Meta’s Leap into the Future: Introducing Llama 3.2 with Celebrity Voices and Visual AI Capabilities

To look at this article again, go to My Profile, and then click on Saved stories to see it

Meta Unveils Llama 3.2, Now Featuring Celebrity Voices

Today, Mark Zuckerberg revealed that Meta, the company that evolved from a social media platform to a metaverse and AI giant, is enhancing its AI helpers by integrating a variety of famous voices, such as Dame Judi Dench's and John Cena's. However, a significant advancement for Meta's future goals is the newly developed capability of its AI to analyze users' images and visual data.

Today, Meta unveiled Llama 3.2, marking the debut of its open-source AI models equipped with visual capabilities. This advancement expands their applicability across robotics, virtual reality, and AI-based agents. Additionally, certain iterations of Llama 3.2 are specifically tailored for mobile platforms, enabling the development of AI-driven applications that can operate on smartphones. These applications can leverage the phone's camera or monitor the device's screen to autonomously interact with other apps for the user.

"Zuckerberg announced at Connect, a Meta event in California today, that this marks our initial venture into open source, multimodal models, paving the way for numerous intriguing applications that necessitate an understanding of visuals."

With its extensive network spanning Facebook, Instagram, WhatsApp, and Messenger, Meta's enhancements to its assistant could introduce a wide audience to the latest wave of AI assistants, which are more advanced in speech and visual capabilities. Meta announced that its AI assistant, known as Meta AI, is currently utilized by over 180 million users on a weekly basis.

At the Connect event, Zuckerberg unveiled various new AI capabilities. He presented clips showcasing Ray Ban smart glasses equipped with Llama 3.2 technology offering culinary suggestions upon recognizing ingredients, as well as critiquing apparel displayed in a shop. Additionally, the head of Meta introduced a range of AI innovations in development. These advancements encompass a program facilitating real-time translation from Spanish to English, the capability to dub videos automatically in multiple languages, and a digital avatar designed for creators to respond to fan inquiries for them.

Recently, Meta has elevated the role of artificial intelligence within its applications, such as incorporating it into the search functions of Instagram and Messenger. Additionally, users can now enjoy celebrity voice features, with options including Awkwafina, Keegan Michael Key, and Kristen Bell.

Previously, Meta attempted to popularize text-based assistants by assigning them celebrity identities, but this strategy did not resonate with users. In July, the company introduced AI Studio, a feature allowing users to design chatbots tailored to any personality they prefer. Meta announced that these new personas will be accessible to users in the United States, Canada, Australia, and New Zealand within the coming month. Additionally, Meta's advancements in AI image technology will be launched in the United States, though the company has yet to disclose a timeline for availability in other regions.

The latest iteration of Meta AI is designed to offer insights and critiques on users' photographs. For instance, it can identify the type of bird in your photo if you're uncertain. Additionally, this version has the capability to modify pictures by adding elements or changing the background upon request. This follows Google's introduction of a comparable feature for its Pixel phones and Google Photos in April.

Enhancing the functionalities of Meta AI is an enhanced iteration of Llama, Meta's leading large language model. The announcement today of the model being available for free could significantly influence its widespread use, considering the extensive adoption of the Llama family by developers and startups thus far.

Unlike OpenAI's offerings, Llama is available for free local installation and use, albeit with certain limitations on extensive commercial deployment. Additionally, Llama offers a more straightforward process for enhancements or task-specific adjustments through extra training.

Patrick Wendell, a founding member and the vice president of engineering at Databricks, a firm known for hosting AI models such as Llama, mentions that numerous businesses are attracted to open-source models as they offer a more secure way to safeguard their proprietary data.

Big language models are progressively evolving to be "multimodal," signifying their training now encompasses handling images and audio in addition to text. This expansion of capabilities paves the way for creators to develop innovative AI applications, such as AI agents designed to perform valuable functions on computers for users. Llama 3.2 aims to simplify the process for developers in creating AI agents that, for example, can navigate the internet, potentially searching for discounts on specific products based on a brief description provided.

"Multimodal models have become crucial as the information utilized by individuals and companies isn't limited to text; it spans a variety of forms such as pictures, sound, or even unique types like protein chains or economic records," states Phillip Isola, an MIT professor. "Recently, we've transitioned from powerful language models to models that effectively handle images and audio too. With each passing year, these systems are gaining the capability to process an increasing range of data types."

"Meta's release of Llama 3.1 has demonstrated that open models are now on par with their proprietary equivalents," states Nathan Benaich, the founder and general partner at Air Street Capital, and a renowned author of an annual AI report. He further mentions that models capable of processing multiple modes of input generally surpass the capabilities of those limited to text. "I'm looking forward to the advancements in version 3.2," he expresses.

Today, the Allen Institute for AI (Ai2) based in Seattle unveiled a sophisticated open source multimodal model named Molmo. This model comes with a more permissive license compared to Llama, and Ai2 is making the training data details available. This move allows researchers and developers to explore and adapt the model further.

Meta announced today its plans to launch various versions of Llama 3.2, each differing in size and capability. In addition to releasing two higher-end versions featuring 11 billion and 90 billion parameters, indicators of both the model's complexity and scale, Meta will also introduce more modest models with 1 billion and 3 billion parameters. These smaller versions are specifically tailored for efficient performance on mobile devices, having been fine-tuned for ARM-based processors made by Qualcomm and MediaTek.

Meta's introduction of advanced AI technologies occurs amid intense competition among major tech firms, all aiming to present the most sophisticated AI solutions. By choosing to make its highly valuable models available at no cost, Meta could secure a competitive advantage, laying the groundwork for numerous AI-based applications and services. This strategy is particularly timely as businesses start to delve into the possibilities offered by AI agents.

Check Out These Recommendations…

Delivered to your email: A selection of the finest and most unusual tales from the archives of WIRED

Elon Musk poses a threat to national security

Discussion: Meredith Whittaker Aims to Disprove Capitalist Ideals

What's the solution for a challenge like Polestar?

Event: Don't miss out on The Grand Conversation happening on December 3 in San Francisco

Additional Content from WIRED

Insights and Tutorials

© 2024 Condé Nast. All rights are protected. WIRED might receive a share of revenue from items bought via our website, as a result of our Affiliate Agreements with sellers. Any content from this site cannot be copied, shared, broadcast, stored, or utilized in any form without explicit consent from Condé Nast. Ad Choices

Choose a global website