{"total":14,"items":[{"citing_arxiv_id":"2607.00924","ref_index":59,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"Graph-Native Reinforcement Learning Enables Traceable Scientific Hypothesis Generation through Conceptual Recombination","primary_cat":"cs.AI","submitted_at":"2026-07-01T13:26:10+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"Graph-PRefLexOR fine-tunes graph-native models with GRPO to organize reasoning into phases, yielding 40-65% gains in traceable hypothesis generation and 2-3x semantic diversity on 100 materials science questions.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2606.25191","ref_index":44,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"To Isolate or to Score? Model-Adaptive Assessment for Cost-Efficient Multi-Agent RAG","primary_cat":"cs.AI","submitted_at":"2026-06-23T21:34:23+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"Empirical study finds isolation drives gains for weak models in multi-agent RAG while scoring matters for strong ones, enabling MADARA for cost-efficient adaptive assessment.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2606.21678","ref_index":17,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"Decodable but Not Faithful: Coupling Natural-Language Rationales to Programmatic Verifiers","primary_cat":"cs.LG","submitted_at":"2026-06-19T18:37:24+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"Consistency training decodes verifier information from rationale representations but does not produce faithful natural-language explanations.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2606.18671","ref_index":43,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"HANSEL: Extracting Breadcrumbs from Web Agent Trajectories for Interactive Verification","primary_cat":"cs.HC","submitted_at":"2026-06-17T04:13:20+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"HANSEL extracts navigable evidence from agent trajectories with 83.7% precision and 88.8% recall on 45 tasks, reduces volume by 61.6%, and improves verification metrics in a 14-participant study.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2606.11445","ref_index":71,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"Forecasting Future Behavior as a Learning Task","primary_cat":"cs.AI","submitted_at":"2026-06-09T20:56:23+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":7.0,"formal_verification":"none","one_line_summary":"Behavior Forecasters trained on LRM trajectories outperform larger models in predicting repeatability and input sensitivity at low cost.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2605.24396","ref_index":28,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"Understanding and Mitigating Premature Confidence for Better LLM Reasoning","primary_cat":"cs.AI","submitted_at":"2026-05-23T04:42:45+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"Premature confidence in LLM chains of thought predicts flawed reasoning and is mitigated by progressive confidence shaping, a label-free RL objective that yields accuracy gains on arithmetic, math, and science tasks.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2605.11746","ref_index":49,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"When Reasoning Traces Become Performative: Step-Level Evidence that Chain-of-Thought Is an Imperfect Oversight Channel","primary_cat":"cs.AI","submitted_at":"2026-05-12T08:24:47+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"CoT traces align with internal answer commitment in only 61.9% of steps on average, dominated by confabulated continuations after commitment has stabilized.","context_count":1,"top_context_role":"background","top_context_polarity":"background","context_text":"arXiv preprint arXiv:2510.04040, 2025. [48] Mirac Suzgun, Nathan Scales, Nathanael Schärli, Sebastian Gehrmann, Yi Tay, Hyung Won Chung, Aakanksha Chowdhery, Quoc Le, Ed Chi, Denny Zhou, and Jason Wei. Challenging BIG-Bench tasks and whether chain-of-thought can solve them. InFindings of the Association for Computational Linguistics: ACL 2023, pages 13003-13051, 2023. [49] Sree Harsha Tanneru, Dan Ley, Chirag Agarwal, and Himabindu Lakkaraju. On the hardness of faithful chain-of-thought reasoning in large language models.arXiv preprint arXiv:2406.10625, 2024. [50] Miles Turpin, Julian Michael, Ethan Perez, and Samuel Bowman. Language models don't always say what they think: Unfaithful explanations in chain-of-thought prompting."},{"citing_arxiv_id":"2605.08942","ref_index":4,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"Decomposing and Steering Functional Metacognition in Large Language Models","primary_cat":"cs.CL","submitted_at":"2026-05-09T13:22:47+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"LLMs have linearly decodable functional metacognitive states that causally modulate reasoning when steered via activation interventions.","context_count":1,"top_context_role":"background","top_context_polarity":"background","context_text":"2025. Linear Control of Test Aware- ness Reveals Differential Compliance in Reasoning Models.arXiv preprint arXiv:2505.14617. [3] Sree Harsha Tanneru, Dan Ley, Chirag Agarwal, and Himabindu Lakkaraju. 