Latest news with #LLaMA
Fast Company
19-05-2025
- Science
- Fast Company
‘LegoGPT' designs Lego models with nothing but a prompt
Researchers at Carnegie Mellon University (CMU) have just answered a question that's probably occurred to Lego fans for decades: What if I could instantly turn any idea into a Lego set? In a paper titled 'Generating Physically Stable and Buildable LEGO Designs from Text,' published last week, six coauthors lay out an invention they're calling 'LegoGPT.' This generative AI model can take a text-based prompt, like 'an acoustic guitar with an hourglass shape,' and determine all of the necessary Lego pieces needed to build that structure and how to assemble them. The LegoGPT demo and code is publicly available through the study, meaning that Lego hobbyists are free to try it out at home. Although outputs are currently limited to around 20 categories (including basic items like chairs, guitars, boats, trains, and cars), the researchers are working to expand the model's capabilities into more complicated categories. Ultimately, they think a LegoGPT-type tool might serve as the basis for a variety of real-world tasks in architecture and product design. How LegoGPT predicts its next block LegoGPT is a fine-tuned version of Meta's LLaMA-3.2-Instruct-1B language learning model, which you can think of as an open source ChatGPT. To teach the model how to make Lego structures, researchers trained it using a database of 47,000 Lego structures and 28,000 unique 3D shapes, each with their own descriptive captions. Based on that vast swath of designs, LegoGPT is able to predict how to build a hypothetical object using only a text prompt. To do that, LegoGPT uses something called an autoregressive model, which is common among the most popular generative AI platforms. 'ChatGPT and Llama are autoregressive models because, given the string of words that they've already outputted, they want to predict the next word,' explains Ava Pun, one of the study's coauthors and a PhD student at CMU. 'So if you ask, 'What is the weather,' and it predicts 'The weather today is,' then it will try to predict the next word: sunny, rainy, and so on. With Lego GPT, instead of predicting the next word, it wants to predict the next brick.' Once LegoGPT has created a 3D model it thinks will work, the LLM needs a way to make sure that the structure will actually be stable. According to Pun, that proved tricky, considering that existing simulators aren't trained to understand the physics of a Lego brick. So, the CMU team built their own physics algorithm for LegoGPT to check its work. 'We developed a customized physics reasoning algorithm that accounts for all the physical forces that the bricks experience: for example, the downward force due to gravity, friction forces, and contact forces from the other bricks that they're touching,' Pun says. 'The algorithm constructs a force model for the structure and then evaluates the forces over the entire structure. If these physical forces sum to zero, that means the structure will not move around.' LegoGPT automatically uses this algorithm to ensure that it's found a viable solution. If any of the block it's chosen is causing the model to turn out wobbly, the model will continue iterating until it lands on a new version that passes the test. A future real-world application So far, researchers have used LegoGPT to create a range of structures, including vintage cars, steamships, and an electric guitar. Currently, the model only works on a 20x20x20 voxel grid, though Pun says the team is already planning on adding more brick types to the model's database and expanding the grid resolution. For Lego fans who want to play around at home, the study's demo, available through a public portal, can turn simple prompts into a buildable 3D Lego model and a list of necessary parts. Because LegoGPT isn't made to be Lego-builder-facing, it doesn't produce step-by-step instructions, meaning the main challenge will be figuring out how to arrange the component parts in the right order. Pun says her team used Lego brick assembly to test AI's 3D-building capabilities because of the blocks' accessibility. Eventually, though, they believe this concept could be applied to real-world scenarios, like helping architects draft buildings or designing custom furniture from a predefined set of parts. 'Today's generative AIs can't offer that—you can generate a cool image or video of a chair, but the model doesn't know how these things can be made in the real world,' Pun says. 'We wanted to address this challenge by integrating physical laws and assembly constraints into generative models and creating objects that function in reality.'
