Category: Artificial intelligence

  • DeepSeek Launches Enhanced V3 AI Model Just 3 Months After Release

    DeepSeek Launches Enhanced V3 AI Model Just 3 Months After Release

    Key Takeaways

    1. DeepSeek introduced the DeepSeek-V3-0324, an updated version of its V3 AI large-language model with improved performance and reduced hardware requirements.
    2. The V3 model is a non-reasoning AI designed for quick answers, featuring 685 billion parameters, making it one of the largest publicly available LLMs.
    3. The updated model scored between 5.3% and 19.8% higher on AI benchmarks compared to its predecessor, competing well with other AIs like GPT-4.5 and Claude Sonnet 3.7.
    4. Enhanced capabilities include improved web page design, user interface creation for online games, and better skills in Chinese searching, writing, and translating.
    5. To run the full DeepSeek-V3-0324 model, users need 700 GB of storage and Nvidia A100/H100 GPUs, while smaller versions can operate on a single Nvidia 3090 GPU.

    DeepSeek, a Chinese artificial intelligence firm, has introduced the DeepSeek-V3-0324, an updated iteration of its V3 AI large-language model (LLM), which was first unveiled in December 2024. This V3 model astonished many due to its significantly reduced hardware needs for training, quicker training durations, and lowered API expenses, while still delivering high performance compared to rival AIs like OpenAI’s GPT models.

    Key Features of the V3 Model

    The revised V3 model is classified as a non-reasoning AI, which means it focuses on providing quick answers without taking extra time to ponder complex issues, in contrast to DeepSeek’s R1 model. With a massive size of 685 billion parameters, it ranks among the largest publicly accessible LLMs. Users can leverage the latest model under the MIT License.

    DeepSeek-V3-0324 has shown improvements, scoring between 5.3% and 19.8% higher on AI benchmarks compared to the previous V3 version. Its performance stands alongside other leading AIs, including GPT-4.5 and Claude Sonnet 3.7.

    Enhanced Capabilities

    The updated model brings several enhancements. One notable upgrade is its enhanced ability to design attractive web pages and user interfaces for online games. Additionally, its skills in searching, writing, and translating in Chinese have seen significant improvements.

    For those interested in testing the complete 685B DeepSeek-V3-0324, it is necessary to have at least 700 GB of available storage and several Nvidia A100/H100 GPUs. However, smaller and distilled versions of the model can be operated on a single GPU, like the Nvidia 3090 (available for purchase on Amazon).

    Source:
    Link

    DeepSeek Launches Enhanced V3 AI Model Just 3 Months After Release
    Tags: , ,
  • Amazon Introduces Interests AI for Personalized Shopping Tips

    Amazon Introduces Interests AI for Personalized Shopping Tips

    Key Takeaways

    1. Amazon has launched Interests AI, an assistant that finds products based on user-described interests.
    2. The AI uses a language model to convert detailed user descriptions into targeted product searches.
    3. Users can create and customize multiple prompts to refine their product searches.
    4. Interests AI operates 24/7, automatically notifying users of new products that match their criteria.
    5. Access to the Interests feature is currently available to a select group of US users, with broader availability expected soon.

    Amazon has introduced its Interests AI, a clever assistant that automatically looks for new products based on what users describe as their interests.

    How It Works

    The Interests AI relies on a large language model designed to convert descriptive user inputs into product searches. Users can narrow down the products by providing multi-sentence descriptions detailing what they want. This may include specific colors, price limits, and design preferences.

    Customizable Prompts

    Users can create multiple prompts, each of which can be refreshed with new specifications and descriptions of what they are searching for on Amazon. The AI is active around the clock, continuously searching for new products that align with the prompts and notifying users about the latest finds, significantly reducing the need for manual searches.

    Those who wish to explore the Amazon Interests AI can open the Amazon app on their smartphones or navigate to the Amazon website using their mobile browsers. By tapping the “Me” tab in the Amazon Shopping app, they can check if the “Interests” feature is available. A select group of US users will have immediate access, while others will gain access in the months to come.

    Amazon’s Announcement

    Amazon’s AI-driven ‘Interests’ function automatically discovers new products that align with your interests and hobbies.

