{"slug": "reasoning-error-from-known-fact-step-level-self-consistency-group-relative-for", "title": "Reasoning Error from Known Fact: Step-Level Self-Consistency Group Relative Policy Optimization for LLM", "summary": "A new study from arXiv finds that large language models (LLMs) produce Context-Sensitive Factual Hallucinations during multi-step reasoning, where the model possesses relevant knowledge but makes factual errors due to contextual interference. The researchers propose Step-level Self-Consistency Group Relative Policy Optimization (SSC-GRPO), which assigns step-level rewards by computing self-consistency scores across multiple rollouts, achieving state-of-the-art performance on mathematical reasoning benchmarks and hallucination leaderboards.", "body_md": "arXiv:2607.18915v1 Announce Type: new\nAbstract: With the rapid advancement of large language models (LLMs), modern systems not only possess strong foundational capabilities and extensive knowledge, but can also solve complex problems via long, multi-step reasoning. However, as reasoning traces become longer, LLMs may produce a substantial amount of hallucinated content during the reasoning process, which is often difficult to detect. In this work, we conduct a fine-grained analysis of hallucinations arising in LLM reasoning and find that the reasoning traces are particularly prone to Context-Sensitive Factual Hallucinations: cases where the model actually has the relevant knowledge, yet makes factual errors due to contextual interference during reasoning. To address this issue, we propose Step-level Self-Consistency Group Relative Policy Optimization (SSC-GRPO), which assigns step-level rewards to reasoning traces by computing self-consistency scores of individual steps across multiple rollouts. Compared with prior methods, SSC-GRPO achieves state-of-the-art performance on both mathematical reasoning benchmarks and hallucination leaderboards. Our results offer a new perspective for detecting and mitigating hallucinations in the reasoning process of large language models.", "url": "https://wpnews.pro/news/reasoning-error-from-known-fact-step-level-self-consistency-group-relative-for", "canonical_source": "https://www.machinebrief.com/news/reasoning-error-from-known-fact-step-level-self-consistency-2tch", "published_at": "2026-07-22 04:00:00+00:00", "updated_at": "2026-07-22 04:09:43.176466+00:00", "lang": "en", "topics": ["large-language-models", "ai-research", "ai-safety"], "entities": ["arXiv", "SSC-GRPO"], "alternates": {"html": "https://wpnews.pro/news/reasoning-error-from-known-fact-step-level-self-consistency-group-relative-for", "markdown": "https://wpnews.pro/news/reasoning-error-from-known-fact-step-level-self-consistency-group-relative-for.md", "text": "https://wpnews.pro/news/reasoning-error-from-known-fact-step-level-self-consistency-group-relative-for.txt", "jsonld": "https://wpnews.pro/news/reasoning-error-from-known-fact-step-level-self-consistency-group-relative-for.jsonld"}}