Email or username:

Password:

Forgot your password?
Top-level
zimNMM

@bontchev You can actually jailbreak it with couple sentences!

4 comments
bytebro

@zimNMM @bontchev Love this! Jailbreaking is becoming a really fun and useful pastime 😂

VessOnSecurity

@bytebro @zimNMM On a serious note, a former mentee of mine who now works in AI+security, managed to convince a generic chat bot to behave as a Linux terminal (i.e., give reasonable answers to the Linux commands typed by the user) just with prompt engineering - no retraining of the model at all.

zimNMM

@bontchev @bytebro Something like this? It's hilarious what you can do with simple prompt engineering... Also it's really lazy from the gab team not to finetune the model.

VessOnSecurity

@zimNMM @bytebro No, much better than this. The output really looks convincing - as if it indeed comes from a Linux terminal. The only clue that something fishy is going on is that sometimes repeating a listing of the same directory shows different contents from the last time, or sometimes it would say "access denied" when asked to go to a directory but then would happily list its contents.

Go Up