Yahoo
18-05-2025
- Science
- Yahoo
AI Chatbots Are Becoming Even Worse At Summarizing Data
Ask the CEO of any AI startup, and you'll probably get an earful about the tech's potential to "transform work," or "revolutionize the way we access knowledge." Really, there's no shortage of promises that AI is only getting smarter — which we're told will speed up the rate of scientific breakthroughs, streamline medical testing, and breed a new kind of scholarship. But according to a new study published in the Royal Society, as many as 73 percent of seemingly reliable answers from AI chatbots could actually be inaccurate. The collaborative research paper looked at nearly 5,000 large language model (LLM) summaries of scientific studies by ten widely used chatbots, including ChatGPT-4o, ChatGPT-4.5, DeepSeek, and LLaMA 3.3 70B. It found that, even when explicitly goaded into providing the right facts, AI answers lacked key details at a rate of five times that of human-written scientific summaries. "When summarizing scientific texts, LLMs may omit details that limit the scope of research conclusions, leading to generalizations of results broader than warranted by the original study," the researchers wrote. Alarmingly, the LLMs' rate of error was found to increase the newer the chatbot was — the exact opposite of what AI industry leaders have been promising us. This is in addition to a correlation between an LLM's tendency to overgeneralize with how widely used it is, "posing a significant risk of large-scale misinterpretations of research findings," according to the study's authors. For example, use of the two ChatGPT models listed in the study doubled from 13 to 26 percent among US teens between 2023 and 2025. Though the older ChatGPT-4 Turbo was roughly 2.6 times more likely to omit key details compared to their original texts, the newer ChatGPT-4o models were nine times as likely. This tendency was also found in Meta's LLaMA 3.3 70B, which was 36.4 times more likely to overgeneralize compared to older versions. The job of synthesizing huge swaths of data into just a few sentences is a tricky one. Though it comes pretty easily to fully-grown humans, it's a really complicated process to program into a chatbot. While the human brain can instinctively learn broad lessons from specific experiences — like touching a hot stove — complex nuances make it difficult for chatbots to know what facts to focus on. A human quickly understands that stoves can burn while refrigerators do not, but an LLM might reason that all kitchen appliances get hot, unless otherwise told. Expand that metaphor out a bit to the scientific world, and it gets complicated fast. But summarizing is also time-consuming for humans; the researchers list clinical medical settings as one area where LLM summaries could have a huge impact on work. It goes the other way, too, though: in clinical work, details are extremely important, and even the tiniest omission can compound into a life-changing disaster. This makes it all the more troubling that LLMs are being shoehorned into every possible workspace, from high school homework to pharmacies to mechanical engineering — despite a growing body of work showing widespread accuracy problems inherent to AI. However, there were some important drawbacks to their findings, the scientists pointed out. For one, the prompts fed to LLMs can have a significant impact on the answer it spits out. Whether this affects LLM summaries of scientific papers is unknown, suggesting a future avenue for research. Regardless, the trendlines are clear. Unless AI developers can set their new LLMs on the right path, you'll just have to keep relying on humble human bloggers to summarize scientific reports for you (wink). More on AI: Senators Demand Safety Records from AI Chatbot Apps as Controversy Grows
Alalam24
07-05-2025
- Business
- Alalam24
An Unexpected Move by Meta Changes the Rules of Artificial Intelligence
Meta, the social media giant, has launched its first standalone application powered by intelligent AI assistance, in a clear move to compete with platforms like ChatGPT by providing users with direct access to generative AI models. Mark Zuckerberg, the company's founder and CEO, announced the launch in a video on Instagram, noting that over one billion users are already interacting with the 'Meta AI' system across the company's various apps. The new release comes in the form of a standalone app, offering users a personalized and direct experience. Zuckerberg explained that the app is designed to serve as a personal assistant for each user, relying primarily on voice interaction and tailoring responses to individual interests. Initially, the app uses minimal contextual information, but over time—and with user consent—it will be able to learn more about users' habits and social circles through Meta's connected apps. The AI is based on the open-source generative model 'LLaMA,' which has garnered significant attention from developers and has been downloaded over a billion times, making it one of the most widely used models in its category. The app features a design aligned with Meta's social nature, allowing users to share AI-generated posts and view them in a personalized feed. It's powered by a newer version of the model known as 'LLaMA 4,' which brings more personalized and flexible interactions. Users can also choose to save shared information to avoid repeating it in future conversations. Additionally, the app offers the ability to search within Facebook and Instagram content—provided prior permission is granted. This app serves as an alternative to the 'Meta View' app used with Ray-Ban Meta smart glasses, enabling seamless interaction across glasses, mobile, and desktop platforms through a unified interface. The launch comes at a time when major tech companies are racing to release intelligent assistants aimed directly at users, with OpenAI still leading the market through the ongoing development of ChatGPT and its continuous integration of advanced features.
