Consider setting up a dedicated red-team environment. Use the models responsibly to identify vulnerabilities before they're exploited.

Let's fucking gooooo!!!!!

This is the way

Whats the cost difference from the regular versions?

Hackers dream come true

One caveat worth flagging for people auditing code with these: the refusal direction is not fully separate from other capabilities, so removing it can degrade instruction following, coherence, and reasoning on standard benchmarks. Worth running the same prompts on the base model too, since uncensore...

awesome