Category: Artificial intelligence

  • OpenAI Unveils Enhanced O3-Mini AI with Free ChatGPT Access

    OpenAI Unveils Enhanced O3-Mini AI with Free ChatGPT Access

    OpenAI has introduced its newest AI large language model, the o3-mini, which is quicker and more efficient than its earlier version, the o1-mini, in delivering precise responses. This model marks the first small reasoning LLM that is freely available for all ChatGPT account holders starting today.

    Three Reasoning Levels

    The o3-mini model offers three levels of reasoning efforts: low, medium, and high. Users with a free ChatGPT account can access o3-mini at the medium reasoning level without any cost, while those with paid accounts have the option to select either the low or high reasoning levels. Subscribers of ChatGPT Plus, Team, and Pro can utilize o3-mini right away, but Enterprise users will need to wait until February. For developers eager to build popular apps utilizing the OpenAI o3-mini API, guidance is available in a book on Amazon.

    Performance Comparison

    Overall, o3-mini in medium or high reasoning modes excels beyond o1-mini according to standardized AI benchmarks, especially in tasks that require reasoning. In high reasoning mode, o3-mini even surpasses the larger o1 model on certain benchmarks; however, it doesn’t have o1’s capability to interpret images or handle visual data.

    Speed and Efficiency

    Interestingly, o3-mini delivers responses 24% faster, which equates to approximately 2.46 seconds quicker than o1-mini. This improvement not only decreases the waiting time for ChatGPT users but also lessens the carbon footprint associated with running o3-mini. While it is more efficient, hackers might find o3-mini disappointing, as its potential for being utilized in cybersecurity attacks has been significantly limited.

    OpenAI news release, OpenAI o3-mini system card

    Source:
    Link


  • UC Berkeley Researchers Replicate DeepSeek AI for Just $30

    UC Berkeley Researchers Replicate DeepSeek AI for Just $30

    AI research has long been dominated by large tech firms with substantial funding. But a team from UC Berkeley has changed the narrative. They successfully replicated the main features of DeepSeek R1-Zero for a mere $30 (yes, you read that right). Their initiative, named TinyZero, shows that sophisticated AI reasoning models can be created without hefty expenses. Plus, AI research is becoming increasingly available to everyone.

    The Team’s Ambition

    Under the leadership of Jiayi Pan, the researchers set out to recreate DeepSeek’s reasoning framework using reinforcement learning (RL). Rather than depending on costly cloud services or enormous computational resources, they trained TinyZero using just a basic language model, a straightforward prompt, and a reward system. Pan expressed his enthusiasm on X (previously known as Twitter), stating, “You can experience the ‘Aha’ moment yourself for < $30.” He also mentioned that TinyZero represents the first open reproduction of reasoning models, emphasizing its capability to verify and enhance its own responses.

    Development Process

    In order to evaluate the model, the researchers utilized a game called Countdown, where players must attain a target number through basic mathematical operations. Initially, TinyZero made random guesses, but over time, it learned to confirm its answers, seek improved solutions, and adjust its strategies. They experimented with various model sizes, ranging from 500 million parameters to 7 billion parameters. The findings were intriguing: smaller models (0.5B parameters) merely guessed answers and halted, while larger models (1.5B+ parameters) began to self-verify, refine their responses, and notably enhance accuracy.

    Impressive Affordability

    What really sets TinyZero apart is its low cost compared to conventional AI models. Here’s a look at the expenses:
    – OpenAI’s API: $15 per million tokens
    – DeepSeek-R1: $0.55 per million tokens
    – TinyZero’s total cost: $30—one-time training expense

    This accessibility means that anyone—not just large tech corporations—can explore AI reasoning models without financial strain.

    Open for Exploration

    TinyZero is open-source and can be found on GitHub, making it possible for anyone to experiment with it. While it has only been tested in the Countdown game, Pan aspires for this project to broaden the reach of reinforcement learning research. However, he acknowledged that it’s still in the early stages, stating, “One caveat, of course, is that it’s validated only in the Countdown task but not the general reasoning domain.” Yet, even with this limitation, the implications are significant: AI development need not be costly. With initiatives like TinyZero, affordable and open-source AI might represent the future.