Time Business News
02-05-2025
- Business
- Time Business News
Exploring DeepSeek: The Future of Open-Source AI in Enterprise Search and Beyond
Artificial Intelligence (AI) has rapidly evolved from research labs to real-world applications across every industry. Among the latest innovations making headlines in the open-source community is DeepSeek, an advanced AI language model that blends the power of large language models (LLMs) with enterprise search capabilities. Developed by DeepSeek AI, this model promises to revolutionize how businesses interact with their internal data, drive productivity, and implement intelligent automation. So, what makes DeepSeek unique, and why should businesses and tech professionals keep it on their radar? Let's delve deeper. DeepSeek is a cutting-edge, open-source large language model (LLM) designed to compete with industry giants like OpenAI's GPT models, Google's Gemini, and Meta's LLaMA. Created by DeepSeek AI, a China-based startup, DeepSeek offers two flagship models—DeepSeek-V2 and DeepSeek-Coder—that cater to general-purpose text generation and code-related tasks respectively. These models are trained on massive multilingual and code datasets, enabling them to perform a wide range of tasks including document summarization, code generation, language translation, and enterprise data search. DeepSeek's primary objective is to offer a free, powerful alternative to proprietary AI tools—enabling greater transparency, adaptability, and innovation. Open-Source and Community-Centric One of DeepSeek's most appealing features is its open-source availability. Developers and businesses can freely access, customize, and deploy DeepSeek's models, promoting innovation across borders. This democratization of AI technology significantly reduces dependence on paid, closed-source models. Multilingual and Multimodal Capabilities DeepSeek's models are trained on 2 trillion tokens, incorporating both natural language and code from diverse sources. This allows the models to support multiple languages, making them a great fit for global applications. Moreover, DeepSeek is continuously evolving toward multimodal integration, such as text-to-image or voice-to-text functionalities. Tailored for Enterprise Search and Knowledge Management Unlike generic LLMs, DeepSeek is optimized for enterprise-level data retrieval. It can understand and search through complex datasets, documents, and knowledge bases—providing meaningful, context-aware answers that can enhance workflows and decision-making. Competitive Performance In several benchmark tests, DeepSeek's models have outperformed other open-source competitors like Mistral and Falcon, and even rivaled GPT-3.5 in some areas. Its code-focused variant, DeepSeek-Coder, demonstrates exceptional skill in generating, refactoring, and debugging code snippets across multiple programming languages. The versatility of DeepSeek makes it ideal for a range of industries and use cases: Enterprise Knowledge Base Search: Businesses can integrate DeepSeek into their internal systems to create smart knowledge management solutions that allow employees to quickly find relevant information from policy documents, technical guides, and databases. Businesses can integrate DeepSeek into their internal systems to create smart knowledge management solutions that allow employees to quickly find relevant information from policy documents, technical guides, and databases. Customer Support Automation: DeepSeek can power intelligent chatbots capable of handling complex queries by pulling data from various internal sources, enhancing customer experience while reducing human workload. DeepSeek can power intelligent chatbots capable of handling complex queries by pulling data from various internal sources, enhancing customer experience while reducing human workload. Software Development: Developers can leverage DeepSeek-Coder for code generation, translation between programming languages, and real-time debugging suggestions—speeding up software development cycles. Developers can leverage for code generation, translation between programming languages, and real-time debugging suggestions—speeding up software development cycles. Research and Content Creation: Writers, analysts, and marketers can use DeepSeek to generate content drafts, summarize research papers, and create multilingual reports, improving productivity across creative and academic domains. In an era where digital monopolies dominate the AI landscape, open-source LLMs like DeepSeek play a crucial role in ensuring accessibility, affordability, and innovation. They empower startups, researchers, and small businesses to compete on a level playing field without being locked into expensive ecosystems. Moreover, open-source AI models are auditable, allowing users to inspect how decisions are made—an essential feature for industries that demand transparency, such as finance, healthcare, and legal sectors. DeepSeek AI continues to evolve its technology with the aim of developing a multimodal general AI agent, capable of seamlessly understanding and generating text, image, and audio content. Its recent model updates include Mixture of Experts (MoE) architecture, which enables the AI to activate only a subset of model parameters per task—improving performance while reducing resource consumption. The DeepSeek team is also actively encouraging contributions from the global open-source community, inviting developers and researchers to fine-tune, extend, and deploy the model for unique use cases. As the world leans increasingly into AI-powered solutions, DeepSeek emerges as a promising, open-source alternative that combines flexibility, enterprise-readiness, and innovation. Whether you're a developer building a smart application, a business looking to streamline internal processes, or a tech enthusiast exploring AI's next frontier—DeepSeek is a project worth watching. By placing the power of advanced language models into the hands of the public, DeepSeek is helping shape a more open, intelligent, and collaborative digital future. Want to integrate AI into your business? At Infowind Technologies, we specialize in building custom AI solutions that align with your strategic goals. From LLM integration to AI-powered chatbots and enterprise automation tools, our expert team is here to help you lead in the age of intelligence. TIME BUSINESS NEWS
