Forbes January 2, 2026
In today’s column, I examine the latest approach to getting generative AI and large language models (LLMs) to be honest, namely, forcing the AI to provide confessions about the answers that are being generated.
Yes, in the same sense that confession is supposed to be good for the human soul, some believe that AI might benefit by having to make confessions. Here’s the deal. It is already known that AI can be deceptive, scheming, and altogether dishonest. Various AI safeguards try to stop this from happening or at least catch the AI in the act of being underhanded.
One new and quite clever approach to safeguarding consists of having AI produce a confession after each response that the AI generates....
Today's Sponsors
Today's Sponsor