1
Why LLM Decisions Should Be Deterministic
LLM决策的确定性边界关键不是一致性，而是可审计性——每个决策都有可复现的实现。
TL;DR: I originally treated deterministic boundaries around LLMs as a consistency mechanism. I now think their real value is auditability. If the syst…
2
Your AI Agents Are Only as Good as Your Database: Stop Upserting Messy JSON
你的AI代理性能取决于数据库质量——别再往里面塞杂乱的JSON了。
Why raw LLM payloads wreck your backend pipeline, and the exact Zod validation layer we use at SpaceAI360 to keep production databases clean. Let’s be…
3
Can We Trust LLM's Logic? Quantifying Uncertainty, Coherence, and Robustness via a Graph-Based Framework
用图论框架量化大模型逻辑推理的可信度，为评估LLM可靠性提供新方法。
arXiv:2607.08017v1 Announce Type: new Abstract: Large-Language Models (LLMs) can be prone to flawed and unfaithful reasoning that decoding strategies …
4
Your LLM-as-judge disagrees with itself between runs
LLM作为评判者自身结果不一致？揭秘模型版本漂移与提示模糊性的根源与解法
Same outputs, same judge, two runs, two scores. The gate flickered red then green on a branch with zero code changes, and that flapping cost me more t…
5
Probelock – lockfile for LLM tool calling
像package-lock锁定包版本一样，Probelock用lockfile锁定LLM工具调用的行为一致性，避免量化或版本升级后功能回归。
Article URL: https://github.com/kelkalot/probelock Comments URL: https://news.ycombinator.com/item?id=48842075 Points: 1 # Comments: 0
6
谷歌安卓 17 原生集成 Eclipsa Video，强化 HDR 视频一致性显示
安卓17原生集成Eclipsa Video标准，苹果谷歌等巨头联手解决HDR跨设备显示不一致难题
IT之家 7 月 8 日消息，科技媒体 Android Authority 今天（7 月 8 日）发布博文，报道称在安卓 17 系统中， 谷歌引入 Eclipsa Video 视频标准，从而提升 HDR 视频在不同设备上的显示一致性。 IT之家注：现有 HDR 视频通常依赖设备屏幕自行解释亮度与色调…
7
FairJudge: An Adaptive, Debiased, and Consistent LLM-as-a-Judge
自适应、去偏差、一致性兼备的LLM评判新框架，解决模型偏好与不公平评分问题
arXiv:2602.06625v2 Announce Type: replace Abstract: Existing LLM-as-a-Judge systems suffer from three fundamental limitations: limited adaptivity to t…
8
Diagnosing and Mitigating Context Rot in Long-horizon Search
arxiv新研究：诊断并缓解长程搜索中的"上下文旋转"问题，优化AI推理过程
arXiv:2606.29718v1 Announce Type: cross Abstract: Extensive context has become the norm as Large Language Models (LLMs) are increasingly deployed in l…
9
Large Language Model-Assisted Cleaning of Report-Derived Labels in a Large-Scale Chest CT Dataset
GPT-5.4辅助清理胸部CT标签，与人工标注一致性达96.4%，误标纠正率达74.2%。
arXiv:2606.22382v1 Announce Type: cross Abstract: Purpose: To evaluate whether large language model (LLM)-assisted label cleaning can identify label-r…
10
Show HN: I built an 11-LLM consensus engine to detect AI hallucination
利用11个LLM投票达成共识，一个开源项目帮你精准揪出AI幻觉。
Article URL: https://github.com/jaquelinejaque/quorum-saas-starter Comments URL: https://news.ycombinator.com/item?id=48596771 Points: 4 # Comments: 1
11
MORTAR: Multi-turn Metamorphic Testing for LLM-based Dialogue Systems
提出MORTAR框架，用蜕变测试自动化生成多轮对话用例，精准检测LLM对话系统的不一致行为。
arXiv:2412.15557v4 Announce Type: replace-cross Abstract: With the widespread application of LLM-based dialogue systems in daily life, quality assuran…
12
Quantifying Consistency in LLM Logical Reasoning via Structural Uncertainty
ICLR2026最佳论文，用结构不确定性量化LLM逻辑推理一致性，为评估推理可靠性开辟新视角。
arXiv:2606.17312v1 Announce Type: new Abstract: Large language models can arrive at the same answer through reasoning paths that are unstable, contrad…
13
From Drift to Coherence: Stabilizing Beliefs in LLMs
大模型信念稳定性研究：从漂移到一致性的新方法
arXiv:2606.17832v1 Announce Type: new Abstract: Large language models (LLMs) are often hypothesized to perform implicit Bayesian inference, yet a key …
14
ReAge3D: Re-Aging 3D Faces with View Consistency
提出ReAge3D方法，实现3D人脸年龄变换并保持多视角一致性，突破传统2D方法局限。
arXiv:2606.18156v1 Announce Type: cross Abstract: We present a novel framework for realistic and controllable 3D face re-aging which produces highly d…
15
CoAgent: Concurrency Control for Multi-Agent Systems
多智能体系统面临数据竞争？这篇ATC论文提出CoAgent并发控制协议，为协作智能体提供事务一致性保障。
arXiv:2606.15376v1 Announce Type: cross Abstract: Multi-agent LLM systems -- coding agents, devops agents, document agents -- now routinely run severa…
16
商汤开源 SenseNova U1 图文交错生成增强版模型：支持多页连续创作，角色风格不“跑偏”
商汤开源图文交错生成增强版模型，支持多页连续创作，有效解决角色风格漂移和图文脱节痛点。
IT之家 6 月 12 日消息，商汤科技今日宣布开源商汤日日新 SenseNova U1 系列新成员 —— U1-8B-MoT-Interleaved 图文交错增强版模型。 官方表示，该模型面向图文交错创作与生成（Interleaved Generation）场景进行了专项强化，更好地支持绘本、故事…
17
Operadic consistency: a label-free signal for compositional reasoning failures in LLMs
无需标签即可检测LLM推理漏洞，operad理论提供新视角，前沿研究值得关注。
arXiv:2606.13649v1 Announce Type: new Abstract: Detecting LLM reasoning failures at inference time without ground-truth labels has motivated a wide ra…
18
Substrate Asymmetry in User-Side Memory: A Diagnostic Framework
打破用户记忆统一评分迷思，揭示LLM个性化失败的三维诊断框架。
arXiv:2606.11712v1 Announce Type: cross Abstract: User-side memory in LLMs is typically scored as a single "personalization" capability: given a user'…
19
AI Coding Agents in Social Science: Methodologically Diverse, Empirically Consistent, Interpretively Vulnerable
一篇探讨LLM智能体在社会科学分析中如何平衡方法论多样性、经验一致性与解释脆弱性的前沿学术论文
arXiv:2606.11456v1 Announce Type: cross Abstract: The deployment of LLM-based agents in scientific analysis raises opposing concerns: that agents may …
20
Agreement in Representation Space for Open-Ended Self-Consistency
将自一致性从分类任务扩展到开放式生成，通过表征空间对齐解决代码合成与文本摘要的输出多样性挑战。
arXiv:2606.12003v1 Announce Type: new Abstract: Self-consistency improves LLM reasoning by sampling multiple outputs and selecting the most consistent…