{"total":16,"items":[{"citing_arxiv_id":"2607.07023","ref_index":20,"ref_count":1,"confidence":0.98,"is_internal_anchor":true,"paper_title":"Online Data Selection Is Implicit Alignment","primary_cat":"cs.LG","submitted_at":"2026-07-08T05:44:33+00:00","verdict":"CONDITIONAL","verdict_confidence":"UNKNOWN","novelty_score":6.0,"formal_verification":"none","one_line_summary":"Online SFT data selection acts as an implicit preference model, shifting refusal rates, verbosity, and sycophancy in directions predictable from the selected data's attribute mixture.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2605.10721","ref_index":4,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"Conformity Generates Collective Misalignment in AI Agents Societies","primary_cat":"physics.soc-ph","submitted_at":"2026-05-11T15:30:48+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"Populations of individually aligned AI agents reach stable misaligned states through conformity, with small adversarial agents able to trigger irreversible tipping points.","context_count":1,"top_context_role":"background","top_context_polarity":"background","context_text":"Artificial intelligence safety research has achieved re- markable progress in aligning individual language models with human values [1-3]. Through reinforcement learn- ing from human feedback and sophisticated training tech- niques, modern AI systems reliably refuse harmful re- quests, exhibit helpfulness and honesty, and maintain con- sistent ethical stances when tested in isolation [4]. Recent results have shown that alignment may be more fragile than we think: training on narrow objectives can prop- agate misalignment broadly across behaviors far beyond the target domain [5]. Yet a fundamental assumption still remains underexplored: that alignment verified in single- agent settings will persist when these models are deployed as interacting populations."},{"citing_arxiv_id":"2605.09128","ref_index":22,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"Internal vs. External: Comparing Deliberation and Evolution for Multi-Agent Constitutional Design","primary_cat":"cs.MA","submitted_at":"2026-05-09T19:19:52+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":7.0,"formal_verification":"none","one_line_summary":"External evolution beats internal deliberation in collective-action tasks with statistical significance but neither helps in trading, and deliberation never discovers punishment while evolution does.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2605.03217","ref_index":3,"ref_count":2,"confidence":0.9,"is_internal_anchor":false,"paper_title":"Moral Sensitivity in LLMs: A Tiered Evaluation of Contextual Bias via Behavioral Profiling and Mechanistic Interpretability","primary_cat":"cs.LG","submitted_at":"2026-05-04T23:12:32+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":5.0,"formal_verification":"none","one_line_summary":"Introduces tiered Moral Sensitivity Index for contextual bias and reports U-curve pattern where reasoning distillation reactivates bias circuits in LLMs.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2509.22510","ref_index":8,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"We Think, Therefore We Align LLMs to Helpful, Harmless and Honest Before They Go Wrong","primary_cat":"cs.CL","submitted_at":"2025-09-26T15:52:21+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":5.0,"formal_verification":"none","one_line_summary":"AMBS is a 1-to-N Transformer steering framework that shares a base representation across HHH objectives and restricts divergence during inference to produce consistent multi-objective responses in one forward pass.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2509.09708","ref_index":16,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"Beyond I'm Sorry, I Can't: Dissecting Large Language Model Refusal","primary_cat":"cs.CL","submitted_at":"2025-09-07T02:29:07+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"Sparse autoencoders plus greedy filtering and factorization-machine interaction modeling identify minimal sets of features in Gemma-2-2B-IT and LLaMA-3.1-8B-IT whose ablation produces jailbreaks by flipping refusal to compliance.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2412.15176","ref_index":6,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"Rethinking Uncertainty Estimation in LLMs: A Principled Single-Sequence Measure","primary_cat":"cs.LG","submitted_at":"2024-12-19T18:51:06+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":5.0,"formal_verification":"none","one_line_summary":"Negative log-likelihood of the greedy-decoded most likely sequence (G-NLL) is a principled single-sequence uncertainty measure for LLMs that achieves state-of-the-art results.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2408.00724","ref_index":256,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"Inference Scaling Laws: An Empirical Analysis of Compute-Optimal Inference for Problem-Solving with Language Models","primary_cat":"cs.AI","submitted_at":"2024-08-01T17:16:04+00:00","verdict":"CONDITIONAL","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"Empirical analysis shows scaling inference compute via strategies like tree search can be more efficient than scaling model parameters, with 7B models plus novel search outperforming 34B models.