揭示思维链并非万能，实证剖析串行深度瓶颈对LLM推理的边界影响。
arXiv:2608.09942v1 Announce Type: cross Abstract: It is widely assumed that chain-of-thought (CoT) prompting universally improves LLM reasoning. We in…
AI缺乏人类的常识、情感和创造性推理，这篇文章通过具体案例剖析了LLM与人类技术工作者的本质差距
When we talk about AI these days, we're usually talking about large language models. To an LLM, the only reality it knows is data, and to be more spec…
AI检测工具对非母语写作者存在高达3.2%的误报率，教育领域依赖技术检测的隐患值得警惕。
Article URL: https://forta.com/blog/ai-in-the-classroom-part-2-the-fears-real-and-imagined Comments URL: https://news.ycombinator.com/item?id=49089268…
最新arXiv研究揭秘：LLM不止能写SQL，调试企业级ETL代码才是真考验
arXiv:2601.18119v2 Announce Type: replace Abstract: SQL is central to enterprise data engineering, yet generating fully correct SQL code in a single a…
AI无法理解“家”的含义，这句话道出了大模型与人类创造力的本质鸿沟。
There’s a line from Rutherford Craze’s latest note that hit me like a piano tumbling out of a six story building: When we ask Midjourney to draw an ox…
DeepMind的AlphaZero在稀疏奖励游戏中表现有限，论文提出辅助监督方法提升性能。
arXiv:2607.08984v1 Announce Type: cross Abstract: AlphaZero has demonstrated that a neural-guided Monte Carlo Tree Search can achieve superhuman perfo…
大型语言模型在形式化验证的代数结构上遭遇机制级路由失败，揭示推理深层次缺陷。
arXiv:2607.04534v1 Announce Type: new Abstract: We present an empirical study of structural routing failure in large language models (LLMs) over a for…
探究大语言模型在结构化CBT情感推理中的局限性，发现其虽能流畅回应却陷入无效验证模式。
arXiv:2607.02885v1 Announce Type: cross Abstract: Cognitive Behavioral Therapy (CBT) provides a structured framework for understanding a user's mental…
当多元价值并存，成对比较为何失效？一篇挑战决策底层逻辑的硬核论文，值得深读。
arXiv:2607.02672v1 Announce Type: new Abstract: Local pairwise comparisons are a standard tool for learning how people want decision rules to work, e.…
从零搭建LLM裁判系统，与人类评分对比，揭露看似好用实则暗藏陷阱的评估局限。
In Part 1 the model's job was to pick one of 77 labels, so I could check it with == . But most real LLM output isn't like that — it's a paragraph, a s…
用一条简单汽车清单提问，揭示当前AI模型常识推理的短板。
Claude: please give me a list of famous cars named after fish, like the Plymouth Barracuda for example. Comments URL: https://news.ycombinator.com/ite…
GPT-5等大模型在简单的人类注意力测试中败下阵来，Stroop效应暴露出AI注意力架构的致命短板。
Article URL: https://scitechdaily.com/even-gpt-5-failed-this-human-attention-test/ Comments URL: https://news.ycombinator.com/item?id=48626330 Points:…
Claude Opus因训练数据限制，无法识别新颖代码模式，陷入重构循环。
I'm having a terrible problem with claude opus constantly clobering some of my codebase ..because I'm doing some things that are novel and arent incld…
展望2026年AI编码助手的能力边界，揭示其在关键性能优化和复杂业务逻辑上的局限。
AI coding assistants are now part of daily developer life. They write boilerplate, explain strange errors, generate tests, and save hours of repetitiv…
LLM在理解棋盘游戏规则上表现糟糕，这个被忽视的bug揭示了大模型逻辑推理的深层缺陷。
I've come across a use case LLMs seem inexplicably bad at: reading and understanding board game rules. It seems like it should be something an LLM wou…
最新研究揭示大模型在法律推理中的忠实性困境，实证分析求解与自动形式化能力的边界
arXiv:2606.16118v1 Announce Type: new Abstract: Large Language Models (LLMs) achieve strong performance on reasoning tasks, but whether this reflects …
一位Hacker News用户断言：AGI无法靠当前LLM实现，需“从零设计”全新的DNA级架构，就像从鸟翼到飞机的跃迁，冲击传统认知。
Fast won't be necessary, and we won't reach AGI with how LLM's work. We will need a complete reconstruction of the architecture. We need a DNA based s…
探讨人类相比AI仍具优势的领域，从创造力到复杂决策，引发对AI能力的再思考。
Comments URL: https://news.ycombinator.com/item?id=48517839 Points: 2 # Comments: 4
顶级会议ICML 2026论文，揭示大模型借助心理意象进行推理时的能力边界，认知科学视角下的深刻洞见。
arXiv:2602.02465v2 Announce Type: replace Abstract: Frontier models are transitioning from multimodal large language models (MLLMs) that merely ingest…
揭秘LLM在科学新颖性评估中的致命短板，颠覆对AI判别的盲目信任
arXiv:2606.12071v1 Announce Type: cross Abstract: LLMs are increasingly used to generate and judge scientific ideas. This makes novelty evaluation a c…