17
PSA: A talk with my friend who works in security made me rethink AI safety
We were having a beer in Denver last week and he said, 'We're building these powerful AI tools faster than we're building the guardrails.' He was talking about a new open-source model his team is testing, and how easy it is to misuse. Are we moving too fast on the cool stuff without thinking about the bad ways people could use it?
3 comments
Log in to join the discussion
Log In3 Comments
leor371mo ago
Whoa, that's a real gut check. Your friend is totally right. I saw a demo where someone took an open-source image generator and stripped the safety filters in an afternoon. Suddenly it could make fake, harmful stuff. The teams building the cool new features are just not the same teams thinking about all the ways it breaks. We gotta slow down and build the lock before we hand out the keys.
2
carter.jennifer1mo ago
The "build the lock first" idea sounds good in theory. But in practice, the bad guys are already picking the locks on everything else. Playing catch-up forever seems like a losing plan.
8
johnson.adam1mo ago
But the lock slows down the good stuff too.
4