Even Google Research themselves use youtube-dl when assembling computer vision datasets. The only way to work on their data and reproduce the results is through youtube-dl, as video datasets are often only distributed as URLs and timestamps, and even the example codes use youtube-dl.

I wonder how this will impact the AI research community if there's some FUD around this tool.

Generally scraping is a gray area. I know researchers who scraped Google street view for large areas to train AI models without Googles approval and got hired by Google for their impressive AI skills instead of threatened.

I know researchers are a tiny group compared to consumers, but it's still interesting I think.