A team of researchers has found a way to steer the output of large language models by manipulating specific concepts inside these models. The new ...