Inspect: A framework for large language model evaluations
An alignment auditing agent capable of quickly exploring alignment hypothesis
Open on GitHub