At Meta, our vision is to bring personal superintelligence to everyone. This vision is deeply inspiring to us as a company, as well as our people. For the Facebook search team, it means relentlessly asking how they can help people not just find information, but truly understand, connect and grow.
Chen Zhou, tech lead manager for Facebook search, is always wondering how AI and LLMs can help in real-world scenarios. For Chen, AI isn’t just complex code or mathematical models — it’s a tool to help people understand the world around them and discover new connections and information. He believes that search is the perfect scenario to capture the spirit of human curiosity. This belief drives Chen and his team to push the boundaries of what’s possible with LLMs for Facebook search.
Chen has witnessed firsthand the profound transformation that LLMs are bringing to the way people discover information. Before LLMs, search was mainly about matching keywords to entities or documents. Today, it has evolved into a more natural, intuitive and personalized experience driven by cutting-edge AI technologies.
The rise of LLMs has been a game-changer for search. These models enable us to better understand complex human language queries, extract rich semantic meaning from content, and generate more relevant and context-aware results. Chen and his team are exploring how to integrate LLMs like Llama into the Facebook search ranking stack by building high-quality pretrained models, fine-tuning LLMs with few-shot learning and leveraging prompt engineering with LLM capabilities. Not only are LLMs enhancing the semantic relevance, location, freshness and quality of search results, they’re also evaluating the end-to-end (E2E) ranking stack to deliver more personalized responses.
“LLMs allow us to do more with search — going beyond people’s initial query to understand the broader context of the information they’re trying to find or the topic that they want to learn more about. We can use AI to stitch together information across the broader web and create a more impactful search experience.”
Looking ahead, Chen and his team plan to double down on LLMs’ reasoning ability to improve Facebook search’s understanding of content and people’s interests. They also want to use Llama and other AI models from Meta to increase ML developer productivity with co-pilot tools and automate more of the recommendations stack with AI agents. This journey will go beyond model training to include E2E ranking evaluation and continuous innovation.
Building state-of-the-art AI models is only part of the challenge. For Chen and the Facebook search team, efficiency is just as critical — especially when serving billions of people worldwide. GPUs are expensive, and optimizing online efficiency is a top priority. By optimizing model architectures and inference pipelines, the team ensures that advanced AI capabilities are accessible to everyone without compromising on speed or cost. Besides the NVIDIA GPU optimization, the team also leverages internal hardware, such as the next-generation Meta Training and Inference Accelerator (MTIA), to deliver high-performance AI at scale.
This focus on efficiency allows Facebook search to democratize access to leading AI models, making search smarter and faster for all.
AI advancements at Meta aren’t limited to Facebook search. The models and technologies developed by Chen and his team are designed to generalize well across Meta technologies, including search on Instagram, Marketplace and Threads. This cross-product generalization ensures that AI innovations benefit people everywhere, creating a unified and intelligent search experience across Meta technologies.
The future of search at Meta is bright. The Facebook search team is redefining what search means — moving beyond keyword matching to create an intelligent discovery engine. This transformation will unlock new ways for people to connect with the information, communities and experiences that matter most to them.
As Facebook search continues to innovate, Chen is excited to be part of a team that is engineering the future of search and realizing our vision of personal superintelligence for everyone. The journey is just beginning, and the possibilities are limitless.