    Source:
    Link

    Amazon Introduces Interests AI for Personalized Shopping Tips

     

    Tags: , ,
  • GPT-4o Can Generate Images with Near-Perfect Text

    GPT-4o Can Generate Images with Near-Perfect Text

    Key Takeaways

    1. GPT-4o now features image generation that allows for precise text rendering, improving the quality of visuals created from textual prompts.
    2. The model uses an interactive approach for dynamic image creation, enabling users to modify images step-by-step based on their ideas.
    3. Users can merge elements from different images and produce clear text, marking a significant advancement over previous AI image generation technologies.
    4. GPT-4o can handle 10-20 elements in a scene, surpassing competitors that typically manage only 5-8, making it easier to visualize complex ideas.
    5. Despite its advancements, limitations like bottom cropping, hallucinations, and challenges with non-Latin scripts still exist.

    OpenAI’s GPT-4o, which was introduced nearly a year ago, just received a significant update: it now includes image generation with incredibly precise text rendering. This new capability allows users to create intricate, high-quality visuals from textual prompts and engage in conversation to adjust these images until they align with their ideas—no more nonsensical signs or strange letters that earlier AI models produced.

    Dynamic Image Creation

    Unlike traditional methods of generating images by simply refining a single prompt, GPT-4o employs a more interactive technique. You begin with a straightforward request—like a cat—and then discuss modifications to capture your vision: perhaps adding a detective hat, a monocle, or any other detail you desire.

    Step-by-Step Modifications

    OpenAI provides examples that illustrate this process: users can construct and alter scenes incrementally, merging elements from various images into a single, unified result. The model excels at producing clear text on signs or items, a significant improvement over the distorted outputs of past AI image generation technologies.

    Impressive Capabilities

    Importantly, OpenAI acknowledges some selective showcasing—many images are labeled as “best of 2” or “best of 8″—but the outcomes remain impressive, particularly given the very user-friendly interface. GPT-4o can even start with your own photo and apply changes, managing 10-20 elements in a scene while competitors struggle with just 5-8. Just last week, I tried to recreate a scene from The Count of Monte Cristo, and it was quite challenging. Now, with GPT-4o’s image generation, not only will the images produced feature readable text, but it will also be significantly easier to turn your imagination into reality.

    Some Limitations

    However, it’s not without its flaws. OpenAI points out issues like bottom cropping, persistent hallucinations, difficulties with non-Latin scripts, and problems when exceeding 20 objects. Still, the capacity to create intricate, text-filled images using simple English distinguishes GPT-4o from its predecessors. If you’re working on a poster design, this tool offers the accuracy and flexibility that older models could only wish for.

    GPT-4o Can Generate Images with Near-Perfect TextGPT-4o Can Generate Images with Near-Perfect TextGPT-4o Can Generate Images with Near-Perfect TextGPT-4o Can Generate Images with Near-Perfect Text
    Tags: , ,
  • Simplify Your Vacation Planning with Google Search and Maps Updates

    Simplify Your Vacation Planning with Google Search and Maps Updates

    Key Takeaways

    1. AI Overviews in Search: Users can create detailed travel itineraries for specific destinations with suggestions for activities, images, and reviews, shareable via Google Docs or Maps.

    2. Integrating Screenshots in Maps: Users can now integrate screenshots of locations into Google Maps, which recognizes the places and allows users to save favorites and compile lists.

    3. Personalized Travel Guides with Gemini: The Gems feature lets users create a personal AI travel assistant for recommendations on destinations, dining, and more, along with a hotel price tracking feature.

    4. Ease of Access: New features are launching in the US for English-speaking users on both mobile and desktop platforms, enhancing the planning experience.

    5. Using Lens for Travel Insights: Travelers can use Google Lens to get detailed AI Overviews by simply pointing their camera at an object and asking questions, currently available for English-speaking users.

    Those who are wanting to take a summer vacation can now find more help from Google. A bunch of new features are here to help with planning trips, booking hotels, and exploring sights. These updates are part of Google Search, Google Maps, Lens, and Gemini. One of the key improvements is in AI Overviews, which can suggest trip ideas for various countries and regions, helping users to plan their days effectively while ensuring they don’t miss any attractions they want to see during their holidays.

