颠覆认知：研究发现大模型在知道正确答案时仍会产生幻觉，本质是「承诺失败」而非知识缺失。
arXiv:2605.22007v1 Announce Type: new Abstract: Hallucination is often viewed as a direct consequence of missing knowledge: a model answers incorrectl…
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…
基于真实开发者查询，揭示LLM生成代码中库幻觉的具体风险与数据支撑
arXiv:2509.22202v3 Announce Type: replace-cross Abstract: Large language models (LLMs) now play a central role in code generation, yet they continue t…
PARALLAX方法如何区分真实大模型幻觉与基准构造伪影？Neurips 2026预印本揭示关键。
arXiv:2605.17028v1 Announce Type: new Abstract: Large language models (LLMs) hallucinate with confidence: their outputs can be fluent, authoritative, …
研究发现LLM的幻觉并非随机，而是与模型规模和主题频率呈可预测的比例关系。
arXiv:2605.18732v1 Announce Type: cross Abstract: While scaling laws govern aggregate large language model performance, no scaling law has linked fact…
研究发现视频MLLMs的音频理解实际依赖视觉线索，揭示模型幻觉问题，挑战多模态真实性。
arXiv:2605.16403v1 Announce Type: new Abstract: Despite rapid progress in video-capable MLLMs, we find that their apparent audio understanding in vide…
关注LLM幻觉？HalluScore填补阿拉伯语基准空白，专测大语言模型问答中的幻觉问题。
arXiv:2605.17007v1 Announce Type: new Abstract: Large language models (LLMs) have achieved remarkable progress in natural language generation, but rem…
最新研究揭示2026年大模型代码生成中包幻觉仍存，最高21.7%的虚假包名成slopsquatting攻击新入口
arXiv:2605.17062v1 Announce Type: cross Abstract: Spracklen et al. (USENIX Security '25) showed that code-generating large language models hallucinate…
闭环验证推理突破复杂视觉生成，用可验证的多步推理解决规划幻觉问题，效果惊艳。
arXiv:2605.14876v1 Announce Type: cross Abstract: Despite rapid advancements, current text-to-image (T2I) models predominantly rely on a single-step g…
用脑电图揭示人脑如何识别或误信AI生成幻觉，神经机制首次解密
arXiv:2605.16953v1 Announce Type: cross Abstract: While AI-generated hallucinations pose considerable risks, the underlying cognitive mechanisms by wh…
5个微分类器从压力、谄媚、幻觉实时监测LLM输出行为，基于心理学漏洞框架
Hey HN! We're Dr. Kashyap Thimmaraju and Giuseppe Canale from Silicon Psyche. We've built Posture Sequence Analysis (PSA), a behavioural health monito…
用强化学习提升长文本生成中置信度表达，直击大模型幻觉难题。
arXiv:2505.23912v2 Announce Type: replace-cross Abstract: Hallucination remains a major challenge for the safe and trustworthy deployment of large lan…
OpenAI最新研究揭示语言模型幻觉根源，用更优评估提升AI可靠性与安全性。
OpenAI’s new research explains why language models hallucinate. The findings show how improved evaluations can enhance AI reliability, honesty, and sa…
程序员被LLM误导，亲手破坏了完美PR，反思人与AI竟有相同“bug”
Article URL: https://www.droppedasbaby.com/posts/2602-02/ Comments URL: https://news.ycombinator.com/item?id=48178732 Points: 1 # Comments: 0
扩散大模型幻觉检测新方法：利用隐藏证据验证，捕捉生成轨迹中的幻觉信号，突破传统输出检测局限。
arXiv:2604.26139v2 Announce Type: replace Abstract: Diffusion large language models generate text through multi-step denoising, where hallucination si…
揭示视觉语言模型语言压倒视觉导致幻觉的根源，提出几何去偏方法提升可靠性。
arXiv:2605.08245v3 Announce Type: replace-cross Abstract: Vision-Language Models (VLMs) increasingly power high-stakes applications, from medical imag…
8B参数模型专用生物实验protocol生成，步骤顺序不乱、剂量无幻觉，超越GPT-4o，已登ICLR2026。
超越GPT-4o
新方法通过正交分解问答表征，实现高效幻觉检测，兼顾准确率与分布迁移鲁棒性。
arXiv:2605.14449v1 Announce Type: cross Abstract: Hallucination detection in large language models (LLMs) requires balancing accu racy, efficiency, an…
DeepSeek 官方回应 <think> 字符触发异常：系模型幻觉，非安全问题或隐私泄露。
IT之家 5 月 19 日消息，今日 DeepSeek 就“<think> 字符触发模型异常回复”发布说明，官方称：属于特殊字符引发的模型幻觉，不涉及安全问题或隐私泄露。 IT之家附声明全文如下： 关于 近期，我们关注到有用户反馈，在与 DeepSeek 模型的对话中输入“<thi…