Meta’s Fundamental AI Research (FAIR) team has announced the public release of several advanced AI models aimed at accelerating innovation and collaboration in the AI community. Over the past decade, Meta has committed to advancing AI through open research, and this release continues that tradition, emphasising the importance of responsible AI development.
One notable model, Chameleon, can simultaneously process and generate both text and images. This mixed-modal capability allows it to handle inputs and outputs in various combinations of text and images, opening up possibilities such as creating creative captions or new scenes based on combined text and image prompts.
Another significant innovation is the multi-token prediction model, which enhances the efficiency of large language models (LLMs). Traditional LLMs predict the next word one at a time, but this new approach trains models to predict multiple future words simultaneously.
This method promises faster and more effective language generation, with pretrained models for code completion available under a research-only license.
images courtesy of Meta
In the realm of music generation, Meta also introduced JASCO, a model that provides greater control over AI-generated music by accepting diverse inputs like chords or beats in addition to text prompts. This multi-input capability allows for more nuanced and versatile music creation.
To address the growing concern of detecting AI-generated speech, Meta has unveiled AudioSeal, an audio watermarking technique designed for localised detection of AI-produced speech segments. AudioSeal offers a significant speed advantage over traditional methods, making it suitable for real-time and large-scale applications. This tool is available under a commercial license and represents Meta’s effort to mitigate the misuse of generative AI technologies.
Meta also focused on improving diversity in text-to-image generation systems. By developing automatic indicators and conducting a large-scale annotation study, Meta aims to ensure that AI-generated images better reflect global geographical and cultural diversity.
Why is this important?
This latest news from Meta follows just after Apple’s recent announcement on its investment in AI tools ‘Apple Intelligence‘ and its introduction of AI training at its Apple Developer Academy.
Through these releases, Meta again demonstrates its commitment to advancing AI technology responsibly while promoting collaboration and innovation within the global AI community. The availability of these cutting-edge models is expected to inspire further research and development, ultimately contributing to the field’s growth and responsible application.