Tech News Hub
Samsung’s Ballie Home Robot Just Got Smarter with Google’s Gemini AI
Samsung is bringing some serious brains to its home robot Ballie by teaming up with Google Cloud to integrate the Gemini AI model. This collaboration means Ballie will now be able to understand and respond using both audio and visual input — making it a much more helpful and intuitive companion around the house.
Smarter Interactions, Thanks to Gemini
By combining Samsung’s AI with Google’s Gemini, Ballie can now do things like analyze how you look and suggest outfit ideas using its built-in camera. So instead of just rolling around and playing music, it can actually see and understand what’s happening around it.
Personalized Health Tips at Home
Ballie will also use Gemini to provide wellness suggestions — things like what exercises you can do or how to improve your sleep routine. You’ll also be able to ask general knowledge questions, and the robot will pull from Gemini’s database to give you useful answers on the spot.
Samsung’s Vision for AI at Home
Yongjae Kim, Executive VP at Samsung’s Visual Display Business, shared that this partnership is about more than just features — it's about changing how AI fits into our daily lives. The goal is to create an assistant that moves with you, understands your needs, and becomes a truly interactive part of your home environment.
When Can You Get One?
Samsung has shown off Ballie at events like CES for a few years now, and earlier this year they announced it would officially launch in South Korea and the U.S. in the first half of 2025. So the wait may finally be over soon.
More to Come from Samsung and Google
This isn't the first time the two tech giants have teamed up — Gemini has already been integrated into Samsung’s Galaxy S24 phones. There are also reports that they’re working together on a mixed reality (XR) device, where Gemini could be a key part of the experience too.
Related Stories:
Top Stories:
Google AI Mode Now Understands Your Photos: Ask Complex Questions About Images
Google is stepping up its AI-powered search game with a new feature that allows users to ask detailed questions about the images they upload or take with their camera. This update brings multimodal search to AI Mode, Google's experimental search experience that supports complex, multi-part queries and follow-up questions.
With this update, users can interact with photos just as easily as they would with text. Whether it’s a picture of a bookshelf, a product, or a room, AI Mode can now analyze the entire scene to provide intelligent, in-depth answers.
Smarter Searches with Photos
The new feature uses Google Lens’ advanced image recognition technology to understand not just what’s in a photo, but how everything in the image relates. It can detect colors, materials, object relationships, and more.
For example, if you take a picture of your bookshelf and ask, “If I enjoyed these, what are some similar books that are highly rated?” — AI Mode will identify the titles, understand your reading preferences, and suggest similar books. You can then follow up with something like, “Which of these is the shortest read?” to narrow things down even further.
This is made possible by a method called query fan-out, which breaks your question into multiple parts and explores each one to offer a more thorough response than a standard search.
Expanded Access
Until now, AI Mode was only available to Google One AI Premium subscribers. But starting this week, Google is expanding access to millions of users who are part of Google Labs, the company’s platform for testing experimental features.
AI Mode was introduced just last month and is already being compared to other advanced tools like Perplexity and OpenAI’s ChatGPT Search. Google says it plans to continue improving the experience, focusing on making searches more natural, visual, and intuitive.
Conclusion
Google’s latest update to AI Mode is a big step forward in making search more interactive and intelligent. By allowing users to ask questions about what they see—not just what they type—Google is opening up new ways to explore the world around us.
Related Stories:
Top Stories:
Microsoft Releases AI-Generated Quake II Demo, but It’s Not Quite the Classic Experience
Microsoft has just released an AI-powered tech demo based on the classic game Quake II, giving gamers a glimpse into the future of gaming with its Copilot AI platform. But while the demo is playable and showcases the potential of AI in game development, it’s clear that this isn't quite the same Quake II you remember.
The demo, available in your browser, allows players to navigate a single level of Quake II, using their keyboard to control movement. However, there’s a time limit, and players can only explore the environment for a few minutes before the demo ends. Despite this, the experience highlights the capabilities of Microsoft's Muse family of AI models, which are designed to let users interact with AI-driven environments in real time.
AI-Powered Gaming: The Vision
The idea behind this demo is to showcase how AI can create dynamic game worlds that respond to player actions instantly. Microsoft’s researchers have trained their AI model on a Quake II level (thanks to the company’s acquisition of ZeniMax, which owns the game). The result? Players can wander around, jump, crouch, shoot, and even blow up barrels just like in the original. In the researchers' words, it’s “playing inside the model” instead of playing the game as we know it.