    AI Overviews in Search

    In Search, AI Overviews can do more than just give a daily schedule; they can also provide suggestions on activities in a specific area or nation with an easy prompt like, “create an itinerary for Costa Rica with a focus on nature.” The output will feature images and reviews to help finalize the chosen spots. When finished, the itinerary can be shared via Google Docs, Gmail, or added to a Google Maps list. This new feature is launching this week in the US for English-speaking users on both mobile and desktop.

    Integrating Screenshots in Maps

    Trip planning usually means taking lots of notes and screenshots. To make this process smoother, Google has added a feature that lets users integrate screenshots into Maps, utilizing Gemini’s abilities. The places shown in the screenshots will be recognized, enabling users to save their favorites and compile a list. These locations will also appear on the map, but Maps will require access to photos for this feature. It will be available in the US on iOS next week, with Android support coming soon thereafter.

    Personalized Travel Guides with Gemini

    Moreover, the Gems feature from Gemini allows users to build their own AI expert at no cost, making it easy to have a travel assistant on their mobile devices at all times. With this tool, users can select a destination, discover dining options, receive recommendations, and much more. On top of that, Google’s hotel search is introducing a price tracking feature for selected dates and locations. This will be available this week on both mobile and desktop platforms.

    Lastly, using Lens, travelers can point their camera at an object and ask a question to receive an AI Overview filled with details and useful links. These overviews in Lens are currently available for English-speaking users.

    Google’s new features are aimed at making travel planning easier and more enjoyable for everyone.

    Source:
    Link

    Tags: , ,
  • Nvidia Project G-Assist: Voice-Controlled AI for Gaming Setup

    Nvidia Project G-Assist: Voice-Controlled AI for Gaming Setup

    Key Takeaways

    1. Project G-Assist AI helper enhances gaming and system settings for RTX desktop users using voice and text commands.
    2. Compatibility is limited to RTX 30, 40, and 50 series GPUs with 12GB or more VRAM.
    3. Activation can be done using the Alt+G shortcut, enabling features like GPU overclocking and game launching.
    4. The assistant may cause a slight temporary decrease in GPU performance while engaged.
    5. Future updates may include support for laptop GPUs and potentially for GPUs with less than 12GB VRAM.

    Nvidia’s Project G-Assist AI helper is now ready to use in the Nvidia app. Initially revealed at Computex 2024, this assistant aims to enhance gaming and system settings for RTX desktop users. It can apply overclocks, start games, and manage various peripheral settings, all done with simple voice and text commands. However, it’s worth noting that this feature is limited to RTX GPUs with 12GB or more VRAM and may slightly affect GPU performance, but this is just a temporary issue.

    Specific GPU Compatibility

    Currently, the Project G-Assist AI helper is exclusive to RTX 30, 40, and 50 series desktop GPUs. It operates using a third-party Small Language Model (SLM) that runs locally on your device. Due to its design, it is not suited for general conversations but is focused on specific tasks. Since it utilizes local GPU resources, users might experience a minor decrease in rendering rate or inference completion speed for a brief moment when the assistant is engaged, which means you could notice a slight drop in performance during gaming or running demanding applications.

    Activation and Functionality

    To activate G-Assist, you can use the Alt+G shortcut and request actions like optimizing graphics settings or monitoring temperatures. It supports a range of functions, including providing information on Nvidia tech, GPU overclocking, launching games, saving gameplay clips, checking for driver updates, and more. Additionally, the assistant has the ability to control peripheral and room lighting through a plugin, but this is only for compatible devices.

    There is a wide array of capabilities that this SLM already possesses, which can be explored further here.

    Future Developments

    At this time, G-Assist is limited to desktop RTX GPUs, with laptop GPU support expected in a future update. Nvidia has not indicated whether it will extend support to GPUs with less than 12GB VRAM, particularly since the new RTX 5060 and 5060 Ti versions will feature 8GB VRAM options.

