Researchers test whether AI models can introspect on their own 'thoughts'

By injecting thoughts directly into large language models, researchers are mapping AI's capacity for introspection.