However, as promising as this sounds, it’s not without its quirks.
Not Quite the Classic Experience
The demo's creators openly acknowledge the limitations of the technology. For one, the enemies in the game are "fuzzy" and lack the sharpness and precision players expect from a well-built game. The damage and health counters also tend to be inaccurate, and the model struggles with object permanence—meaning it forgets objects that are out of view for a mere 0.9 seconds. This leads to some amusing and unexpected outcomes, like spawning enemies by briefly looking away or "teleporting" around the map by glancing at the sky and then back down.
While these bugs may seem like fun glitches, they do highlight the gap between what AI models can currently do and what makes games like Quake II so memorable. In a statement, Microsoft framed the demo as a “research exploration,” emphasizing that this is more about experimenting with AI’s potential rather than replicating the classic game experience.
Criticism and Concerns: Does AI Understand Games?
Not everyone is impressed by this attempt to bring classic games into the future through AI. Austin Walker, a writer and game designer, shared his thoughts on the demo in a gameplay video where he spent much of his time stuck in a dark room, unable to make much progress. Walker argued that Microsoft’s push to use AI for game preservation and portability misses a crucial point: the internal mechanics of games like Quake II are what make them work.
According to Walker, the core elements of a game—its code, design, 3D art, and audio—combine to create specific play experiences, often leading to surprising and unpredictable moments. If AI can’t fully reconstruct these core mechanics, then it can’t truly capture the magic of these classic games. Essentially, he warns that using AI to “port” games could result in a watered-down experience that lacks the edge cases that make them unique.
The Future of AI in Gaming: A Work in Progress
While the AI-powered Quake II demo may not replace the original, it certainly demonstrates the potential of AI in reshaping the gaming landscape. The technology is still in its early stages, and the glitches may be frustrating, but they also show us where AI can go. With continued development, Microsoft’s Muse models might one day power games that are more immersive, dynamic, and responsive than ever before.
In the meantime, however, Quake II fans may want to stick with the classic if they’re looking for the real experience. The demo is fun as a tech showcase, but it still has a long way to go before AI can truly recreate the magic of a well-crafted game.
Related Stories:
Top Stories:
Meta Launches Llama 4: Smarter AI Models Built for the Next Generation of Tasks
Meta has officially introduced its latest collection of AI models under the Llama 4 banner. The release, which surprisingly happened over the weekend, includes three major models: Llama 4 Scout, Llama 4 Maverick, and the still-in-training Llama 4 Behemoth. Designed to handle complex reasoning across text, images, and videos, these models mark a significant step forward in Meta's open AI strategy.
What’s New in Llama 4
Llama 4 Scout
Scout is built for tasks like document summarization, reasoning over large codebases, and understanding long-form content.
- 109 billion total parameters
- 16 expert models, with 17 billion active parameters
- Massive context window of 10 million tokens
- Efficient enough to run on a single Nvidia H100 GPU
Llama 4 Maverick
Maverick is aimed at general assistant and chatbot use cases, such as creative writing and coding help.
- 400 billion total parameters
- 128 experts, also with 17 billion active parameters
- Requires more computing power, such as an Nvidia H100 DGX system
- Outperforms OpenAI’s GPT-4o and Google’s Gemini 2.0 in certain benchmarks, but not quite at the level of the latest GPT-4.5 or Gemini 2.5 Pro
Llama 4 Behemoth (coming soon)
Behemoth is Meta’s most powerful model yet.
- Nearly 2 trillion total parameters
- 16 experts with 288 billion active parameters
- Outperforms GPT-4.5, Claude 3.7 Sonnet, and Gemini 2.0 Pro in tasks like math and problem-solving
All three models use a Mixture of Experts (MoE) architecture. Instead of having one large model handle everything, MoE allows different parts of the model — called experts — to focus on specific subtasks. This approach improves performance and efficiency, especially when dealing with complex queries.
Availability and Licensing
Scout and Maverick are available now through Llama.com and platforms like Hugging Face. Behemoth is still being trained.
Meta has already integrated Llama 4 into Meta AI, the company’s virtual assistant used across apps like WhatsApp, Messenger, and Instagram in 40 countries. For now, the assistant’s ability to handle image and video input is only available in the United States, and only in English.