    Source:
    Link


  • TinyZero: Affordable DeepSeek AI Clone Developed for $30

    TinyZero: Affordable DeepSeek AI Clone Developed for $30

    While it’s quite difficult to confirm the supposed development expenses of this new AI model, even the hours put in by PhD students are likely worth much more than a mere $30. Nevertheless, this recent narrative demonstrates that the upheavals in the AI industry weren’t merely a result of exaggerated reactions. It’s worth noting that some AI models for personal use that require minimal computing power have been around for a while, even prior to the emergence of DeepSeek.

    The Power of DeepSeek

    Clearly, DeepSeek operates with a significantly larger database, whereas the researchers at the University of Berkeley zero in on “reinforcement learning.” Their software doesn’t rely on human-generated data, as the AI can only confirm its own findings.

    Self-Verification Process

    This self-checking mechanism can produce extensive chains of reasoning, but it does necessitate a certain amount of processing time. This strategy is effective for straightforward mathematical problems and programming tasks, due to the simplicity of the verification involved in these scenarios.

    The AI adjusts its methods gradually to arrive at the right answer with the least number of steps possible. Users can access this AI on GitHub, where the source code and usage examples are available.

    Open Access to Data

    The creators have also made available all the data that was used in this programming trial. Since the model doesn’t incorporate extra data, TinyZero AI stands out as a very streamlined AI model. However, it is expected to perform comparably to more advanced models in specific tasks, such as solving a mathematical puzzle game.

    Regardless of whether this AI was truly developed for just $30, once foundational technologies are made publicly accessible, there will always be individuals who can enhance or refine them. After all, today’s smartphones require only a small fraction of the computing power that early 2000s desktop PCs needed and still outperform them. In the realm of AI, this progress appears to be accelerating even more rapidly.

    Source:
    Link

  • Most Galaxy AI Features May Stay Free Forever, But There’s a Catch

    Most Galaxy AI Features May Stay Free Forever, But There’s a Catch

    Most of the updates in the Galaxy S25 lineup are focused on Galaxy AI. There were rumors that Samsung might start charging for Galaxy AI features in 2026. However, tipster PandaFlash X suggests that key features like Photo Assist, AI Instant Slow-mo, and Writing Tools will likely remain free.

    Free Features for Now

    Initially, it was stated that these features would be “free until 2025,” but the new information does not specify a time limit, implying that they could be free indefinitely. This seems logical when you look at the improvements: Galaxy AI along with only minor hardware changes.

    Pricing and Subscription Details

    We are uncertain about how much Samsung will charge for Galaxy AI or how its plans will be structured. However, we do know that the Gemini Advanced subscription, which is complimentary for the first six months for Galaxy S25 series users, costs $20 per month. The tipster also noted that features within Gemini Advanced will require payment.

    There has been no clarity on which specific features will rely on the Gemini Advanced subscription.

    What’s Next After Six Months?

    What will you have to pay for or lose after the six-month free trial? Samsung has been just as vague about this as it was regarding the Galaxy S25 Edge. They highlighted the AI functionalities and the improvements in user experience that the S25 series offers, but did not detail which functions will need the Gemini Advanced subscription.

    The VP of ‘Google Gemini Experiences’ mentioned at the Galaxy Unpacked event, “Gemini Advanced comes with our most capable AI models and priority access to the newest features. It will also include ‘Screen Sharing’ and ‘live video viewing’ in the future.”

    Most of the AI features, like summarization, leverage the latest model available. Even if you don’t subscribe, these features will still work but with somewhat lower quality results.

    According to the Google Store, Gemini Advanced is particularly good at solving complex issues due to its improved reasoning skills. Some of the more challenging queries may not perform as effectively after the first six months, or they might cease to work entirely. It’s unfortunate that Samsung left so many questions open for speculation on the internet instead of clarifying them at the launch.


