Latent Lab: LLMs for Knowledge Exploration

Kevin Dunnell

Latent Lab is an AI-driven knowledge exploration system designed to navigate through the 4,000+ research projects within the Media Lab. Utilizing GPT, Latent Lab generates embeddings for project descriptions and extracts relevant topics and subtopics from unstructured datasets, providing an immersive 2D visualization that maps the connections between diverse projects. The contour lines emphasize the density of projects surrounding various themes, while the embedding-based search feature allows users to input their ideas and uncover related research endeavors. Explore the evolution of the Media Lab by scrubbing the interactive timeline. As users traverse the virtual space, the system automatically summarizes samples of projects currently in view, enhancing the user's understanding of the research landscape. The sidebar offers a dedicated space for delving into the specifics of selected projects and crafting unique "recipes" by combining project elements, ultimately inspiring the generation of innovative new research ideas.

Interested users are invited to experience this exploration tool directly by visiting, where you can navigate through various datasets or upload and explore a dataset of your own.