Friday, July 12, 2024
- Advertisement -
More

    Latest Posts

    Meta Publicly Releases Five AI Models

    Meta’s Fundamental AI Research Team (FAIR) has publicly released five AI models, including image-to-text and text-to-music generation models, a multi-token prediction model, and a technique for detecting AI-generated speech. Announced through an official company blog post on June 18, Meta hopes to aid AI research through this move.

    “For more than a decade, Meta’s Fundamental AI Research (FAIR) team has focused on advancing the state of the art in AI through open research. As the field rapidly innovates, we believe collaboration with the global AI community is more important than ever,” said the company. “By publicly sharing this research, we hope to inspire iterations and ultimately help advance AI in a responsible way.”

    The models released are:

    • Chameleon: The company is releasing key components of its Chameleon models under a research-only license. Chameleon is a family of mixed-modal models that can understand and generate images and text, taking any combination of text and images as input and providing any combination of both as an output.
    • Multi-token prediction models:  Multi-token prediction models predict multiple future words, making them significantly more efficient than most LLMs, which can predict only the next word. The company is releasing pre-trained AI models, to be used for code completion on research-only non-commercial licence.
    • JASCO: JASCO is a music-generative AI that accepts text and musical input to produce musical output. 
    • Audio Seal: To reduce AI-generated misinformation, Meta is releasing AudioSeal, which they believe is the first audio watermarking technique designed specifically for localized detection of AI-generated speech. Released under a commercial license, AudioSeal pinpoints AI-generated segments within longer audio snippets. Unlike traditional complex decoding methods, AudioSeal’s localized approach enables up to 485 times faster detection speed, suiting it for large-scale and real-time applications. 
    • Geographic disparities in AI: Meta developed automatic indicators to evaluate potential geographical disparities in text-to-image models to ensure their outputs reflect global diversity. Meta is releasing the geographic disparities evaluation code and annotations to help the community improve diversity across generative models.

    In its previous comments to the US government, Meta had advocated for minimal restrictions on open-source AI, suggesting that responsible development of all AI models is key to balancing the benefits of open foundation models with their marginal risks. In its yearly earnings report, Meta stated that scaling its AI offerings into profitable businesses may require multi-year investment cycles, but it was optimistic about the results.

    Also Read:

    The post Meta Publicly Releases Five AI Models appeared first on MEDIANAMA.

    Latest Posts

    - Advertisement -

    Don't Miss

    Stay in touch

    To be updated with all the latest news, offers and special announcements.