Yahoo
10-04-2025
- Business
- Yahoo
Why META Platforms Stock is a Smart Buy for Black Swan Portfolios
Meta Platforms (META) might still be best known as the social media giant behind Facebook, Instagram, and WhatsApp, but it's quickly evolving into something far more significant. What's flying under the radar for many is its transformation into a leader in AI and hardware. Despite the wide-ranging stock plunges in all sectors, I remain bullish on META. Discover outperforming stocks and invest smarter with Top Smart Score Stocks. Filter, analyze, and streamline your search for investment opportunities using Tipranks' Stock Screener. Yes, the recent sell-off has rattled some investors—especially those unprepared for sudden macro shocks. META stock has swandived from around $700 in mid-February to ~$500 per share today. But there's a lesson here. Build your portfolio with resilience in mind to survive and thrive through unexpected turmoil. If you've kept some cash on hand or have been dollar cost averaging all along, now is the time to lean in and start putting money to work in Meta. The AI train hasn't derailed—it's just pausing before accelerating again. Meta's ascent in artificial intelligence is nothing short of extraordinary. At the heart of this surge are two major initiatives. First, there's LLaMA, the company's open-source large language model. Then there's Meta AI, the user-facing assistant already integrated into its core apps. LLaMA was only released in 2023, yet by early 2025, it had surpassed one billion downloads. That kind of scale is staggering. By open-sourcing the models, Meta isn't just participating in the AI race—it's changing how the game is played. Instead of keeping its technology behind paywalls, Meta embeds AI within its vast social media ecosystem, accelerating internal development and third-party innovation. AI is also quietly reshaping the user experience. More than half of the content consumed on Instagram has now surfaced via AI-driven recommendations. On Facebook, it's about 30%. This translates into more intelligent ad targeting and, ultimately, more substantial revenue growth as marketers follow user engagement. But what's even more interesting is the efficiencies Meta is gaining. By automating internal processes, the company is starting to chip away at its operating expenses. For a tech-heavy business like Meta, those savings can scale fast. Despite short-term macro headwinds and global trade fears, the AI foundation Meta is building remains undervalued. The stock's price-to-earnings ratio currently sits around 21—well below its historical average. That's not a reflection of performance–it's skepticism–and for forward-looking investors, it could be an opportunity. With the stock now trading below its 50-week moving average, those with a risk-on appetite may want to look closer. Meta's ad platform is getting a significant boost from its AI capabilities. The Advantage+ campaign system, driven by machine learning, has more than doubled revenue. Meanwhile, the company's Reality Labs division—though currently a drag on margins—is laying the groundwork for what could be a monumental hardware breakthrough. Between Ray-Ban smart glasses and the upcoming Orion AR headset, Meta is edging toward a potential iPhone moment of its own. Still, it's not without cost. Reality Labs posted a $17.7 billion operating loss in 2024. Yet Meta is forging ahead, budgeting between $60 billion and $65 billion for capital expenditures in 2025. That's a hefty jump from the $39 billion it spent in 2024. The good news? META can afford it. The company generated over $52 billion in free cash flow last year and holds nearly $78 billion in cash on its balance sheet—against just $29 billion in debt. What's more, profitability is already on the rebound. Operating margins climbed back to roughly 42% in 2024, up from a low of 29% two years ago. Even so, the market hasn't caught up. Meta continues to trade well below its decade-long average P/E of 32. That means the business's full earnings power and AI-led potential simply aren't priced in—yet. When the market gets nervous, I stay calm. That's because I run a black swan portfolio strategy designed to handle exactly these kinds of events. Going into the recent volatility, I had about 30% of my portfolio in cash. A diversified portfolio, including an appropriate cash allocation, allows me to absorb sudden shocks without resorting to panic. The term 'black swan' comes from Nassim Taleb, who described it as an unpredictable, high-impact event that people try to explain in hindsight. Think of the 2008 financial crisis, 9/11, COVID-19, or today's spiraling trade war. You can't see them coming. But you can prepare. The easiest way to protect yourself is by holding cash or cash equivalents. During the recent market sell-off, I trimmed my cash position from 30% to 22.5%, reinvesting in undervalued names like Nvidia (NVDA). Meta is also now firmly on my radar as a company poised to lead a global economy increasingly driven by AI and robotics. If we enter a deeper recession, I plan to reduce my cash holding to ~15%. If things get worse—say, a full-scale Taiwan crisis that crashes the S&P 500 (SPX) by 90%—I plan to deploy everything. The upside from buying into that kind of drop when the economy recovers could be remarkable. Wall Street is super bullish on META stock. Meta has earned a consensus Strong Buy rating, with 42 analysts recommending Buy and only three recommending Hold. Only one analyst is bearish on META. The average Meta Platforms price target is $748 per share, suggesting a 46.5% upside from current prices. Even amid geopolitical and economic uncertainty, that kind of bullishness speaks volumes. When markets become rough and volatile, the impulse to sell and retreat into cash is understandable—but usually misplaced. The smart money already had cash or hedges in place before the trouble started. The goal isn't just to survive market shocks but to come out stronger–the epitome of resilience and anti-fragility. Meta is one of the best-positioned stocks in this new AI-driven world, and there may not be many better chances to establish a long META position at the discounts offered by a frightened market. Disclaimer & DisclosureReport an Issue
