Meta Unveils Suite of Advanced AI Models and Tools, Emphasizing Open-Source Collaboration

Meta Unveils New AI Models and Tools to Drive Innovation

Collaborations: Meta partners with filmmakers like Blumhouse to refine Movie Gen tools. Meta, the owner of Facebook, announced on Friday that it was releasing a batch of new AI (Artificial Intelligence) models from its research division, including a "Self-Taught Evaluator," which could reduce the need for human involvement in the AI development process. Meta's Fundamental AI Research (FAIR) team introduced a series of new AI models and tools aimed at advancing machine intelligence (AMI). Notable releases include Meta Segment Anything (SAM) 2.1, an updated model designed for improved image segmentation, and Meta Spirit LM, a multimodal language model that blends text and speech for natural-sounding interactions. Meta claims that Meta Spirit LM is its first open-source multimodal language model that freely mixes text and speech. Also Read: Jio Showcases AI Tools, Industry 5.0 and More Innovations at IMC2024 Other innovations include Layer Skip, a solution that accelerates large language models (LLMs) generation times on new data, and SALSA, a tool for testing post-quantum cryptography. Meta also released Meta Open Materials 2024, a dataset for AI-driven materials discovery, along with Meta Lingua, a streamlined platform for efficient AI model training. Meta Open Materials 2024 provides open-source models and data based on 100 million training examples, offering an open-source option for the materials discovery and AI research community. The Self-Taught Evaluator is a new method for generating synthetic preference data to train reward models without relying on human annotations. Reportedly, Meta's researchers used entirely AI-generated data to train the evaluator model, eliminating the need for human input at that stage. "As Mark Zuckerberg noted in a recent open letter, open source AI "has more potential than any other modern technology to increase human productivity, creativity, and quality of life," all while accelerating economic growth and advancing groundbreaking medical and scientific research," Meta said on October 18. Also Read: OpenAI Raises USD 6.6 Billion to Accelerate AI Research and Expansion Earlier, on October 4, Meta introduced Movie Gen, a suite of AI models capable of generating 1080p videos and audio from simple text prompts. These models generate HD videos, personalised content, and precise edits, outperforming similar industry tools, according to Meta. Movie Gen also supports audio syncing to visuals. While still in development, Meta is collaborating with filmmakers to refine the tool, which could have future applications in social media and creative content. "Our first wave of generative AI work started with the Make-A-Scene series of models that enabled the creation of image, audio, video, and 3D animation. With the advent of diffusion models, we had a second wave of work with Llama Image foundation models, which enabled higher quality generation of images and video, as well as image editing. Movie Gen is our third wave, combining all of these modalities and enabling further fine-grained control for the people who use the models in a way that's never before been possible," Meta said. Movie Gen has four key capabilities: video generation, personalised video generation, precise video editing, and audio generation. Meta says that these models are trained on a combination of licensed and publicly available datasets. Meta says it continues to improve these models, which are designed to enhance creativity in ways people might never have imagined. For instance, users could animate a "day in the life" video for Reels or create a personalised animated birthday greeting for a friend to send via WhatsApp, all using simple text prompts. Also Read: Meta AI Expands to 21 New Countries, Including the UK and Brazil Continuing, on October 17, Meta announced that, as part of a pilot program, it is collaborating with Blumhouse and other filmmakers to test the tool before its public release. According to the company, early feedback suggests that Movie Gen could help creatives quickly explore visual and audio ideas, though it is not intended to replace hands-on filmmaking. Meta plans to use feedback from this program to refine the tool ahead of its full launch. "While we're not planning to incorporate Movie Gen models into any public products until next year, Meta feels it's important to have an open and early dialogue with the creative community about how it can be the most useful tool for creativity and ensure its responsible use," says Connor Hayes, VP of GenAI at Meta. "These are going to be powerful tools for directors, and it's important to engage the creative industry in their development to make sure they're best suited for the job," added Jason Blum, founder and CEO of Blumhouse. Meta is extending the Movie Gen pilot into 2025 to continue developing the models and user interfaces. In addition to collaborating with partners in the entertainment industry, Meta plans to work with digital-first content creators, the company said.

New Meta AI Models released - SAM 2.1 Spirit LM, MEXMA and More

Meta has unveiled a suite of innovative AI models and datasets, signaling a significant leap forward in machine intelligence. Developed by Meta's AI Research (FAIR) team, these innovations are set to transform various fields, from speech processing to material science. By open-sourcing many of these models, Meta is fostering a collaborative ecosystem that enables researchers and developers worldwide to push the boundaries of AI technology. At the heart of this initiative is a commitment to open-source collaboration, a move that provides widespread access to access to innovative technology and enables a diverse community to tackle some of the world's most pressing challenges. Whether it's enhancing video editing capabilities with SAM 2.1 or transforming language processing with Metal Lingua, these models are designed to inspire and enable. The Segment Anything Model 2.1 (SAM 2.1) represents a substantial upgrade in object tracking capabilities for video content. This advanced model enhances segmentation precision, allowing for more accurate and reliable object tracking across frames. By making SAM 2.1 open-source, Meta enables developers to integrate this powerful technology into a wide array of applications. Key applications of SAM 2.1 include: Meta Spirit LM stands out as a new model that seamlessly integrates speech-to-speech and text processing. This innovative approach preserves the nuances and expressiveness of speech while converting it into text, opening up new possibilities for communication technologies. While currently under a non-commercial research license, Meta Spirit LM has vast potential for: The Layer Skip Technique introduces a novel approach to enhancing the performance of large language models. By selectively using layers for response generation and verification, this technique significantly accelerates processing times without compromising accuracy. Benefits of the Layer Skip Technique: The Salsa Project addresses critical vulnerabilities in post-quantum cryptography, focusing on strengthening systems against increasingly sophisticated AI-based attacks. As the landscape of digital threats evolves, the Salsa Project's contributions are essential for maintaining robust protection of sensitive data and critical infrastructure. Key focus areas of the Salsa Project include: Metal Lingua offers a framework for training large language models with unprecedented efficiency. By significantly reducing computational demands, this framework enables organizations to achieve substantial performance gains without the need for extensive hardware resources. Advantages of Metal Lingua: Met Open Materials 2024 provides a comprehensive suite of open-source models and datasets for inorganic materials discovery. This initiative accelerates research in material science, potentially leading to breakthroughs in various technological fields. Potential impacts of Met Open Materials 2024: Maxima is an advanced crosslingual sentence encoder that significantly enhances translation accuracy. By employing sophisticated token-level objectives, it offers improved multilingual representation, making it an invaluable tool for global communication and content localization. Applications of Maxima include: The Self-Thought Evaluator is a new model for generating high-quality synthetic preference data used in training reward models. This innovative approach delivers faster processing times and outperforms larger models in evaluation tasks, making it a valuable asset in AI research and development. Key benefits of the Self-Thought Evaluator: Meta's decision to open-source these models, or at least provide open weights, underscores its commitment to advancing AI research on a global scale. By facilitating non-commercial research and local experimentation, Meta enables the AI community to explore new frontiers in machine intelligence collectively. This collaborative approach is crucial for: As these new Meta AI models and techniques are adopted and refined by the global research community, we can expect to see rapid advancements across various sectors, from healthcare and education to environmental conservation and beyond. Meta's contributions represent not just technological progress, but a step towards a more collaborative and innovative future in AI research and development.