1 paper
Ishaan Kelkar, Nebras Alam, Vikram Kakaria +3
Language models are often sycophantic: they agree with a user's stated opinion whether or not it is correct. Prior work has shown that this trait can be controlled by steering a mo…