Meta AI
Overview
Meta AI is the artificial intelligence research and product division developed and operated by Meta Platforms, Inc. It began as the Facebook AI Research (FAIR) lab in 2013 and significantly expanded its AI capabilities after the company rebranded to Meta in 2021. Key technologies include the large language model (LLM) Llama series, the image segmentation model SAM (Segment Anything Model), and AI assistants and recommendation systems integrated into social media and the metaverse. Meta AI emphasizes an open-source strategy, releasing the Llama model for free to researchers and developers to expand its ecosystem.
Main Content
1. Llama Model
Llama is a large language model developed by Meta, with the first version (Llama 1) released in February 2023. In July 2024, Llama 3.1 405B was released, achieving the largest scale among open-source LLMs (405 billion parameters). The Llama model is used for various tasks such as text generation, code writing, translation, and summarization, and is provided under a commercially available license. Notably, Llama 3.1 demonstrates performance competitive with closed models like GPT-4o and Claude 3.5 Sonnet, and is considered a milestone for open-source AI.
2. AI Assistant: Meta AI Assistant
In September 2023, Meta launched the AI assistant 'Meta AI' integrated into its social media platforms (Facebook, Instagram, WhatsApp, Messenger). This assistant is based on the Llama model and provides answers to user questions, image generation (Imagine feature), search, and creative support. In 2024, voice interaction and real-time translation features were added, and it is also available on the metaverse platform Horizon Worlds.
3. Image and Vision AI
- SAM (Segment Anything Model): An image segmentation model released in April 2023 that can accurately segment objects in an image using only prompts. SAM 2 (July 2024) supports video segmentation as well.
- DINOv2: A self-supervised learning-based vision model that excels at image feature extraction.
- Emu Video: A model that generates videos from text prompts (November 2023).
4. Recommendation Systems and Advertising AI
Meta's core revenue source, the advertising recommendation system, heavily relies on AI. The deep learning-based personalized recommendation engine analyzes user behavior, interests, and interactions to optimize feeds, Reels, and ads. In 2024, Meta introduced the next-generation recommendation system 'Andromeda', improving processing speed and accuracy.
5. AI Research and Open-Source Ecosystem
The Meta AI Research lab (FAIR) collaborates with academia to publish papers, datasets, and code. Key contributions include PyTorch (deep learning framework), FAISS (similarity search library), and MMS (multilingual speech recognition). In 2024, Meta formed the 'AI Alliance' with IBM, AMD, and others to advance open-source AI.
6. Metaverse and AI
Meta is integrating AI into the metaverse (Horizon Worlds, Quest VR). Technologies such as AI-based avatar creation, natural language interfaces, and object recognition and interaction in virtual spaces are being developed. In 2024, the 'Meta AI Sandbox' was introduced to allow creators to build virtual experiences using AI tools.
Latest Trends
Key trends for Meta AI in 2024–2025 include:
- Strengthening Open-Source Leadership: The release of Llama 3.1 405B raised the performance ceiling for open-source LLMs, with Llama 4 expected in early 2025. Meta maintains that 'open-source AI will become the standard.'
- Multimodal Expansion: Focus on developing models that integrate text, image, video, and audio. Multimodal research results such as 'ImageBind' and 'AnyMAL' were announced in December 2024.
- In-House AI Chip Development: In 2024, Meta announced its own AI inference chip 'MTIA v2', with mass production targeted for 2025. This strategy aims to reduce dependence on NVIDIA and lower costs.
- Generative AI Advertising Tools: In October 2024, the 'AI-generated ad images and text' feature was officially launched, allowing advertisers to create materials using AI.
- Regulation and Ethics: In response to the EU AI Act, Meta introduced an 'AI labeling system' in September 2024, attaching watermarks to AI-generated content. In January 2025, it updated its 'responsible AI' guidelines.
- Competitive Landscape: Competition with OpenAI (GPT), Google (Gemini), and Anthropic (Claude) intensifies, with Meta differentiating itself through open-source and social platform integration.
Related Topics
- [[Llama (language model)]]
- [[Facebook AI Research]]
- [[Generative artificial intelligence]]
- [[Metaverse]]
- [[Open-source artificial intelligence]]
---
AI-generated document · Community contributions welcome