2024. On the Hardness of Faithful Chain-of-Thought Reasoning in Large Language Models.arXiv preprint arXiv:2406.10625. https://arxiv.org/abs/2406.10625. [4] James Fodor. 2025. Line Goes Up? Inherent Limitations of Benchmarks for Evaluating Large Language Models.arXiv preprint arXiv:2502.14318. https: //arxiv.org/abs/2502.14318. [5] Miles Turpin, Julian Michael, Ethan Perez, and Samuel R. Bowman. 2023. Language Models Don't Always Say What They Think: Unfaithful Explana- tions in Chain-of-Thought Prompting."},{"citing_arxiv_id":"2605.05329","ref_index":22,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"Understanding Annotator Safety Policy with Interpretability","primary_cat":"cs.AI","submitted_at":"2026-05-06T18:01:40+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"Annotator Policy Models learn safety policies from labeling behavior alone, accurately predicting responses and revealing sources of disagreement like policy ambiguity and value pluralism.","context_count":1,"top_context_role":"method","top_context_polarity":"use_method","context_text":"(1) Non-Negative Logistic Regression (NNLR).We restrict logistic regression weights to be non-negative, w ∈R 𝑐 +. Combined with non-negative binary features, this naturally learns a negative bias term, defaulting to safe and only predicting unsafe when positive features accumulate sufficient evidence. (2) Disjunctive Normal Form (DNF).We learn boolean rules in \"ORs-of-ANDs\" form [22]. DNF is especially interpretable for safety: ifanyindividual rule (a conjunction of features) fires, the model predicts unsafe. This aligns with intuition as a single policy violation is sufficient to render text unsafe. Both options yield lightweight, interpretable models that describe where annotators deem text unsafe using onlysafety-relevantfeatures."},{"citing_arxiv_id":"2605.01704","ref_index":25,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"The Reasoning Trap: An Information-Theoretic Bound on Closed-System Multi-Step LLM Reasoning","primary_cat":"cs.CL","submitted_at":"2026-05-03T04:12:21+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"Closed-system multi-step LLM reasoning is subject to an information-theoretic bound where mutual information with evidence decreases, preserving accuracy while eroding faithfulness, with EGSR recovering it on SciFact and FEVER.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2605.08142","ref_index":25,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"Reasoning emerges from constrained inference manifolds in large language models","primary_cat":"cs.LG","submitted_at":"2026-05-02T10:41:20+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":5.0,"formal_verification":"none","one_line_summary":"Reasoning in LLMs emerges from inference dynamics forming constrained low-dimensional manifolds that preserve non-degenerate information volume, rather than from compression alone.","context_count":1,"top_context_role":"background","top_context_polarity":"background","context_text":"evolve within a regime that supports meaningful in- termediate computation, independent of whether a particular output is correct [23, 24]. From this perspective, reasoning health character- izes how a model reasons, not what it knows or how well it performs on a given dataset. Models with sim- ilar task accuracy may operate in fundamentally dif- ferent internal regimes [25, 18], while models with 8 Reasoning emerges from constrained inference manifolds in large language models structurally healthy inference dynamics may fail spe- cific benchmarks due to misalignment, insufficient supervision, or domain mismatch. The diagnostic therefore complements, rather than replaces, exter- nal evaluation [7] by isolating intrinsic properties"},{"citing_arxiv_id":"2605.01048","ref_index":46,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"Compared to What? Baselines and Metrics for Counterfactual Prompting","primary_cat":"cs.CL","submitted_at":"2026-05-01T19:23:33+00:00","verdict":"CONDITIONAL","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"Counterfactual prompting effects on LLMs are often indistinguishable from those caused by meaning-preserving paraphrases, causing most previously reported demographic sensitivities to disappear under proper statistical comparison.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2510.24941","ref_index":26,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"Can Aha Moments Be Fake? Towards Quantifying Decorative and True Thinking in Chain-of-Thought","primary_cat":"cs.LG","submitted_at":"2025-10-28T20:14:02+00:00","verdict":null,"verdict_confidence":null,"novelty_score":null,"formal_verification":null,"one_line_summary":null,"context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2412.16720","ref_index":23,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"OpenAI o1 System Card","primary_cat":"cs.AI","submitted_at":"2024-12-21T18:04:31+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":4.0,"formal_verification":"none","one_line_summary":"OpenAI reports that chain-of-thought reasoning in o1 models enables deliberative alignment, yielding state-of-the-art results on selected safety benchmarks for illicit advice, stereotypes, and jailbreaks.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null}],"limit":50,"offset":0}