Can We Read the Mind of an Audio LLM?

TL;DR — An audio LLM is a black box in a specific way: we see what it says, never what it worked out on the way there, and chain-of-thought monitoring only helps if the model writes its reasoning down. Reading a base Qwen3-Omni with a logit lens at the audio-token positions, the answer to a spoken question becomes legible — in words — in the middle layers, before a single token is emitted.

A hidden multi-hop chain, with no chain-of-thought

On a clip whose verbatim transcription is empty garbling, the readout reconstructs Watergate and scandal, passes through the role president, and resolves to Nixon. None of those words are in the question, the options, or the model's own transcription. The reasoning happened; it simply was never written down.

Figure 1. Where the task-relevant concept becomes readable, for seven curated cases. The strongest readout lands in the middle band of the network — and the word that surfaces is often not in the language of the question.

What the workspace looks like

Verbalizable

The intermediate state reads out as ordinary words, not as an opaque vector.

Multilingual

One audio-inferred concept surfaces in several scripts at once; 38% of top-1 readouts are Chinese on English inputs.

Paralinguistic

It forms the sound source, speaker role or affect that a text caption of the same clip throws away.

Causal

Activation patching shows the signal is used, and committed before the last fifth of the layers.

A waveform-swap control runs throughout: identical text, only the sound changed. That is what separates a genuinely audio-driven signal from a prior over the printed answer options.

Deleting layers maps the pipeline

Reading is correlational; deleting is causal. Bypassing one layer at a time gives a clean functional map: listening is localized at the entry, delivery at the output, and retrieval is spread across the interior — no single interior layer is essential.

Figure 2. Accuracy lost when a single layer is deleted (identity skip, averaged over clips). Only the encoder-to-text entry layer is critical; every interior layer costs at most ten points.

The same tooling shows where hallucinations come from: in text mode the wrong concept usually takes over early, and is never revised afterwards.

Why it matters, and what it is not

If what a model is about to say is legible mid-stack, monitoring need not wait for it to speak, and need not trust it to narrate its own reasoning. But this is a deliberately qualitative account of one model family: a readable correlate is not a complete mechanism, and the paper is conservative about what the lens can and cannot establish.

BibTeX

@article{fan2026readmind,
  title   = {Can We Read the Mind of an Audio {LLM}? A Verbalizable,
             Multilingual Middle-Layer Workspace},
  author  = {Jiajun Fan and Jingyuan Li and Prashanth Gurunath Shivakumar
             and Qi Luo and Jia-Hong Huang and M. Maruf and Roger Ren
             and Yile Gu and Rahul Pandey and Ge Liu and Ivan Bulyko},
  journal = {arXiv preprint arXiv:2608.24958},
  year    = {2026}
}