Comment by telotortium

Threads like this want me to throw my laptop into a volcano and lobotomize myself. I have a bridge if you think an LLM can reliably generically redact passwords. Especially given the intersection of password quality and the type of person to let their passwords leak into output of a recorded terminal session.

I love leaving my baseline security in the hands of a hallucinating, unreliable token generator.

I'm sorry, I don't even know how to explain that this type of feature would be such a mis-feature, it's hard to even know how to explain it to somehow for whom its not obvious.

ksherlock 10 hours ago

s/mysekretpassword/••••••••••••••••/g

What about that should involve a large language model?

Reply View 2 replies

telotortium 6 hours ago

Obviously the part that determines what part of the output might be a password or other sensitive value.

Reply View | 1 reply
- nixosbestos 6 hours ago
  
  Threads like this want me to throw my laptop into a volcano and lobotomize myself. I have a bridge if you think an LLM can reliably generically redact passwords. Especially given the intersection of password quality and the type of person to let their passwords leak into output of a recorded terminal session.
  I love leaving my baseline security in the hands of a hallucinating, unreliable token generator.
  I'm sorry, I don't even know how to explain that this type of feature would be such a mis-feature, it's hard to even know how to explain it to somehow for whom its not obvious.
  
  Reply View | 0 replies