News for hackers. Stuff that matters.

A comment on Laion-400M: open-source dataset of 400M image-text pairs.

Re:Laion-400M: open-source dataset of 400M image-text pairs by heeen2 on Monday September 13, 2021 @06:57AM (#28509896)
entered "woman" and started to wonder where they crawled this from
[ Parent | on HN ]
Re:Laion-400M: open-source dataset of 400M image-text pairs by Yenrabbit on Monday September 13, 2021 @08:47AM (#28510816)
Even leaving the query blank shows a few NSFW things that suggest they really did just look for any and all image-caption pairs they could find!
[ Parent | on HN ]
Re:Laion-400M: open-source dataset of 400M image-text pairs by rom1504 on Wednesday September 15, 2021 @04:07PM (#28543718)
a safe mode is now in place
[ Parent | on HN ]
Re:Laion-400M: open-source dataset of 400M image-text pairs by Fnoord on Monday September 13, 2021 @12:24PM (#28513415)
I tried a few keywords which should not provide NSFW, but they did, such as "thumb". Not cool. This needs a 'safe mode'.
[ Parent | on HN ]