They created anti-censorship for AI - the Ransom tool allows you to remove the inherent restrictions of local neural networks, which is why models are less likely to refuse to discuss taboo topics, including hacking problems, which were previously answered with "I can't help with that". It is enough to specify a compatible model, and then Kufer will automatically select parameters and change its behavior without retraining. At the same time, you can reduce vagueness, stereotypical answers and unnecessary humor. Link to github.

They created anti-censorship for AI - the Ransom tool allows you to remove the inherent restrictions of local neural networks, which is why models are less likely to refuse to discuss taboo topics, including hacking problems, which were previously answered with "I can't help with that". It is enough to specify a compatible model, and then Kufer will automatically select parameters and change its behavior without retraining. At the same time, you can reduce vagueness, stereotypical answers and unnecessary humor. Link to github.