However, there are some notable licensing restrictions:
- Developers and companies located in the European Union are not allowed to use or distribute the models, likely due to regional AI and data privacy regulations.
- Companies with more than 700 million monthly active users must request special permission from Meta to use Llama 4.
A More Balanced and Responsive AI
One major update in Llama 4 is how the models respond to politically and socially sensitive questions. Meta says it has tuned the models to:
- Answer a broader range of questions — even controversial ones — more consistently
- Avoid outright refusal unless truly necessary
- Offer factual responses without judgment or favoritism
These changes come in response to criticism that AI models are too selective or biased in their responses. Some political figures have accused AI platforms of promoting certain viewpoints, but Meta’s latest update attempts to make its models more neutral and flexible across different perspectives.
How They Compare
Image Credits:chatgpt
Final Thoughts
Llama 4 is a major step forward for Meta in the competitive AI landscape. With powerful new models, support for multimodal input, and a more open approach to content, Meta is pushing the boundaries of what open-source AI can do.
Despite some limitations in availability and licensing, Llama 4 is likely to attract strong interest from developers, researchers, and companies looking for high-performance AI tools that can be adapted to real-world applications.
Related Stories:
Top Stories:
Google Unveils Gemini 2.5 Pro: Its Most Advanced AI Model Yet
Google has just pulled back the curtain on Gemini 2.5 Pro, its most capable and high-performing AI model to date — and it’s built for power users who need cutting-edge performance in coding, reasoning, and math.
But this kind of capability doesn’t come cheap.
A Premium AI Experience, With a Premium Price
Gemini 2.5 Pro is now available through Google’s API, and its pricing reflects the model’s advanced capabilities. For inputs up to 200,000 tokens (roughly equivalent to 750,000 words), developers will pay $1.25 per million input tokens and $10 per million output tokens.
If you're working with longer prompts — anything over 200,000 tokens, which few other models can even support — the rate doubles to $2.50 for input and $15 for output per million tokens.
This makes Gemini 2.5 Pro Google’s most expensive AI model to date, well above the cost of its lighter sibling, Gemini 2.0 Flash, which is priced at just $0.10 for input and $0.40 for output per million tokens.
How It Stacks Up
Compared to other major AI offerings, Gemini 2.5 Pro sits somewhere in the middle:
- More expensive than OpenAI’s o3-mini and DeepSeek’s R1
- More affordable than Anthropic’s Claude 3.7 Sonnet and OpenAI’s GPT-4.5
- Far cheaper than OpenAI’s newest o1-pro, which tops the charts at $150 for input and $600 for output per million tokens
In short, Gemini 2.5 Pro is built for developers who need serious power — and are willing to pay for it.
Developer Response and Rising Demand
Despite its high cost, developers seem eager to adopt it. Google says Gemini 2.5 Pro is now the most in-demand model on its AI Studio platform, with usage jumping by 80% this month alone.
The reason? It delivers. Whether you’re building tools that solve complex math problems or writing intricate code, Gemini 2.5 Pro is designed to handle it with ease.
A Sign of Things to Come?
Gemini 2.5 Pro is part of a broader trend: AI models are getting more powerful — and more expensive. Major players like Google, OpenAI, and Anthropic are all shifting toward premium offerings, likely due to the high cost of compute and growing demand for more sophisticated AI.
Still, there’s a silver lining. Gemini 2.5 Pro is available for free (with usage limits), allowing developers to test its capabilities before making any commitments.
Conclusion
Gemini 2.5 Pro represents a big step forward for Google in the AI arms race. It’s smart, scalable, and packed with potential — but it comes with a price tag to match. For developers and businesses looking to leverage top-tier AI for their most demanding tasks, Gemini 2.5 Pro might be just the tool they’ve been waiting for.
Related Stories:
Top Stories:
Microsoft Copilot Just Got Smarter: Now It Can Take Action, See What You See, and Remember What You Like
To mark its 50th anniversary, Microsoft has given Copilot a big upgrade — and it’s more helpful than ever.
Copilot, Microsoft’s AI assistant, can now browse the web for you, perform tasks like booking tickets or sending flowers, and even remember your preferences. Whether you’re planning a weekend trip or just trying to organize your desktop, Copilot is ready to help out in ways that feel more personal and useful.
