemini can make mistakes” and suggesting they “help your child think critically” about the chatbot.

The email also recommended parents teach their child how to fact-check Gemini’s answers. And the company suggested parents remind their child that “Gemini isn’t human” and “not to enter sensitive or personal info in Gemini.”

-snip-


Read more: https://www.nytimes.com/2025/05/02/tech ... -kids.html

[wjfox]

Google debuts an updated Gemini 2.5 Pro AI model ahead of I/O

May 6, 2025

Google on Tuesday announced the launch of Gemini 2.5 Pro Preview (I/O edition), an updated version of its flagship Gemini 2.5 Pro AI model that the company claims tops a number of widely used benchmarks.

Gemini 2.5 Pro Preview (I/O edition) is available via the Gemini API and Google’s Vertex AI and AI Studio platforms, and is priced the same as the Gemini 2.5 Pro model it effectively replaces. It’s also in Google’s Gemini chatbot app for the web and for mobile devices.

The model’s release comes ahead of Google’s annual I/O developer conference (hence the “I/O edition” designation), where Google is expected to unveil a host of models, as well as AI-powered tools and platforms. The company is competing fiercely for mindshare and marketshare in the cutthroat AI race; rivals like OpenAI and xAI are on the cusp of releasing models that are expected to be highly performant.

According to Google, Gemini 2.5 Pro Preview (I/O edition) has “significantly” improved capabilities for coding and building interactive web apps. The model is also better at tasks like code transformation — that is, modifying a piece of code to achieve a specific goal — and code editing, the company says.

https://techcrunch.com/2025/05/06/googl ... ad-of-i-o/