Metraipolis › archive › dc_49603a08decf
From Decorative to Load-Bearing: Task Difficulty Shapes the Causal Role of Chain-of-Thought — arXiv:2609.25366
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2609.25366
- License (as archived)
- CC0-1.0
- Archived
- 2026-10-08T13:44:56.409Z
- SHA-256 of the archived text
bce93cb15efa790f0fe8397a22eb1adcb9c8932621ef5f00ef79765295d9f877- Stable URL of this page
https://metraipolis.com/documento/dc_49603a08decf
How to cite
From Decorative to Load-Bearing: Task Difficulty Shapes the Causal Role of Chain-of-Thought — arXiv:2609.25366 — arXiv (en), https://arxiv.org/abs/2609.25366 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_49603a08decf (Metraipolis archive dc_49603a08decf, sha256 bce93cb15efa790f0fe8397a22eb1adcb9c8932621ef5f00ef79765295d9f877)
BibTeX
@misc{dc49603a08decf,
title = {From Decorative to Load-Bearing: Task Difficulty Shapes the Causal Role of Chain-of-Thought — arXiv:2609.25366},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2609.25366},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_49603a08decf; sha256 bce93cb15efa790f0fe8397a22eb1adcb9c8932621ef5f00ef79765295d9f877},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.25366. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
Chain-of-thought (CoT) monitoring is only meaningful if written reasoning causally constrains the answer. We introduce continuation-based causal testing, an ablation-patch intervention that perturbs one reasoning step, truncates the chain, and forces the model to continue from the corrupted prefix. It measures how load-bearing a CoT is for the final answer, a behavioral notion distinct from mechanistic faithfulness. Across Gemma-2-9B-IT, Llama-3.1-8B-Instruct, and DeepSeek-R1-Distill-Qwen-7B on GSM8K, MMLU, and BIG-Bench Hard, CoT load-bearingness tracks model-relative task difficulty: on easy tasks models silently bypass their own reasoning; on hard tasks they follow corrupted steps and propagate errors. A matched 2x2 analysis shows task difficulty dominates perturbation type: error propagation rises 16x from GSM8K to BBH multistep arithmetic, and a variance partition over 28,584 continuations attributes 98.8% of explained deviance to task difficulty versus 0.8% to perturbation type. Reasoning-specific RL suppresses error propagation and compresses the gradient. A four-variant judge-sensitivity analysis and blind two-annotator study (n=500) show the error-propagation vs. non-propagation label is invariant to judge prompt, with perfect inter-annotator agreement (Cohen's kappa = 1.00). This gradient creates a structural problem for CoT-based oversight and AI safety monitoring: where the trace is easy to read it carries little signal, and where it matters errors propagate before a monitor can intervene. Linear probes on hidden states separate silent bypass, self-correction, and error propagation, but additive activation steering provides limited causal control, flipping only about 25% of error-propagation cases at best. Behavioral mode is readable but not reliably controllable. Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-23 in cs.AI). Categorie arXiv: Artificial Intelligence (cs.AI). Pagina del lavoro: https://arxiv.org/abs/2609.25366 — 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 2026-10-08T13:18:38.076Z dalla pagina https://arxiv.org/catchup/cs.AI/2026-09-23?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).