用Reddit海量评论追踪话题漂移，看社区讨论如何悄然转向。
arXiv:2608.19133v1 Announce Type: new Abstract: We present a novel application of embedding-based dynamic topic modeling techniques to detect and quan…
首个衡量LLM驱动代码编辑安全漂移的基准，揭示AI改码时隐藏的安全退化风险。
arXiv:2608.15092v1 Announce Type: cross Abstract: In this work, we introduce WeSCE, a benchmark for quantifying security drift in code editing under w…
RAG成本砍6倍的关键，在于提前筛掉不该进大模型的内容，而非单纯优化推理。
Most teams building retrieval augmented generation (RAG) systems for high stakes classification make the same architectural bet: Route every ambiguous…
用空间画布可视化对话节点，直观解决大模型上下文漂移痛点
Article URL: https://treequence.ai Comments URL: https://news.ycombinator.com/item?id=49310343 Points: 4 # Comments: 0
同一份AI指令文件竟有18个版本，揭示配置漂移的惊人真相与治理思路。
Your team runs Claude Code, Cursor, and Copilot. Each of them reads a file before it acts — CLAUDE.md , .cursor/rules/*.mdc , .github/copilot-instruct…
医学影像AI落地最怕模型悄悄“变笨”，MMC+框架用可扩展监控提前捕捉漂移，保障长期可靠性。
arXiv:2410.13174v3 Announce Type: replace-cross Abstract: The integration of artificial intelligence (AI) into medical imaging has advanced clinical d…
大模型生成代码安全研究新基准，揭示领域提示引发的漏洞风险变化
arXiv:2607.25225v1 Announce Type: cross Abstract: LLMs are increasingly used for code generation in critical infrastructure, yet the security effect o…
针对微调后CLIP模型的分布漂移，提出利用图像-文本对齐的漂移感知校准方法，提升跨领域零样本性能。
arXiv:2501.19060v4 Announce Type: replace-cross Abstract: Vision-language models (VLMs), such as CLIP, adapt effectively to downstream tasks through p…
追踪开源AI模型从Hugging Face到GitHub的许可证变更，揭示生态中潜藏的合规风险与治理挑战。
arXiv:2509.09873v2 Announce Type: replace-cross Abstract: Hidden license conflicts in the open-source AI ecosystem pose serious legal and ethical risk…
LLM复合系统中模块分工可能偏离预设角色，揭示“角色漂移”这一新失败模式。
arXiv:2607.21627v1 Announce Type: new Abstract: End-to-end reinforcement learning can improve the accuracy of compound LLM systems, but it does not co…
同步Claude Code与Codex配置，可视化漂移监控，YAML双向无损兼容，附带SHA256校验保障数据一致性。
Article URL: https://github.com/slash9494/ai-config-sync-manager Comments URL: https://news.ycombinator.com/item?id=49044662 Points: 1 # Comments: 0
前沿大模型存在“响应漂移”？这篇论文系统研究了不同版本LLM输出随时间一致性的关键问题。
arXiv:2607.20454v1 Announce Type: cross Abstract: All frontier large language models (LLMs) exhibit response drift -- producing outputs that deviate f…
AI代理在操作中会产生幻觉并出现安全漂移，这篇论文深入剖析了风险机制与防护策略。
arXiv:2607.18366v1 Announce Type: new Abstract: Large language models (LLMs) serving as planners in tool-using autonomous agents introduce dynamic rel…
LLM时代维基维护面临知识漂移与内容矛盾，本文系统梳理了审核机制与系统架构的挑战与解法
Article URL: https://www.glukhov.org/knowledge-management/knowledge-systems-architectures/compiled-knowledge/llm-wiki-maintenance-knowledge-drift/ Com…
追踪大模型后训练中价值对齐的动态变化，揭秘AI价值观漂移的关键时刻
arXiv:2510.26707v2 Announce Type: replace-cross Abstract: As LLMs occupy an increasingly important role in society, they are more and more confronted …
用版本化Markdown规则锁定AI Agent行为，告别会话间一致性漂移
Originally published on hexisteme notes . If you use AI agents daily, you have hit this: the agent gives a correct, well-reasoned answer on Monday. On…
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…
追踪开源聊天LLM十二个版本的信任度变化，揭示模型行为随时间漂移的潜在风险
arXiv:2607.02587v1 Announce Type: cross Abstract: Model cards quote trust-benchmark scores without recording when they were measured, and the same num…
大模型生成的量子代码能否跟上SDK版本迭代？最新基准测试揭示API漂移的严重程度。
arXiv:2607.04072v1 Announce Type: cross Abstract: Large language models can generate plausible quantum code, but it is unclear whether they can reliab…
开源监测大语言模型API无声漂移，提前预警保障输出稳定性。
Article URL: https://github.com/Tania-coder/SEISMOGRAPH Comments URL: https://news.ycombinator.com/item?id=48773957 Points: 1 # Comments: 0