Tag
This paper demonstrates that frontier language models can perform 'invisible reasoning' using semantically irrelevant filler tokens, improving accuracy on synthetic reasoning tasks by up to 13 percentage points, which undermines the assumption that chain-of-thought monitoring captures all reasoning.
This paper shows that frontier LLMs can perform multi-step reasoning over content-free filler tokens, and that hidden states in the residual stream can be decoded to recover intermediate values with high accuracy, challenging the assumption that chain-of-thought monitoring is the only audit tool.