研究人员正在开发新的框架来解决大型语言模型（LLM）中的幻觉问题。一种称为“LLM 精神病”的方法将严重的现实边界失败进行分类，并提出了一种评估它们的诊断量表，其中记录了 ChatGPT 5 的发现。另一种方法 KARL 使用强化学习将弃权行为与模型的知识边界对齐，旨在在不牺牲准确性的情况下减少幻觉。此外，PRISM 提供了一个基准，将幻觉分解为知识、推理和指令遵循错误，以帮助理解其根源。对于视觉语言模型，AVES-DPO 专注于自我纠正，利用分布内数据来减轻幻觉。 AI
影响 新的 LLM 幻觉诊断工具和缓解策略可以提高已部署 AI 系统的可靠性和可信度。
排序理由 多篇学术论文介绍了用于理解和缓解 LLM 幻觉的新框架和基准。
AI 生成摘要 · Google Gemini · 来自 8 个来源。 我们如何撰写摘要 →
arXiv:2604.25934v1 Announce Type: cross Abstract: The deployment of large language models (LLMs) as interactive agents has exposed a category of behavioral failure that prevailing terminology, principally hallucination, fails to adequately characterize. This paper introduces LLM …
arXiv:2604.14246v2 Announce Type: replace-cross Abstract: Sparse Mixture-of-Experts (MoE) models have achieved remarkable scalability, yet they remain vulnerable to hallucinations, particularly when processing long-tail knowledge. We identify that this fragility stems from static…
arXiv:2604.22779v1 Announce Type: cross Abstract: Enabling large language models (LLMs) to appropriately abstain from answering questions beyond their knowledge is crucial for mitigating hallucinations. While existing reinforcement learning methods foster autonomous abstention, t…
arXiv:2604.16909v2 Announce Type: replace Abstract: As large language models (LLMs) evolve from conversational assistants into agents capable of handling complex tasks, they are increasingly deployed in high-risk domains. However, existing benchmarks largely rely on mixed queries…
arXiv:2604.24395v1 Announce Type: new Abstract: Large Vision-Language Models (LVLMs) frequently suffer from hallucinations. Existing preference learning-based approaches largely rely on proprietary models to construct preference datasets. We identify that this reliance introduces…
Large Vision-Language Models (LVLMs) frequently suffer from hallucinations. Existing preference learning-based approaches largely rely on proprietary models to construct preference datasets. We identify that this reliance introduces a distributional mismatch between the proprieta…
KARL: RL Framework Cuts LLM Hallucinations Without Accuracy Loss KARL introduces a reinforcement learning framework that dynamically estimates an LLM's knowledge boundary to reward abstention only when appropriate, achieving a superior accuracy-hallucination trade- https:// genti…
Japan Deploys Unitree G1 Robots at Haneda Airport Amid Labor Shortage Japan is testing Unitree G1 and taller humanoid robots at Tokyo Haneda Airport to tackle its labor shortage crisis, marking a real-world deployment of AI-driven robotics. https:// gentic.news/article/japan-depl…