  • Google Launches Gemini 2.0 Flash: Faster and More Efficient AI

    Google Launches Gemini 2.0 Flash: Faster and More Efficient AI

    Google has rolled out Gemini 2.0 Flash as the new standard model for its Gemini application, offering enhancements in both speed and efficiency. This update, which follows its launch in December, boosts performance in crucial areas, making activities like brainstorming, studying, and writing much quicker and smoother. Thanks to faster response times and improved processing, users can effortlessly generate ideas, dive into concepts, and produce written content.

    Features for Advanced Subscribers

    For those who have subscribed to Gemini Advanced, the 1 million token context window is still available, enabling document uploads of up to 1,500 pages. Subscribers also keep access to unique features like Deep Research, which aids in discovering insights and relationships, as well as Gems, which delivers curated resources for more streamlined research.

    Enhanced Image Generation

    Moreover, the image generation now employs Imagen 3, which delivers more intricate textures and precise interpretations of prompts. The updated model excels at handling fine details, realistic elements, and stylistic choices, simplifying the creation of photorealistic images and imaginative artwork.

    Gemini 2.0 Flash is now accessible on both web and mobile platforms. To aid in the transition, Google will still offer support for the earlier models, Gemini 1.5 Flash and 1.5 Pro, for a limited duration, allowing users to finish any ongoing chats.

    Microsoft Updates

    In other news, Microsoft is launching distilled versions of DeepSeek R1 to Windows 11 Copilot+ PCs, enabling AI functionalities to operate efficiently on-device without requiring an internet connection. This change enhances privacy and performance by keeping data processing local. The rollout will commence with Snapdragon X-powered devices from manufacturers such as Asus, Lenovo, Dell, HP, and Acer, and will soon include Intel Core Ultra 200V and AMD Ryzen AI 300 series.

    Source:
    Link


  • Host Your Own AI Image Generator with Invoke AI & Stable Diffusion

    Host Your Own AI Image Generator with Invoke AI & Stable Diffusion

    There are plenty of reasons you might consider setting up your own AI image generator. You may want to skip watermarks and ads, create multiple images without paying for a subscription, or even explore image generation in ways that might not align with the ethical guidelines of the service. By hosting your own instance and utilizing training data from companies like Stable Diffusion, you can maintain complete control over what your AI produces.

    Getting Started

    To kick things off, download the Invoke AI community edition from the provided link. For Windows users, most of the installation is now automatic, so all necessary dependencies should install smoothly. However, you might encounter some challenges if you’re using Linux or macOS. For our tests, we used a virtual machine running Windows 11, with 8 cores from a Ryzen 9 5950, an RTX 4070 (available on Amazon), and 24GB of RAM on a 1TB NVMe SSD. While AMD GPUs are supported, that’s only for Linux systems.

    Once the installation is complete, open Invoke AI to generate the configuration files, and then close it. This step is essential as you’ll need to modify some system settings to enable “Low-VRAM mode.”

    Configuring Low-VRAM Mode

    Invoke AI doesn’t clearly define what “low VRAM” means, but it’s likely that the 12GB RAM on the RTX 4070 won’t be efficient enough for a 24GB model. To adjust this, you should edit the invokeai.yaml file located in the installation directory using a text editor and add the following line:

    “`
    enable_partial_loading: true
    “`

    After this change, Windows users with Nvidia GPUs need to adjust the CUDA – Sysmem Fallback Policy to “Prefer No Sysmem Fallback” within the Nvidia control panel global settings. You can tweak the cache amount you want for VRAM, but most users will find that simply enabling “Low-VRAM mode” is sufficient to get things running.

    Downloading Models

    Some models, like Dreamshaper and CyberRealistic, can be downloaded right away. However, to access Stable Diffusion, you’ll need to create a Hugging Face account and generate a token for Invoke AI to pull the model. There are also options to add models via URL, local path, or folder scanning. To create the token, click on your avatar in the top right corner and select “Access Tokens.” You can name the token whatever you prefer, but make sure to grant access to the following:

    Once you have the token, copy and paste it into the Hugging Face section of the models tab. You might have to confirm access on the website. There’s no need to sign up for updates, and Invoke AI will notify you if you need to allow access.

    Be aware that some models can take up a significant amount of storage, with Stable Diffusion 3.9 requiring around 19 GB.