It Can Now Take Action on the Web
This is one of the biggest updates so far. Copilot can now complete real-world tasks on websites. For example, if you say, “Send flowers to my friend,” it can actually do it — thanks to partnerships with popular platforms like OpenTable, Expedia, Booking.com, Tripadvisor, and others.
No more switching tabs or clicking around. Just type what you need, and Copilot gets to work.
It Remembers What You Like
Copilot now remembers things you’ve shared in past conversations — like your favorite foods, movies, or places to stay — and uses that information to make smarter suggestions next time.
If that sounds a bit too personal, don’t worry. Microsoft has added privacy controls that let you review, delete, or turn off memory features at any time. You’re always in control.
It Can See What You’re Seeing
If you’ve ever looked at a plant and thought, “What is that?”, Copilot can now help. On mobile, it can look at what’s in your camera view or photo gallery and answer questions about it.
On Windows, it’s even more powerful — it can check what’s on your screen and help you search, change settings, or organize files. This will first be available to Windows Insider Program users.
Turn Articles into Conversations
Another cool feature: Copilot can now turn long articles or studies into short podcast-like conversations between two AI hosts. You can listen to the summary and even jump in with questions as they go.
It’s a great way to learn on the go or catch up on a topic without reading walls of text.
Stay Organized with Pages
If you’re working on a project, Copilot can now help you collect notes, research, and ideas in a single space — kind of like a digital whiteboard. From there, it helps turn your rough notes into a document or report.
Plus, the new Deep Research feature can dig through websites, files, and images to help answer more complex questions — like comparing sources or gathering background info for a presentation.
It Can Track Deals, Too
Looking to buy something? Ask Copilot to track a product, and it’ll keep an eye out for price drops. When there’s a deal, it lets you know and shares a link to buy.
What’s the Catch?
While these features sound great, Microsoft hasn’t shared much about how reliable they are just yet. And like other AI tools, Copilot could face limits — some websites might block it from doing certain things.
There are also valid concerns around privacy, especially when it comes to accessing your screen or personal files. Microsoft says it’s building in safeguards, but we’ll have to see how that plays out.
Final Thoughts
With these upgrades, Copilot is turning into more than just a chatbot — it’s a real assistant. From helping you plan your day to managing your digital life, it’s clearly designed to save time and make things easier.
Related Stories:
Top Stories:
Amazon’s New AI Shopping Agent Can Buy from Third-Party Websites for You
Amazon is testing a groundbreaking AI shopping agent called “Buy for Me”, designed to help users purchase products even when they’re not available on Amazon. This new feature, announced in a blog post, allows a subset of users to shop from external websites directly within the Amazon Shopping app—without needing to visit those sites themselves.
How “Buy for Me” Works
If Amazon doesn’t sell an item a user is searching for, Buy for Me will display relevant products from third-party websites. Users can then select a product and request a purchase without ever leaving the Amazon app.
Behind the scenes, Amazon’s AI agent will:
- Visit the external website.
- Select the requested product.
- Auto-fill shipping and payment details.
- Complete the purchase on the user’s behalf.
This hands-off shopping experience could reshape e-commerce by making Amazon a central hub for even more online transactions.
AI-Powered Shopping Expansion
Amazon joins the ranks of OpenAI, Google, and Perplexity, which have also introduced AI shopping assistants capable of browsing websites and helping users make purchases. However, Amazon’s approach is different. Instead of requiring users to manually enter credit card details, Buy for Me automatically processes transactions using encrypted billing information—keeping Amazon itself from seeing purchase details.
The AI agent relies on Amazon Nova AI models and Anthropic’s Claude to function. One of these, Nova Act, is an autonomous AI agent Amazon recently introduced, capable of interacting with websites.
Potential Risks and Concerns
While AI-driven shopping offers convenience, it also raises concerns. Handing over purchasing power to AI—especially a system prone to errors—could make some users hesitant. There’s always the possibility of unintended purchases, like ordering 100 items instead of one.
Additionally, returns and exchanges will be handled through the original third-party retailer, not Amazon, which may complicate customer service interactions.
A New Era of E-Commerce?
By integrating AI into third-party shopping, Amazon is positioning itself as an even bigger player in the e-commerce space. However, whether users embrace this level of automation—or prefer to retain direct control over their purchases—remains to be seen.
Related Stories:
Top Stories:
