I don't think that's feasible for anyone but someone like Google or another tech giant. They could use Wikimedia Commons to get a smaller collection of better

I don't think that's feasible for anyone but someone like Google or another tech giant. They could use Wikimedia Commons to get a smaller collection of better labeled images. Currently it's images from Common Crawl extracted by Laion, not sure if it already includes Wikimedia Commons.