AI Watchdog Datasets

AI Watchdog: MiraData-330K

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.

MiraData-330K is a collection of 330,000 video clips taken from 225,544 videos from YouTube, Videvo, Pixabay, and Pexels, and paired with text captions. This data set features long video clips (averaging 72.1 seconds each) and captions (averaging 318 words). It was compiled by researchers at Tencent’s Applied Research Center and released in 2024. The researchers used GPT-4V to generate captions. The data set is hosted on Hugging Face, an AI-development hub, and has been downloaded more than 2,500 times.