1
Steam 喜加一：原价 70 元方块消除对战游戏《Crystal Crisis》免费领
IT之家 9 月 15 日消息，由 Nicalis 开发的方块消除益智对战游戏《Crystal Crisis》目前正在 Steam 平台开启限时免费“喜加一”活动，本作在国区定价为 70 元，限免活动截止至 9 月 22 日 14:59，IT之家附游戏 Steam 商品页（ https://stor…
2
MoRFI: Monotonic Sparse Autoencoder Feature Identification
揭秘大模型幻觉根源：MoRFI方法用单调稀疏自编码器精准锁定知识特征，为可解释性带来新突破。
arXiv:2604.26866v2 Announce Type: replace-cross Abstract: Large language models (LLMs) acquire most of their factual knowledge during the pre-training…
3
OPPO / 一加 ColorOS 16 八月体验升级：桌面 App 小红点秒消除，智能到站提醒上新
ColorOS 16八月更新上线，桌面红点一键清除，智能到站提醒更贴心
IT之家 8 月 17 日消息，OPPO 官方今日发布了 ColorOS 16 八月体验升级，覆盖多款 OPPO 和一加机型。 本次升级主要内容包括： 新增双指外扩清空桌面全部应用角标 ，小红点秒消除，桌面一秒清爽； 锁屏岛全屏音乐歌词 新增支持酷狗概念版 ，喜欢的歌词抬眼就能看到； 交通优行卡 覆…
4
Chain-of-Models: Cross-Model Auditing for Bias-Robust LLM Judges
用多模型交叉审计替代提示工程去偏，为LLM裁判公正性提供新思路。
arXiv:2607.28636v1 Announce Type: new Abstract: LLMs increasingly serve as automated judges, but their judgments remain vulnerable to cognitive biases…
5
Alert Grouping by DrDroid
智能分组告警，彻底消除噪音干扰，让运维团队专注于真正重要的警报
Reduce Your Alerts Noise Completely Discussion | Link
6
Show HN: E– – a language you dial between English and Python
一种介于英语和Python的新语言，消除歧义并让LLM可选执行，重构人机交互语法。
Article URL: https://github.com/frmoded/e-- Comments URL: https://news.ycombinator.com/item?id=48863891 Points: 1 # Comments: 2
7
Show HN: Bais.News – News Neutralization Piepline
一个独立开发者用7个月打造新闻“去偏”管道，聚合多方信源只留事实共识，值得围观。
Solo dev here, been working on this project the last 7 months and I need some feedback. Bais aggregates news stories from a broad range of sources and…
8
FORGE: Fused On-Register Gradient Elimination for Memory-Efficient LLM Training
融合反向传播与优化器阶段的梯度处理，大幅降低LLM训练的内存峰值，突破显存瓶颈的新方案。
arXiv:2606.22932v1 Announce Type: new Abstract: Reverse-mode differentiation computes every weight gradient, writes it to memory, and only then lets t…
9
Ask HN: How do you make AI writing usable?
网友热议如何让AI写作摆脱模板化废话，核心在于提示工程与后处理技巧。
Every time I ask the latest models to write something, it defaults to LLMisms like contrastive negation and long unnecessary lists. Is there a way to …
10
Show HN: Lastwordonearth.com
全球网友共同消除174k单词，直到只剩最后一个，一场慢节奏的社交文字实验。
This is a silly idea I had a couple years ago that I just recently launched. It's a slow, globally shared word "game" (or experiment?) to find the Las…
11
Fair Cognitive Impairment Detection Through Unlearning
用「去学习」技术消除模型偏见，实现公平的认知障碍检测，AI医疗伦理前沿探索。
arXiv:2606.18571v1 Announce Type: new Abstract: Mild Cognitive Impairment (MCI) is a medical condition characterized by a noticeable decline in memory…
12
Tensor-Coord: Algebraic Decomposition of Joint Plan Tensors for Conflict-Free Multi-Agent LLM Planning
代数分解联合计划张量，实现多智能体LLM无冲突规划，开创性解决协同失败难题。
arXiv:2606.16478v1 Announce Type: new Abstract: Large language models (LLMs) remain limited in multi-agent planning because independently generated pl…
13
Disentangling Hallucinations: Orthogonal Semantic Projection for Robust Interpretability
正交语义投影技术，从根源降低大模型幻觉，提升可解释性
arXiv:2606.14758v1 Announce Type: cross Abstract: As Vision-Language Models are increasingly deployed in safety-critical applications, the trustworthi…
14
Diffusion Policy Optimization without Drifting Apart
您提供的内容是一篇学术论文，而非在线工具。请提供您想要推荐的互联网工具名称或链接，我将按照格式为您撰写推荐语、分类、标签和评分。
arXiv:2606.13795v1 Announce Type: new Abstract: RL post-training has become increasingly pivotal for improving diffusion policies, but existing diffus…
15
Building Better Activation Oracles
改进激活预言机训练，解决幻觉和模糊性，可提升残差流激活的可解释性研究价值。
arXiv:2606.02609v1 Announce Type: cross Abstract: Activation Oracles (AOs) are promising methods for interpreting residual stream activations. However…
16
Counterfactually Fair Regression via Optimal Transport
用最优传输理论实现反事实公平回归，为机器学习模型去除偏见提供新方法。
arXiv:2605.28251v1 Announce Type: cross Abstract: We consider the problem of learning a counterfactually fair regressor. We adopt a causal uncertainty…
17
Cearvol’s Wave Design Fights Off Hearing Loss and Aging Stigma
时尚设计打破助听器刻板印象，轻巧耳塞兼顾听力辅助与美观，对抗老化偏见。
Cearvol’s Wave Lite hearing aids take on the stigma of aging with tech-inspired design, but they lack the nuance to be truly useful.
18
消除“罪证”：给写作去除“AI味”的不完全手册（2026版）
上传或粘贴文本后一键检测和消除“AI味”，支持自定义禁用句式与高频词，让你的写作回归真人风格。
图源：公众号「有三思 U Sense」 做个测试。读一下这段话： “熊猫是最可爱的动物，它最爱吃竹子，样子最憨态可掬，是世界上最珍贵的宝藏。” 如果你笑了，或者皱了眉，很明显，你的 “鉴AI雷达” 已经觉醒。 近期把网友笑到打鸣的“豆包体”里，“最”是高频词汇。大…
19
Beyond Superficial Unlearning: Sharpness-Aware Robust Erasure of Hallucinations in Multimodal LLMs
多模态大模型消除幻觉新方法：锐度感知鲁棒擦除，超越浅层遗忘，提升模型可靠性
arXiv:2601.16527v2 Announce Type: replace Abstract: Multimodal LLMs are powerful but prone to object hallucinations, which describe non-existent entit…