Research tests whether AI models can know their own thoughts

By injecting thoughts directly into large language models, researchers aim to map AI's capacity for introspection.