    Source:
    Link

    Tags: , ,
  • Pixel Sense Enhances, But Doesn’t Replace Google Gemini Assistant

    Pixel Sense Enhances, But Doesn’t Replace Google Gemini Assistant

    Key Takeaways

    1. Google’s anticipated digital assistant, initially called ‘Pixie’, has been renamed to ‘Pixel Sense’ for its launch alongside the Pixel 10 series.
    2. The project has been split into two parts: one released as Pixel Screenshots, and the other potentially evolving into Pixel Sense.
    3. Pixel Sense is expected to enhance Gemini by offering additional functionalities and suggestions for Pixel users.
    4. There is uncertainty about whether Pixel Sense will be exclusive to the Pixel 10 series at launch, similar to Pixel Screenshots’ limitation to Pixel 9.
    5. Recent insights indicate a shift in the narrative surrounding Google’s digital assistant, with new developments emerging from Android Authority and 9to5Google.

    The narrative surrounding Google’s anticipated ‘Pixie’ digital assistant has shifted once again. Earlier this month, Android Authority reported finding new insights regarding what they termed ‘Google’s speculative new…[digital] assistant’.

    New Developments

    This report came on the heels of rumors from late 2023 and early 2024 indicating that Google planned to unveil a new digital assistant using the Pixel 9 and Pixel 9 Pro (curr. $849 on Amazon). However, Android Authority claimed that Google has renamed Pixie to ‘Pixel Sense’ in preparation for a broader launch later this year alongside the Pixel 10 series.

    Project Breakup

    Now, according to 9to5Google, these assertions might not be correct. It seems that Pixie has been divided into two separate projects, with one having already been released as Pixel Screenshots. The other initiative might evolve into Pixel Sense, aiming to enhance Gemini with additional functionalities.

    In essence, Pixel Sense is not likely to take the place of Gemini. Instead, 9to5Google anticipates it will help Pixel users by providing suggestions based on inputs from Google’s own apps. Unfortunately, it’s still unclear if Pixel Sense will be limited to the Pixel 10 series at its debut, much like Pixel Screenshots has been with the existing Pixel 9 smartphones.

    Source:
    Link

    Tags: , ,
  • Philips Hue App Update Introduces New AI Assistant Icon

    Philips Hue App Update Introduces New AI Assistant Icon

    Key Takeaways

    1. Philips Hue is set to introduce a new AI helper in its app, as revealed by Signify.
    2. The AI assistant button has replaced the search icon in the app’s latest update (version 5.38.1).
    3. The AI feature will allow users to create custom smart lighting scenes using voice and text commands.
    4. The app update does not mention the AI assistant and focuses on a QR code scanning feature instead.
    5. The release date for the AI assistant is currently unknown, but it may be linked to the upcoming Hue Secure doorbell.

    Philips Hue seems to be getting ready to introduce a new AI helper in its application. This new feature was revealed by Signify, the firm that runs the smart home brand, back in January. Along with this, the company also discussed a sound detection feature for smoke alarms and additional ways to access live video from a Secure camera.

    Update Brings New Features

    The AI assistant button has shown up in the Philips Hue application after the latest update to version 5.38.1. Users have noticed an icon for the assistant at the lower right corner of the app’s Home page. This new icon has taken the place of the magnifying glass icon, which was used to search for various scenes, devices, and rooms. However, when users click on the new icon, they are still directed to the same search feature.

    Expected Capabilities

    When the AI assistant was initially revealed, Signify mentioned that it would let users create smart lighting scenes tailored to their preferences. It will support both voice and text commands, and there will be a feedback option to enhance the lighting effects generated. This AI assistant is likely to be akin to the Govee AI Lighting Bot, which can be found in products like the Curtain Lights 2 (currently priced from $89.99 at Amazon).

    App Changes and Future Updates

    The change log for the Philips Hue app version 5.38.1 does not reference the AI assistant; it is the same as version 5.38.0, which was released a few days earlier and introduced a new QR code scanning feature. It looks like this update was made to eliminate the leaked Hue Secure doorbell, which is anticipated to be released later this year. It remains uncertain when the new AI assistant will officially be available in the Philips Hue app.

