you are viewing a single comment's thread
view the rest of the comments
[–] 11 points 9 months ago* (5 children)

Sorry to double post but this is potentially relevant:

So, yeah, the FBI maintains a large database of CSAM, for their... CSAM content ID style system.

People who know that that exists have occasionally pointed out that... that is kind of weird, lets just put it that way.

Well, I've just had a horrifying realization.

We currently have AI/LLM corpos on the record going to places like AnnasArchive to acquire their vast troves of data to use in training an LLM.

... If somebody hooks up this FBI CSAM database... into an LLM... well, you now basically have a machine that produces CSAM, and likely also gore/torture videos.

... hooray ...

Like uh, Pete Hegseth just ... you know, hooked up Grok/XAI into... apparently I guess all of the US's military systems.

The scenario I am describing is unfortunately plausible.

  • source
  • parent
  • hideshow 5 child comments
  • [–] 8 points 9 months ago (2 children)

    You may have missed how there's a massive CSAM problem with AI because the models are all trained on CSAM. "Plausible" is a bit behind the times.

  • source
  • parent
  • hideshow 2 child comments
  • [–] 3 points 9 months ago (1 child)

    Uh... yes, I apparently did miss that.

    ... what?

    How... did that happen?

    Can you link an article or something that substantiates that?

  • source
  • parent
  • hideshow 1 child comment
  • [–] 7 points 9 months ago (1 child)

    Good news, that's not how the CSAM database works. What happens is the image is hashed (for non technical people, hashing is a way of always generating the same "ID" from the file data) and the image is discarded. Now the article I found to verify my claim also talks about how the FBI used to distribute CSAM to catch people, so they have some probably... But not a database wide amount.

    https://cybernews.com/editorial/war-on-child-exploitation/

  • source
  • parent
  • hideshow 1 child comment