    Accessing the Interface

    If everything is set up correctly, you should be ready to start. Access the interface through a web browser on the host machine by navigating to http://127.0.0.1:9090. You can also make this accessible to other devices on your local network.

    In the “canvas” tab, you can enter a text prompt to generate an image. Just below that, you can adjust the resolution for the image; keep in mind that higher resolutions will take longer to process. Alternatively, you can create at a lower resolution and use an upscale tool later. Below that, you can choose which model to use. Among the four models tested—Juggernaut XL, Dreamshaper 8, CyberRealistic v4.8, and Stable Diffusion 3.5 (Large)—Stable Diffusion created the most photorealistic images, yet struggled with text prompts, while the others offered visuals resembling game cut scenes.

    The choice of model really comes down to which one delivers the best results for your needs. While Stable Diffusion was the slowest, taking about 30 to 50 seconds per image, its results were arguably the most realistic and satisfying among the four.

    Exploring More Features

    There’s still a lot to explore with Invoke AI. This tool lets you modify parts of an image, create iterations, refine visuals, and build workflows. You don’t need high-end hardware to run it; the Windows version can operate on any 10xx series Nvidia GPU or newer, though expect slower image generation. Despite mixed opinions on AI model training and the associated energy use, running AI on your own hardware is an excellent means to create royalty-free images for various applications.

    Source:
    Link


  • Irish DPA Targets DeepSeek Over Data Practices Concerns

    Irish DPA Targets DeepSeek Over Data Practices Concerns

    Founded in May 2023, DeepSeek is working on a large language model that claims to be way more efficient than those from OpenAI or Meta. Now, the company is facing scrutiny from European regulators. While the European Commission hasn’t gotten involved yet, the Chinese AI firm must explain how its software operates to Irish officials. These authorities are worried about how DeepSeek manages the personal information of users in Ireland.

    Concerns from Irish Authorities

    According to a report from the Irish Data Protection Commission (DPC), cited by Reuters,
    “The Data Protection Commission (DPC) has written to DeepSeek requesting information on the data processing conducted in relation to data subjects in Ireland.”

    Currently, there’s no fixed timeline for DeepSeek to respond, and no public details about potential penalties the company could face if they fail to reply or if their response indicates non-compliance with European and Irish data laws.

    Launch of DeepSeek-R1 Chatbot

    DeepSeek recently launched its first chatbot app, based on the DeepSeek-R1 model, earlier this month. The app is free and available on both iOS and Android platforms, and it has quickly become a rival to ChatGPT. It was the most downloaded free app on the US App Store shortly after its release. One immediate effect of this launch was a significant drop in Nvidia shares, which fell by 18%.

    Open-Source Data Accessibility

    All data connected to DeepSeek-R1, including its generative AI algorithms, is open-source, allowing anyone to develop tailored solutions at a low cost. Nevertheless, it’s crucial that DeepSeek ensures its personal data handling is transparent enough to facilitate easier access to the European market.

    Source:
    Link

  • Why DeepSeek Stands Out Among AI Models

    Why DeepSeek Stands Out Among AI Models

    The AI sector has been largely led by American firms such as OpenAI, Google, and Meta for quite some time. Nevertheless, the rise of DeepSeek, a new AI startup from China, has changed the international AI scene.

    DeepSeek-R1 Model Breakthrough

    DeepSeek’s newest product, the DeepSeek-R1, is gaining attention due to its impressive performance, open-source framework, and affordable development costs. As artificial intelligence continues to play a vital role in tech advancements, it’s important to recognize how DeepSeek sets itself apart from other prominent models.

    Open-Source Advantages

    Unlike U.S. AI leaders like OpenAI, DeepSeek embraces an open-source strategy. By offering its DeepSeek-R1 model under an MIT license, it allows researchers, developers, and businesses to freely access, alter, and use the technology. In contrast, OpenAI has shifted away from its original commitment to open-source, choosing to keep its newer GPT models closed off. This open-source model promotes higher transparency, encourages cooperative enhancements, and reduces the obstacles to AI integration, making DeepSeek an appealing choice for companies and independent developers who wish to avoid being confined to proprietary systems.

