The \”inner thoughts of the lie detector\” are revealed by a new AI.
Jacob Andreas, a MIT professor, lamented, \”I wish I had this to cite.\” He had just published a study exploring the extent to the which language models reflect the motivations of humans.
Burns initially refused the job offer, but after a personal plea from Sam Altman (cofounder and CEO of OpenAI), Burns changed his mind.
Leike states that \”Collin’s work on ‘Discovering latent knowledge in language models without supervision’ is a new approach to determining the truth about what language models believe about the universe.\” Leike says that the work of Colin is exciting because it could be applied to systems smarter than humans, even if they don’t know what is true.
Source:
https://www.freethink.com/robots-ai/ai-lie-detector