Close Menu
Technology Mag

    Subscribe to Updates

    Get the latest creative news from FooBar about art, design and business.

    What's Hot

    It’s a great time to pick up a Kindle, Echo speaker, and other Amazon devices

    July 11, 2025

    You can buy Tapo’s terrific robot vacuum for just $200 right now

    July 11, 2025

    Tornado Cash Made Crypto Anonymous. Now One of Its Creators Faces Trial

    July 11, 2025
    Facebook X (Twitter) Instagram
    Subscribe
    Technology Mag
    Facebook X (Twitter) Instagram YouTube
    • Home
    • News
    • Business
    • Games
    • Gear
    • Reviews
    • Science
    • Security
    • Trending
    • Press Release
    Technology Mag
    Home » A New Kind of AI Model Lets Data Owners Take Control
    Business

    A New Kind of AI Model Lets Data Owners Take Control

    News RoomBy News RoomJuly 11, 20253 Mins Read
    Facebook Twitter Pinterest LinkedIn Reddit WhatsApp Email

    A new kind of large language model, developed by researchers at the Allen Institute for AI (Ai2), makes it possible to control how training data is used even after a model has been built.

    The new model, called FlexOlmo, could challenge the current industry paradigm of big artificial intelligence companies slurping up data from the web, books, and other sources—often with little regard for ownership—and then owning the resulting models entirely. Once data is baked into an AI model today, extracting it from that model is a bit like trying to recover the eggs from a finished cake.

    “Conventionally, your data is either in or out,” says Ali Farhadi, CEO of Ai2, based in Seattle, Washington. “Once I train on that data, you lose control. And you have no way out, unless you force me to go through another multi-million-dollar round of training.”

    Ai2’s avant-garde approach divides up training so that data owners can exert control. Those who want to contribute data to a FlexOlmo model can do so by first copying a publicly shared model known as the “anchor.” They then train a second model using their own data, combine the result with the anchor model, and contribute the result back to whoever is building the third and final model.

    Contributing in this way means that the data itself never has to be handed over. And because of how the data owner’s model is merged with the final one, it is possible to extract the data later on. A magazine publisher might, for instance, contribute text from its archive of articles to a model but later remove the sub-model trained on that data if there is a legal dispute or if the company objects to how a model is being used.

    “The training is completely asynchronous,” says Sewon Min, a research scientist at Ai2 who led the technical work. “Data owners do not have to coordinate, and the training can be done completely independently.”

    The FlexOlmo model architecture is what’s known as a “mixture of experts,” a popular design that is normally used to simultaneously combine several sub-models into a bigger, more capable one. A key innovation from Ai2 is a way of merging sub-models that were trained independently. This is achieved using a new scheme for representing the values in a model so that its abilities can be merged with others when the final combined model is run.

    To test the approach, the FlexOlmo researchers created a dataset they call Flexmix from proprietary sources including books and websites. They used the FlexOlmo design to build a model with 37 billion parameters, about a tenth of the size of the largest open source model from Meta. They then compared their model to several others. They found that it outperformed any individual model on all tasks and also scored 10 percent better at common benchmarks than two other approaches for merging independently trained models.

    The result is a way to have your cake—and get your eggs back, too. “You could just opt out of the system without any major damage and inference time,” Farhadi says. “It’s a whole new way of thinking about how to train these models.”

    Share. Facebook Twitter Pinterest LinkedIn WhatsApp Reddit Email
    Previous ArticlePrime Day Ends Tonight. We Have Nearly 300 Last-Chance Deals So You Can Save
    Next Article HMD is ‘scaling back’ in the US, killing Nokia all over again

    Related Posts

    Tornado Cash Made Crypto Anonymous. Now One of Its Creators Faces Trial

    July 11, 2025

    Linda Yaccarino Tried to Tame X. Now She’s Out as CEO

    July 10, 2025

    ‘People Are Going to Die’: A Malnutrition Crisis Looms in the Wake of USAID Cuts

    July 10, 2025

    Grok Is Spewing Antisemitic Garbage on X

    July 9, 2025

    OpenAI Poaches 4 High-Ranking Engineers From Tesla, xAI, and Meta

    July 9, 2025

    Volodymyr Zelensky’s Clothing Has Sparked a Polymarket Rebellion

    July 9, 2025
    Our Picks

    You can buy Tapo’s terrific robot vacuum for just $200 right now

    July 11, 2025

    Tornado Cash Made Crypto Anonymous. Now One of Its Creators Faces Trial

    July 11, 2025

    The smartphone-sized Boox Palma 2 is cheaper than ever for Prime Day

    July 11, 2025

    It’s the final day of Prime Day 2025, and the deals are still hot

    July 11, 2025
    • Facebook
    • Twitter
    • Pinterest
    • Instagram
    • YouTube
    • Vimeo
    Don't Miss
    Security

    Android May Soon Warn You About Fake Cell Towers

    By News RoomJuly 11, 2025

    In recent years, North Korea has deployed thousands of so-called IT workers to infiltrate Western…

    The Verge’s favorite travel gear for 2025

    July 11, 2025

    Prefer Walmart to Amazon? Check Out These Competing Prime Day Deals

    July 11, 2025

    Some of the stuff we actually bought during Prime Day

    July 11, 2025
    Facebook X (Twitter) Instagram Pinterest
    • Privacy Policy
    • Terms of use
    • Advertise
    • Contact
    © 2025 Technology Mag. All Rights Reserved.

    Type above and press Enter to search. Press Esc to cancel.