  • Free Up to 7GB iPhone Storage by Disabling Apple Intelligence

    Free Up to 7GB iPhone Storage by Disabling Apple Intelligence

    With the launch of iOS 18.3, iPadOS 18.3, and macOS 15.3, Apple has introduced its newest AI-driven feature, Apple Intelligence, which is now automatically active on devices that are compatible. This collection of AI capabilities includes features like message summaries, image creation, and improved Siri interactions, but it also requires a substantial amount of storage. According to Apple’s official documentation, Apple Intelligence can take up to 7GB of space on iPhones, iPads, and Macs.

    Managing Storage Needs

    For users facing storage limitations or those who may not find the AI features beneficial, there is a way to completely disable Apple Intelligence. Once you turn it off, the local AI models will be deleted from the device, freeing up the space they occupied. This is particularly helpful for iPhone users who have limited storage, especially for those who don’t actively utilize AI-generated features or content.

    Steps to Disable Apple Intelligence

    To turn off Apple Intelligence, users can go to Settings on an iPhone or iPad, or System Settings on a Mac, and find the “Apple Intelligence & Siri” section. By toggling off Apple Intelligence, a confirmation prompt will appear, after which the system will erase AI-related resources. Some AI capabilities, like Writing Tools, Genmoji, and Image Playground, will no longer be accessible. However, it’s worth noting that certain tools, like the Clean Up function in the Photos app, might still be available.

    Apple Intelligence is unavailable in mainland China at this time, and the company has not shared a timeline for when it might be launched there. Additionally, only devices with Apple’s newest hardware, including the iPhone 15 Pro, iPhone 16 series, and M-series iPads and Macs, can support Apple Intelligence. Users upgrading to iOS 18.2 must also opt-in to access the AI features.

    Customizing Your Experience

    For those who are hesitant about fully disabling Apple Intelligence, Apple allows users to selectively turn off specific features through app settings. Options like message summaries in notifications and writing assistance tools can be disabled without shutting down the entire AI suite. By managing these settings, users can customize their experience while still maintaining control over the storage and functionality of their devices.

  • Windows 11 Copilot+ PCs to Feature Distilled DeepSeek R1 Models

    Windows 11 Copilot+ PCs to Feature Distilled DeepSeek R1 Models

    Microsoft is set to introduce a simplified version of its DeepSeek R1 AI models to Windows 11 Copilot+ PCs. This announcement follows the recent launch of DeepSeek R1 on Azure AI Foundry, which is Microsoft’s platform aimed at developers to promote innovation through AI tools and machine learning models in a secure and responsible way. Initially, this integration will be available on laptops, tablets, and PCs powered by Snapdragon X, with support for Lunar Lake and AMD Ryzen AI 300 processors to be added later.

    New Optimized Versions

    In a blog post on its Windows Developer Blog, Microsoft revealed that the Neural Processing Unit (NPU) optimized versions of DeepSeek-R1 would be introduced to Copilot+ PCs. These versions will be included in lightweight laptops from leading brands like Asus, Lenovo, Dell, HP, and Acer, all of which offer devices equipped with Snapdragon X Plus and X Elite processors featuring 45 TOPS NPUs. It’s worth noting that all Copilot+ PCs come with NPUs that have 40+ TOPS.

    Developer Access and Future Releases

    The initial release, named DeepSeek-R1-Distill-Qwen-1.5B, will be accessible to developers through the AI Toolkit, with the 7B and 14B versions expected to follow soon after. Microsoft claims that these optimized models will allow developers to create and deploy AI-driven applications that run effectively on devices, leveraging the powerful NPUs found in Copilot+ PCs.

    Timeline and AMD Updates

    Although Microsoft has not provided a specific timeline for the rollout, it appears that Intel Core Ultra 200V devices will receive these enhancements sooner than their AMD Ryzen AI equivalents. Meanwhile, AMD has already released guides related to DeepSeek R1 Distill for its newest Ryzen AI 300 series APUs, along with information for Ryzen 8040/7040 chips, even though the latter only have 16/10 TOPS NPUs, respectively.

    Source:
    Link