The study investigates whether the reasoning steps a language model writes are causally responsible for its answers. Using a causal intervention method on the activation stream, the authors find that for Qwen3-4B, about 77% of stated steps are causally load‑bearing, while behavioral tests overestimate this by roughly 11 percentage points. The faithfulness of reasoning decreases with model size and depth of reasoning, especially for the smaller Qwen3-1.7B.
By Abhiram Bhupatiraju, Rayan Nyaupane
arXiv:2606. 30128v1 Announce Type: new Abstract: Chain-of-thought (CoT) prompting improves LLM reasoning, but the source is contested: do the intermediate steps help because they carry useful semantic content, or because conditioning on more tokens buys extra computation before the model commits to an answer?
By Wenlong Wang, Fergal Reid
arXiv:2607. 13069v1 Announce Type: new Abstract: Large language models produce chain-of-thought (CoT) reasoning that appears logically sound yet may not genuinely depend on its stated premises.
By Hironao Nakamura
arXiv:2606. 05633v1 Announce Type: new Abstract: Retrieval-augmented QA pipelines often route retrieved passages through an LLM \emph{rewriter} before a smaller reader, lifting F1 by tens of points on multi-hop benchmarks; this gain is typically credited to improved evidence quality.
By Yuejie Li, Yueying Hua, Ke Yang, Li Zhang, Yueping He, Yueping He, Ruiqi Li, Bolin Chen, Tao Wang, Bowen Li, Chengjun Mao
The paper audits whether synthetic distractors in RLVR corpora act as shortcuts for learning policies. A classifier using only surface statistics barely outperforms chance, and manual inspection reveals that code distractors are almost identical to correct answers. Experiments with a paraphrase‑matched control show no exploitation advantage for the unmodified data, indicating that the detectable artifact was not used by the policy.
By Esther Xin
arXiv:2608. 01575v1 Announce Type: new Abstract: Whether large language models perform genuine algorithmic reasoning or mere pattern completion is hard to test, because most benchmarks lack a ground truth for correct inductive inference.
By Hector Zenil, Luan Ozelim
arXiv:2610.29052v1 Announce Type: cross
Abstract: A typed decision maps evidence, a question and a fixed set of options to a probability for every option, computed in one forward pass, by reading a l...
By Yuchuan Weng, Chi Liu, Xin Chen, Srinivasan Manoharan
arXiv:2608.29956v1 Announce Type: new
Abstract: Large language models often answer complex reasoning questions without revealing intermediate steps, raising whether they reason latently or complete p...
By Armaan Singh, Ryan Trinh Le, Jasmine Kaur, Abdullah Sultan, Edward Lue Chee Lip, Kiran Nijjer, Adnan Ahmed, Vasu Sharma
The paper investigates whether latent chain-of-thought (latent CoT) models truly perform reasoning by examining twenty-one model-task configurations across different training recipes, tasks, and backbones. Using chain deletion, content replacement, probes, and trajectory analysis, the authors find that on logical tasks the latent chain often contains an answer before the chain starts and can be deleted with little effect, while on arithmetic tasks the chain’s content is essential. The study introduces a taxonomy that distinguishes what a chain represents from how a model uses it, clarifying that readable latent states alone do not prove reasoning.
By Jingjie Qiu, Sheldon Yu, Xunyi Jiang, Jiajin Liu, Zihan Huang, Zhendong Chu, Tong Yu, Julian McAuley, Junda Wu
Whether large language models perform genuine algorithmic reasoning or mere pattern completion is hard to test, because most benchmarks lack a ground truth for correct inductive inference. We introduce F-ICL, an in-context-learning benchmark that supplies one exactly.
arXiv:2605. 03534v2 Announce Type: replace-cross Abstract: Retrieval-augmented generation (RAG) grounds answers in retrieved passages, yet relevance does not guarantee sufficiency: a topical passage may still fail to justify the answer.
By Jingxi Qiu, Zeyu Han, Cheng Huang
The paper introduces AgentDiff, a metric that quantifies how much LLM agents’ answers differ when inputs are altered by meaning‑bearing rewrites (paraphrases, synonym substitutions) versus presentation changes (reordering, formatting, distractors). Across 68 model–benchmark–scaffold combinations involving ten LLMs and over 1,500 questions, meaning‑bearing rewrites consistently produce a roughly 20‑percentage‑point higher inconsistency rate than presentation changes, a gap that persists across severity proxies and remains significant even outside the Qwen family. Trace analysis reveals that meaning‑bearing rewrites preserve the first action but reduce thought similarity from the second step onward, extending the divergence cascade—a phenomenon termed “stealth divergence.”
By Liyun Zhang, Jiayi Guo