62
Hardening Firefox with Anthropic’s Red Team (blog.mozilla.org)
submitted 8 months ago by to c/technology@lemmy.world
9 comments fedilink hide all child comments
 

Lobsters.

you are viewing a single comment's thread
view the rest of the comments
[–] 10 points 8 months ago (6 children)

Yeah I'm actually kinda into this. Even if the AI vomits up a bunch of hallucinated vulnerabilities, there's a team of (presumably) capable people there to figure that out. Seems like a pretty valid use for the technology.

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 6 child comments
    •
  • [–] -3 points 8 months ago (5 children)

    Hallucinated? From researched and documented code spelunking?

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 5 child comments
    •
  • [–] 2 points 8 months ago (4 children)

    That's...exactly my point though...

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 4 child comments
    •
  • [–] -1 points 8 months ago (3 children)

    What is?

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 3 child comments
    •
  • [–] 3 points 8 months ago (2 children)

    That even though the team is using AI to check for vulnerabilities, they're trained and know when their AI is hallucinating and when it's not.

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 2 child comments
    •
  • [–] -1 points 8 months ago (1 child)

    I guess I’m not sure how hallucinating and reading from source code are overlapping. Do you think these models are just barfing back garbage nonsense?

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 1 child comment
    •