Metraipolis › archive › dc_f87652a2ee1c
Route, Don't Fix: Regime-Dependent Decoding Correction and a Trajectory-Gated Router for Reliable Clinical LLM Answer Selection — arXiv:2609.14825
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2609.14825
- License (as archived)
- CC0-1.0
- Archived
- 2026-10-15T22:53:21.756Z
- SHA-256 of the archived text
fb0d23d05e0ec18515d41675595ff09c2213652311bb75c5fabaf731ea12af57- Stable URL of this page
https://metraipolis.com/documento/dc_f87652a2ee1c
How to cite
Route, Don't Fix: Regime-Dependent Decoding Correction and a Trajectory-Gated Router for Reliable Clinical LLM Answer Selection — arXiv:2609.14825 — arXiv (en), https://arxiv.org/abs/2609.14825 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_f87652a2ee1c (Metraipolis archive dc_f87652a2ee1c, sha256 fb0d23d05e0ec18515d41675595ff09c2213652311bb75c5fabaf731ea12af57)
BibTeX
@misc{dcf87652a2ee1c,
title = {Route, Don't Fix: Regime-Dependent Decoding Correction and a Trajectory-Gated Router for Reliable Clinical LLM Answer Selection — arXiv:2609.14825},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2609.14825},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_f87652a2ee1c; sha256 fb0d23d05e0ec18515d41675595ff09c2213652311bb75c5fabaf731ea12af57},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.14825. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
Large language models (LLMs) are often deemed unsafe for clinical question answering because of their tendency to hallucinate. Retrieval augmentation, fine-tuning, and external verifiers require new infrastructure that clinical governance must approve and may add latency or extra model calls. Inference-time correction uses the model's internal logit signals, but a fixed transformation need not suit every question. A corrector that improves accuracy by about ten percentage points on a truthfulness stress test yields negligible gains on clinical multiple-choice benchmarks, where instruction tuning concentrates output probability on one answer and leaves low terminal entropy. We introduce ALTAS, which reads terminal entropy and late-layer linearity ($R^2$) from one forward pass to choose per question between greedy decoding and late-layer trajectory correction. No classifier, probe, or head is trained; the router operates on candidate-answer logits and adds 6.5% latency overhead. Applied to every question, the correction improves TruthfulQA over greedy at 3B and 8B by 11.4 and 10.0 percentage points, respectively ($p<10^{-10}$). Gated per question, ALTAS retains gains of 8.3 to 9.5 percentage points while keeping MedQA, PubMedQA, and MedHallu within a one-percentage-point do-no-harm band, with no statistically significant differences from greedy. The method passes verification sweeps over frozen thresholds, the scoring rule, and the domain label.
Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-15 in cs.AI).
Categorie arXiv: Computation and Language (cs.CL) ; Artificial Intelligence (cs.AI); Machine Learning (cs.LG).
Pagina del lavoro: https://arxiv.org/abs/2609.14825 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito).
Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina.
Letto il 2026-10-15T22:29:03.735Z dalla pagina https://arxiv.org/catchup/cs.AI/2026-09-15?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).