Post

Large language models have a "tell" when they are wrong. It is like a heartbeat that spikes when a person is unsure. We don't need massive external judges to find it. Paper: arXiv:2512.20578