    Source:
    Link

    Tags: , ,
  • Ant Group Cuts Costs by 20% with AI Models on Chinese Chips

    Ant Group Cuts Costs by 20% with AI Models on Chinese Chips

    Key Takeaways

    1. Ant Group is using domestic chips from Alibaba and Huawei to reduce AI training costs by about 20%.
    2. Ant’s AI models, Ling-Plus and Ling-Lite, have outperformed Meta’s models in some tests, indicating a potential breakthrough for Chinese AI.
    3. Training costs for 1 trillion tokens have decreased from roughly $880,000 to around $700,000 with Ant’s optimized strategies.
    4. Ant plans to apply its AI models in healthcare and finance, enhancing services through acquisitions and existing applications.
    5. Both Ling models are open source, but they have fewer parameters compared to leading models like GPT-4.5, which presents challenges in stability during training.

    Jack Ma’s financial technology giant is using domestic chips from Alibaba and Huawei to develop AI models that seem to compete with Nvidia’s H800 GPUs.

    Cost-Effective AI Training

    Sources familiar with the situation have shared that Ant Group has discovered a method to train AI models on semiconductors made in China, reducing costs by approximately 20 percent compared to traditional techniques. While Ant still relies on Nvidia hardware for certain AI tasks, it is increasingly focusing on AMD processors and local alternatives for its newer models.

    Competitive Edge in AI

    According to a research paper released this month by Ant, its Ling-Plus and Ling-Lite models have outperformed Meta Platforms Inc. in some benchmark tests. If these results are verified, it could mark a significant breakthrough for Chinese AI, greatly lowering the costs associated with training and deploying AI solutions.

    The paper indicates that training 1 trillion tokens on high-performance hardware costs about 6.35 million yuan (roughly $880,000). However, with Ant’s optimized strategy and lower-spec gear, this expense reduces to about 5.1 million yuan (around $700,000). Tokens are the basic units of information that these models use to learn and create outputs.

    Future Applications

    Looking forward, Ant plans to apply these AI models in the fields of healthcare and finance. Earlier this year, the company purchased the Chinese online platform Haodf.com to enhance its AI services aimed at healthcare. Ant also runs an AI “life assistant” application named Zhixiaobao and a financial advisory AI tool called Maxiaocai.

    Both Ling models are open source: Ling-Lite has 16.8 billion parameters, while Ling-Plus boasts 290 billion. Although these numbers are substantial, they are still smaller than those of other prominent AI models—experts estimate that GPT-4.5 has around 1.8 trillion parameters, and DeepSeek-R1 is at 671 billion.

    Ant has recognized some challenges, especially with stability during the training phase. The research paper pointed out that minor adjustments in hardware or model design can lead to significant increases in error rates.

    Source:
    Link

    Tags: , ,
  • China’s PLA Boosts Support Ops with DeepSeek AI Technology

    China’s PLA Boosts Support Ops with DeepSeek AI Technology

    Key Takeaways

    1. DeepSeek’s AI systems are currently used in healthcare and logistics within the People’s Liberation Army (PLA), with potential future applications in military intelligence.
    2. The PLA employs DeepSeek’s open-source large language models (LLMs) in various military hospitals and support units, ensuring patient data remains secure.
    3. DeepSeek’s technology assists the People’s Armed Police in physical training and psychological support, allowing for initial testing before sensitive military applications.
    4. DeepSeek plays a significant role in military intelligence, enabling real-time analysis of battlefield information to improve situational awareness and strategic planning.
    5. Experts predict that continued integration of AI in the PLA could lead to direct combat applications, highlighting the importance of effective AI incorporation in military operations.

    DeepSeek’s AI systems are currently operational in hospitals and support units of the People’s Liberation Army, with experts predicting a future shift towards battlefield intelligence functions.

    Current Usage in Healthcare

    Reports from Chinese media indicate that the PLA is utilizing DeepSeek’s AI technology for support roles that do not involve combat. At this time, the focus is primarily on healthcare and logistics. However, defense analysts anticipate that it won’t take long before AI will be deployed in military intelligence and crucial decision-making processes.