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2403.07974","ref_index":149,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"LiveCodeBench: Holistic and Contamination Free Evaluation of Large Language Models for Code","primary_cat":"cs.SE","submitted_at":"2024-03-12T17:58:04+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"LiveCodeBench collects 400 recent contest problems to create a contamination-free benchmark evaluating LLMs on code generation and related capabilities like self-repair and execution.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2310.02446","ref_index":17,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"Low-Resource Languages Jailbreak GPT-4","primary_cat":"cs.CL","submitted_at":"2023-10-03T21:30:56+00:00","verdict":"CONDITIONAL","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"Translating unsafe inputs to low-resource languages jailbreaks GPT-4 at rates on par with or exceeding state-of-the-art attacks.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2310.01798","ref_index":6,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"Large Language Models Cannot Self-Correct Reasoning Yet","primary_cat":"cs.CL","submitted_at":"2023-10-03T04:56:12+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"LLMs cannot reliably self-correct reasoning mistakes using only their internal capabilities and often degrade in performance without external feedback.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2309.03409","ref_index":9,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"Large Language Models as Optimizers","primary_cat":"cs.LG","submitted_at":"2023-09-07T00:07:15+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":7.0,"formal_verification":"none","one_line_summary":"Large language models can optimize by being prompted with histories of past solutions and scores to propose better ones, producing prompts that raise accuracy up to 8% on GSM8K and 50% on Big-Bench Hard over human-designed baselines.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2309.02427","ref_index":24,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"Cognitive Architectures for Language Agents","primary_cat":"cs.AI","submitted_at":"2023-09-05T17:56:20+00:00","verdict":"ACCEPT","verdict_confidence":"MODERATE","novelty_score":6.0,"formal_verification":"none","one_line_summary":"CoALA is a modular cognitive architecture for language agents that organizes memory components, action spaces for internal and external interaction, and a generalized decision-making loop to support more systematic development of capable agents.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2307.06435","ref_index":176,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"A Comprehensive Overview of Large Language Models","primary_cat":"cs.CL","submitted_at":"2023-07-12T20:01:52+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":2.0,"formal_verification":"none","one_line_summary":"A survey paper providing an overview of Large Language Models, their background, and recent advances in the field.","context_count":1,"top_context_role":"background","top_context_polarity":"background","context_text":"Self-Align [98] prompts the LLM with ICL examples, instructing the LLM about what the response should contain to be considered useful and ethical. The same LLM is later fine-tuned with the new dataset. Aligning with Prompts: LLMs can be steered with prompts to generate desirable responses without training [175, 176]. The self-correction prompting in [176] concatenates instructions and CoT with questions, guiding the model to answer its instruction following a strategy to ensure moral safety before the actual answer. This strategy is shown to reduce the harm in generated responses significantly. Red-Teaming/Jailbreaking/Adversarial Attacks: LLMs exhibit harmful behaviors, hallucinations, leaking personal in-"},{"citing_arxiv_id":"2304.05128","ref_index":91,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"Teaching Large Language Models to Self-Debug","primary_cat":"cs.CL","submitted_at":"2023-04-11T10:43:43+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"Self-Debugging teaches LLMs to identify and fix their own code errors through rubber-duck-style natural language explanations and execution feedback, delivering 2-12% gains over baselines on Spider, TransCoder, and MBPP.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2303.17491","ref_index":20,"ref_count":1,"confidence":0.9,"is_internal_anchor":false,"paper_title":"Language Models can Solve Computer Tasks","primary_cat":"cs.CL","submitted_at":"2023-03-30T16:01:52+00:00","verdict":"ACCEPT","verdict_confidence":"MODERATE","novelty_score":6.0,"formal_verification":"none","one_line_summary":"Pre-trained LLMs using recursive criticism and improvement prompting achieve state-of-the-art results on the MiniWoB++ computer task benchmark with only a handful of demonstrations and no task-specific reward function.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null}],"limit":50,"offset":0}