Internal Representation, Not Clinical Knowledge: Where Apparent LLM Triage Failures Originate
Fuente:
arXiv
Saved in:
| Main Authors: | , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2026
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866917543856308224 |
|---|---|
| author | Navarro, David Fraile Como, Berardino Sheng, Jialei Ananthan, Soundariya Berkovsky, Shlomo |
| author_facet | Navarro, David Fraile Como, Berardino Sheng, Jialei Ananthan, Soundariya Berkovsky, Shlomo |
| contents | Patient-voiced clinical-triage benchmarks report high under-triage rates for consumer LLMs for constrained multiple-choice output, yet the same cases score differently with free-text. We ask whether output format changes the model's \emph{clinical representation} or only the mapping from a preserved representation to an answer. Using sparse-autoencoder (SAE) features in Gemma 3 4B/12B IT and Qwen3-8B, we find the same medical features fire on the shared clinical narrative under both formats but go {silent} at the multiple-choice decision token in all the cases at every model. Three independent methods (natural-language autoencoder verbalization, decision-token logit attribution, and top-feature characterization) agree that scaffold and format features, but not medical features, drive the decision logits. Behaviorally, the multiple-choice penalty inverts under both structured and natural-language input, option-order shuffle rules out positional bias, and the gap is dominated by off-by-one decision (the model picks an adjacent acuity letter to the gold answer) rather than knowledge failure. Thus, the failure originates in the output format and not in the clinical representation. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2605_29889 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | Internal Representation, Not Clinical Knowledge: Where Apparent LLM Triage Failures Originate Navarro, David Fraile Como, Berardino Sheng, Jialei Ananthan, Soundariya Berkovsky, Shlomo Computation and Language Artificial Intelligence Patient-voiced clinical-triage benchmarks report high under-triage rates for consumer LLMs for constrained multiple-choice output, yet the same cases score differently with free-text. We ask whether output format changes the model's \emph{clinical representation} or only the mapping from a preserved representation to an answer. Using sparse-autoencoder (SAE) features in Gemma 3 4B/12B IT and Qwen3-8B, we find the same medical features fire on the shared clinical narrative under both formats but go {silent} at the multiple-choice decision token in all the cases at every model. Three independent methods (natural-language autoencoder verbalization, decision-token logit attribution, and top-feature characterization) agree that scaffold and format features, but not medical features, drive the decision logits. Behaviorally, the multiple-choice penalty inverts under both structured and natural-language input, option-order shuffle rules out positional bias, and the gap is dominated by off-by-one decision (the model picks an adjacent acuity letter to the gold answer) rather than knowledge failure. Thus, the failure originates in the output format and not in the clinical representation. |
| title | Internal Representation, Not Clinical Knowledge: Where Apparent LLM Triage Failures Originate |
| topic | Computation and Language Artificial Intelligence |
| url | https://arxiv.org/abs/2605.29889 |