Identifying vulnerabilities is good for public safety, industry, and the scientists making these models.
A team of researchers has found a way to steer the output of large language models by manipulating specific concepts inside ...