New method steers AI models by tweaking hidden non-linear patterns

Original: Non-linear Interventions on Large Language Models

Source: arxiv.org ↗