Danilo Bzdok

Under input uncertainty, transformer models exhibit a systematic exploration of input‑agnostic conceptual representations, increasing the likelihood of hallucinations.