Science ❯ Computer Science ❯ Artificial Intelligence Research ❯ Model Training
By isolating unwanted AI personas through linear activation-space directions, the toolkit inoculates models against harmful traits
