Llama didn’t just save Meta; it made AI meta again. Llama goes beyond being a metaphor for CEO Mark Zuckerberg’s AI vision, who’s meeting founders alongside real llamas and teaching his daughter about multimodal companions.
The year 2024 would have been boring without Llama. The Llama ecosystem experienced rapid growth, with models reaching nearly 350 million downloads on Hugging Face by early this year—a remarkable 10x increase since its launch in 2023.
For Meta, open source is at the heart of innovation, enabling startups and enterprises to build advanced solutions on-premises and in the cloud. “I think, in the next generation of computing, we’re going to return to a zone where the open ecosystem is the leading one again,” Zuckerberg said in an interview.
In an open letter before the launch of Llama 3.2, Zuckerberg said open-source AI will surpass closed models—just like Linux overtook Unix—and become the foundation for future AI. Developers and founders in India and across the world have embraced the Llama models to build new products and companies.
Leaders like Mukesh Ambani and Jensen Huang have praised Llama for driving widespread AI innovation.
In August this year, Llama 3.2 was launched to improve edge AI and vision tasks, providing vision LLMs (11B and 90B) and lightweight models (1B and 3B) optimised for on-device use, with strong support for Qualcomm and MediaTek hardware. This was followed by the quantised versions of Llama 3.2 to enhance on-device AI performance with up to four times faster inference speeds, a 56% model size reduction, and a 41% decrease in memory usage.
Going Beyond Llama
Meta’s Movie Gen, a model that generates and edits videos and creates sound effects—all using simple text prompts—stole the show in 2024. This video-generation tool is a direct competitor to Sora and Runway. Though not available to the public yet, its use cases have gained widespread attention.
Another AI tool led us to Spirit LM, which integrates speech and text for natural interactions, while MarDini, developed with KAUST, was also noteworthy. This tool enables high-quality, flexible video creation.
Google’s NotebookLM alternative NotebookLlama, was also timely. This open-source tool based on Meta’s Llama models, provides easy-to-use summaries from text files.
That’s not all,CoTracker 3 by Meta improves video tracking by handling objects that move out of view, using a method called ‘pseudo labelling’ to enhance accuracy without needing fully labelled data. Meanwhile, AI Studio makes AI tools accessible to more people, encouraging creativity and interaction across Meta’s apps.
Meta Goes MEGA
Zuckerberg announced that pre-training for Llama 4 has begun, supported by advanced compute clusters and data infrastructure, marking a leap beyond Llama 3. VP Ragavan Srinivasan also echoed that next-gen Llama models, set for 2025, will feature improved memory, context, cross-modality, integrations, coding, and hardware.
Under Meta AI chief Yann LeCun‘s leadership, the company is advancing AI with systems like Layer Skip and V-JEPA to improve reasoning and interaction, alongside self-supervised learning to ultimately achieve artificial machine intelligence (AMI).
Robotics and embodied AI are also going to be a huge theme, with technologies that improve how humans interact with robots and support autonomous machine intelligence. Meta made strides in this with Meta Sparsh, Meta Digit 360, and Meta Digit Plexus – which enhance robot touch and dexterity.
Meta’s projects in the Metaverse—like mixed reality and smart glasses—rely on advanced AI models. Soon, Meta will integrate AR/VR into Llama models to create accessible, everyday platforms. The focus for 2025 will remain towards building on open-source and decentralised models.
“There should be a lot of different AIs and AI services out there, not just one singular AI. Every person on our platforms can create their own AI agents that they want to interact with,” said Zuckerberg, emphasising his distinct vision compared to companies like OpenAI and Google, which are focused on developing unified AI systems like ChatGPT and Gemini for a range of tasks.
As next year will be defined by agents, Meta envisions billions of personalised AI agents—changing customer support, community engagement, and personal productivity. Zuckerberg believes the goal should be for every small business and creator to have their own AI agent, as essential and accessible as email or social media today.
The post Imagine Meta Without Llama appeared first on Analytics India Magazine.
