AI Watchdog Datasets

AI Watchdog: YT-Temporal-180M

Editor’s note: This search tool is part of The Atlantic’s investigation into how YouTube videos are taken to train AI tools. This work is part of AI Watchdog, The Atlantic’s ongoing investigation into the generative-AI industry.

YT-Temporal-180M is a collection of 5,494,771 YouTube videos. The data set was compiled by a team of researchers at the University of Washington and the Allen Institute for AI to train a multimodal model called Merlot, and was released in 2021. YT-Temporal-180M specifies the time stamp of every word spoken in a video. A script for downloading the data set is hosted by Hugging Face, an AI-development hub, and the actual data set is hosted by Google. The data set has been downloaded more than 1,200 times.