How Language Models Process Out-of-Distribution Inputs: A Two-Pathway Framework
Fuente:
arXiv
Salvato in:
| Autore principale: | |
|---|---|
| Natura: | Preprint |
| Pubblicazione: |
2026
|
| Soggetti: | |
| Accesso online: | |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866911638012035072 |
|---|---|
| author | Saghir, Hamidreza |
| author_facet | Saghir, Hamidreza |
| contents | Recent white-box OOD detection methods for LLMs -- including CED, RAUQ, and WildGuard confidence scores -- appear effective, but we show they are structurally confounded by sequence length (|r| >= 0.61) and collapse to near-chance under length-matched evaluation. Even raw attention entropy (mean H(alpha) across heads and layers), a natural baseline we include for completeness, shows the same confound. The confound stems from attention's Theta(log T) dependence on input length. To identify genuine OOD signals after deconfounding, we propose a two-pathway framework: embeddings capture what text is about (effective for topic shifts), while the processing trajectory -- hidden-state evolution across layers -- captures how the model processes input. The relative power of each pathway varies along a vocabulary-transparency spectrum: embedding methods excel on vocabulary-distinctive OOD, while trajectory features detect covert-intent inputs that share vocabulary with normal text (0.721 avg AUROC; Jailbreak: 0.850). Three evidence lines support this framework: (1) a crossover between k-NN and trajectory scoring across 6 tasks, where each pathway wins on different OOD types; (2) a per-layer analysis showing that layer-0 k-NN signal is almost entirely a length artifact (Jailbreak: 0.759 raw -> 0.389 matched) -- processing constructs genuine OOD signal from near-chance embeddings; and (3) circuit attribution showing adversarial tasks engage attention circuits more than semantic tasks (p = 0.022; Jailbreak patching p < 0.001), with partial cross-model replication. Code release upon publication. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2605_00269 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | How Language Models Process Out-of-Distribution Inputs: A Two-Pathway Framework Saghir, Hamidreza Computation and Language Machine Learning I.2.7; I.2.6 Recent white-box OOD detection methods for LLMs -- including CED, RAUQ, and WildGuard confidence scores -- appear effective, but we show they are structurally confounded by sequence length (|r| >= 0.61) and collapse to near-chance under length-matched evaluation. Even raw attention entropy (mean H(alpha) across heads and layers), a natural baseline we include for completeness, shows the same confound. The confound stems from attention's Theta(log T) dependence on input length. To identify genuine OOD signals after deconfounding, we propose a two-pathway framework: embeddings capture what text is about (effective for topic shifts), while the processing trajectory -- hidden-state evolution across layers -- captures how the model processes input. The relative power of each pathway varies along a vocabulary-transparency spectrum: embedding methods excel on vocabulary-distinctive OOD, while trajectory features detect covert-intent inputs that share vocabulary with normal text (0.721 avg AUROC; Jailbreak: 0.850). Three evidence lines support this framework: (1) a crossover between k-NN and trajectory scoring across 6 tasks, where each pathway wins on different OOD types; (2) a per-layer analysis showing that layer-0 k-NN signal is almost entirely a length artifact (Jailbreak: 0.759 raw -> 0.389 matched) -- processing constructs genuine OOD signal from near-chance embeddings; and (3) circuit attribution showing adversarial tasks engage attention circuits more than semantic tasks (p = 0.022; Jailbreak patching p < 0.001), with partial cross-model replication. Code release upon publication. |
| title | How Language Models Process Out-of-Distribution Inputs: A Two-Pathway Framework |
| topic | Computation and Language Machine Learning I.2.7; I.2.6 |
| url | https://arxiv.org/abs/2605.00269 |