Reading Between the Dots: Decoding Hidden Computation across Filler Tokens

arXiv:2607.03502v1 Announce Type: new Abstract: Frontier LLMs can perform multi-step reasoning over content-free filler tokens like dots or counting sequences, producing correct answers with no visible chain-of-thought (CoT). This is a limit case for behavioral oversight, where surface tokens carry no information about the underlying reasoning. But hidden from the output is not the same as hidden from us. On four task families (fact retrieval, parallel numeric composition, string manipulation, a...

arXiv cs.CL ·Kaley Brauer, Claudio Mayrink Verdun, Samuel Marks ·
compartilhar: