LLMS KNOW MORE THAN THEY SHOW: ON THE INTRINSIC REPRESENTATION OF LLM HALLUCINATIONS

Summary from the TLDR AI newsletter

Another study that shows language models encode internally when they aren't factual. This evidence can help us steer them and mitigate hallucinations.