Meta has announced the release of four new AI models and other research findings on Meta FAIR, part of its commitment to fostering an open ecosystem. These releases aim to inspire community innovation and advance AI in a responsible manner.
Highlights of the new AI models are as follows:
Meta Chameleon: This series includes 7B and 34B-sized language models that support mixed-mode inputs and pure text outputs, bringing new possibilities to the field of natural language processing.
Meta Multi-token Prediction: An innovative pre-trained language model that, through multi-token prediction technology, can predict multiple future words simultaneously, significantly enhancing the model's capabilities, training efficiency, and prediction speed.
Meta JASCO: A generative text-to-music model that accepts various conditional inputs, offering higher controllability and introducing new innovative tools to the field of music creation.
Meta AudioSeal: An audio watermark model designed specifically for detecting AI-generated speech content, supporting commercial licensing use.
In addition to these models, Meta has also released other Responsible AI (RAI) achievements, including research, data, and code aimed at measuring and improving geographical and cultural preferences and diversity in AI systems.
Meta emphasizes that the most advanced AI should be accessible to everyone, not just a few large tech companies. The company is eager to see how the community will leverage these technologies.