1
工信部严查汽车生产一致性：2025 年 10 余款车型参数不符，涉及东风、中联重科等
IT之家 9 月 5 日消息，工业和信息化部装备工业一司昨晚通报了 2025 年度汽车产品生产一致性监督检查的部分典型案例。此次检查旨在规范汽车产业竞争秩序，守牢汽车产品生产一致性和质量安全底线。 通报显示，检查发现的问题共涉及五类情形。其中，样车个别参数与《道路机动车辆生产企业及产品公告》（简称《…
2
Decomposing Wrong-Consensus Agreement in LLM Self-Consistency: A GPT-4.1 Case Study
多数投票为何在难题上翻车？新指标Gamma精准拆解LLM自洽性失效机制
arXiv:2608.18795v1 Announce Type: cross Abstract: Majority voting over multiple LLM samples is widely used to raise answer accuracy, yet its gain vari…
3
How I Built Intent Detection for a Chat-Based AI Comic Tool
聊天式AI漫画工具，用智能意图识别兜底模糊输入，角色一致性靠状态管理，5次免费体验免绑卡。
Most AI comic generators are single-shot: prompt in, image out. That works fine until you need the same character to show up consistently across five …
4
Reinforcing Consistency in Video MLLMs with Structured Rewards
视频多模态模型常“看走眼”？这项研究用结构化奖励强化推理一致性，为视频理解立新标尺。
arXiv:2604.01460v2 Announce Type: replace Abstract: Multimodal large language models (MLLMs) have achieved remarkable progress in video understanding.…
5
Eval4Sim: An Evaluation Framework for Persona Simulation
首个面向LLM人格模拟的评估框架，专测虚拟人设是否忠实还原真实人类行为。
arXiv:2603.02876v2 Announce Type: replace Abstract: Large Language Model personas, explicit profiles specifying a user's attributes, preferences, and …
6
Incoherent by Design? On the Moral Self-Consistency of LLMs
揭示大模型道德观内在矛盾，88页深度研究拷问AI价值对齐的可靠性。
arXiv:2608.15354v1 Announce Type: new Abstract: LLMs are increasingly used in morally sensitive contexts, yet it is unclear whether they apply ethical…
7
PL-Guard: Probabilistic Logic Reasoning for LLM Guardrails
用概率逻辑推理给大模型护栏做政策一致性判断，比提示词和裁判模型更严谨。
arXiv:2608.15673v1 Announce Type: cross Abstract: Large language model guardrails can be viewed as policy-consistency problems: a system must determin…
8
Model your business once – for humans and AI alike
一套模式让人类与AI同读业务，建模一次，处处可用。
Article URL: https://klr-pattern.github.io/nexusx/ Comments URL: https://news.ycombinator.com/item?id=49310720 Points: 2 # Comments: 0
9
Making Your LLMs More Objective: Stabilizing LLM Safety Behavior Across Traits with Trait-Invariant Safety Tuning
让大模型安全行为不再因用户特质波动，Trait-Invariant Safety Tuning 实现跨场景稳定拒答。
arXiv:2608.11705v1 Announce Type: new Abstract: Aligned large language models (LLMs) are expected to exhibit safety behavior based on the content of t…
10
I found 18 versions of the same CLAUDE.md on one laptop
同一份AI指令文件竟有18个版本，揭示配置漂移的惊人真相与治理思路。
Your team runs Claude Code, Cursor, and Copilot. Each of them reads a file before it acts — CLAUDE.md , .cursor/rules/*.mdc , .github/copilot-instruct…
11
Counterfactual Benchmarking and Training for Factuality Consistency and Order-Robust Grounded Reasoning in LLMs over Heterogeneous Knowledge
用反事实基准和训练方法破解大模型事实一致性与异质知识推理的排序难题
arXiv:2608.07838v1 Announce Type: new Abstract: Large language models (LLMs) have increasingly supported response generation grounded in user-provided…
12
Position: It's Time to Optimize LLMs for Self-Consistency
别再只追准确率，让大模型为自一致性而优化——这份立场论文提出了重塑训练范式的关键方向。
arXiv:2608.05188v1 Announce Type: cross Abstract: Despite ever-increasing sophistication in language model (LM) pre- and post-training pipelines, many…
13
华为昇腾宣布已支持 RL 训推一致性，实测获得最高 60% 性能收益
华为昇腾实现RL训推一致性，关键算子统一路径，实测性能提升最高60%，AI训练推理更高效。
IT之家 8 月 6 日消息，强化学习已成为大模型训练主流范式，推动模型技术向行业核心场景工程化落地。其中，训推一致至关重要，因推理与训练过程耦合，基础设施差异易放大不确定性。 华为计算官方 8 月 5 日宣布，昇腾基于 Ascend C 编程语言提供完整训推一致算子集，在 Qwen3-30B Mo…
14
Verifiable Checks for Business Rule Consistency
用形式化方法验证业务规则一致性，让规则冲突在运行前无所遁形。
arXiv:2608.00396v1 Announce Type: cross Abstract: Maintaining consistency between natural language documentation of business rules and their evolving …
15
TrAC: Trace-Conditioned Answer Consistency for Efficient Uncertainty Quantification in LLMs
用追溯条件优化答案一致性，为LLM不确定性量化提供高效新方案，值得关注。
arXiv:2608.00422v1 Announce Type: new Abstract: Large language models (LLMs) can generate fluent reasoning traces that nevertheless lead to incorrect …
16
PixVL: Self-Supervised Training of Pixel-Level MLLMs via a Unified Mask--Text Consistency Cycle
像素级多模态大模型新训练范式，掩码—文本一致性循环让模型精修细节，值得细读。
arXiv:2608.01354v1 Announce Type: new Abstract: Recent studies develop pixel-level multimodal large language models (MLLMs) that support both Region S…
17
MUD as AI Evaluation and LLM-judge distortion in ways aggregate κ misses
用MUD搭建AI评估场，揭露LLM裁判在聚合kappa指标下隐藏的失真，挑战传统评估框架的盲区。
Article URL: https://www.lesswrong.com/posts/GPbWyHgx9hCLMdAjc/mud-as-ai-evaluation-and-llm-judge-distortion-in-ways Comments URL: https://news.ycombi…
18
工信部赴奇瑞、蔚来、江汽开展监督检查，现场随机抽取样车、动力电池等
工信部突查车企样车与电池，生产一致性监管再升级，质量安全底线不容挑战。
IT之家 7 月 31 日消息，工信部今日发文称，7 月 30 日至 31 日，为进一步规范汽车产业竞争秩序、提升汽车产品生产一致性和质量安全水平，工业和信息化部装备工业一司赴 奇瑞汽车股份有限公司、蔚来汽车科技（安徽）有限公司、安徽江淮汽车集团股份有限公司 开展道路机动车辆生产企业产品安全保障能力…
19
The Reliability of LLMs for Medical Diagnosis: An Examination of Consistency, Manipulation, and Contextual Awareness
研究评估了Google Gemini和ChatGPT-4o在医疗诊断中的一致性、抗操纵性和上下文敏感性，揭示LLM可靠性关键挑战。
arXiv:2503.10647v2 Announce Type: replace-cross Abstract: This study evaluated the diagnostic reliability of two Large Language Models (LLMs), Google …
20
Model-Driven Requirements Configuration with Three-Valued Uncertainty Scoring
提出三值不确定性评分方法，模型驱动下消除LLM生成需求的逻辑不一致与结构问题。
arXiv:2607.26220v1 Announce Type: cross Abstract: Context: Large Language Models (LLMs) offer natural-language flexibility for automated requirements …