    Implementation in Defense

    DeepSeek’s open-source large language models (LLMs) are being utilized in PLA hospitals, the People’s Armed Police, and various defense mobilization departments. The Central Theatre Command general hospital has recently mentioned its “embedded deployment” of the R1-70B LLM, which provides treatment suggestions while ensuring patient data remains on local servers. Additionally, Beijing’s renowned “301 Hospital,” which caters to high-ranking officials, has collaborated with Huawei to integrate the DeepSeek-R1 model into its operations.

    Training and Counseling Applications

    The People’s Armed Police also employs DeepSeek’s technology for physical training and psychological support initiatives. Sam Bresnick, a research fellow at Georgetown University’s Centre for Security and Emerging Technology, emphasizes that these limited applications allow the PLA to test the technology before applying it to more sensitive missions.

    A state-run outlet, Guangming Daily, underscored DeepSeek’s “increasingly vital role in military intelligentization,” highlighting its ability to quickly analyze real-time battlefield information for improved situational awareness during conflicts. The publication discussed how AI could be applied in amphibious landings, where it might adjust strategic plans in response to changing battlefield conditions.

    Outlook on Future Integration

    Chinese military expert Fu Qianshao stated that DeepSeek illustrates “the PLA’s dedication to staying current and fully leveraging AI technology to enhance overall combat capabilities.” He noted that the ongoing integration of AI in command systems might eventually lead to direct combat applications.

    Recently, the Nanjing National Defence Mobilisation Office published a user manual for DeepSeek, detailing its applications for emergency evacuation strategies, defense education, and resource evaluations. While many experts still consider the US to be ahead in AI advancements, Bresnick points out that the true advantage will go to whichever nation effectively incorporates AI into its military operations.

    Source:
    Link

    Tags: , ,
  • Tesla to Launch 5,000 Optimus Robots in 2025, Says Elon Musk

    Tesla to Launch 5,000 Optimus Robots in 2025, Says Elon Musk

    Key Takeaways

    1. Tesla aims to produce a “legion” of Optimus humanoid robots by the end of 2025, with an initial goal of 5,000 units this year.
    2. Elon Musk expressed optimism about reaching production goals, hoping for 10,000 to 12,000 components to create the robots.
    3. Musk referred to the production of 5,000 robots as the size of a Roman legion, highlighting the significance of this milestone.
    4. The Optimus robot has been in development since 2021 and was showcased in a prototype during the Cybercab reveal in October 2024.
    5. Musk considers the Optimus robot to be potentially Tesla’s most important product ever created.

    Tesla is a company with many different divisions, one of which focuses on creating robots. Elon Musk, the CEO, has stated that the company plans to astonish the world with a “legion” of Optimus humanoids before the end of 2025.

    Musk’s Announcement

    The billionaire made this statement during an interactive meeting with staff, which was broadcasted live, allowing the public to prepare for the initial production of the two-legged robot. He mentioned that Tesla aims to gather enough components to produce up to 12,000 units this year, but he would be satisfied with 5,000 Optimuses, or a legion.

    Production Goals

    “This year, we’re hopeful to produce around 5,000 Optimus robots. We’re technically shooting for enough parts to create 10,000, maybe even 12,000, but since it’s a completely new product with entirely new everything, I’ll say we’ll consider it a success if we reach half of the 10,000 goal.

    “Even 5,000 robots, that’s about the size of a Roman legion, which is kind of a creepy thought. Just imagining a whole legion of robots, I’ll be like ‘whoa.’ But I really think we will actually create at least one legion of robots this year, and then perhaps 10 legions next year. It’s a cool way to measure it, you know? Units of legion. So maybe around 50,000 next year.”

    Musk’s Optimism

    Musk is always positive about his company’s innovations. He has called the bipedal robot likely the most significant product ever created.

    The Optimus has been in the works since its introduction in 2021, initially presented with an actor designed to resemble a robot. Tesla showcased a prototype the next year but emphasized it during the Cybercab reveal in October 2024, providing a sneak peek of how it interacts with full-sized robots.

    Source:
    Link

    Tags: , ,