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…
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
On-Policy蒸馏中策略漂移的块级门控机制，为强化学习知识迁移提供新思路。
arXiv:2606.24084v1 Announce Type: cross Abstract: On-policy distillation (OPD) trains a student policy using teacher signals computed on trajectories …
多LLM API用户必看：如何快速察觉响应变慢、错误率升高或模型漂移？这是普遍难题还是运气不佳？
If you are building on top of multiple LLM APIs or even a single one amongst OpenAI, Claude, Gemini, etc. what do you do when the API starts degrading…
大模型信念稳定性研究：从漂移到一致性的新方法
arXiv:2606.17832v1 Announce Type: new Abstract: Large language models (LLMs) are often hypothesized to perform implicit Bayesian inference, yet a key …
思维链推理预算越高，大模型反而越“迷之自信”？新研究揭示校准漂移现象与CABStop缓解方案
arXiv:2606.11211v1 Announce Type: cross Abstract: The ability of large language models (LLMs) to express calibrated uncertainty is important for safe …
自动检测并缓解AI模型漂移的开源控制器，保障模型长期可靠。
Article URL: https://github.com/pberlizov/adaptive-reliability-layer Comments URL: https://news.ycombinator.com/item?id=48463969 Points: 1 # Comments:…
IT之家 6 月 8 日消息，据法新社今天报道，任天堂现已同意支付 3500 万欧元（IT之家注：现汇率约合 2.74 亿元人民币）罚款，以解决法国消费者保护组织针对 Switch 1 游戏机手柄故障提出的指控。 据报道，该案最初由法国消费者权益保护组织（DGCCRF）于 2020 年提起。该组织指…
一种零样本的LLM提示注入防御方法，通过检测嵌入漂移实现轻量级保护，已被NeurIPS研讨会接收。
arXiv:2601.12359v1 Announce Type: cross Abstract: Prompt injection attacks have become an increasing vulnerability for LLM applications, where adversa…
一个符号化的AI安全层，专治价值观漂移和对齐难题，开源研究项目值得关注。
Article URL: https://github.com/Mankirat47/Dao-Heart-3.13 Comments URL: https://news.ycombinator.com/item?id=48442582 Points: 1 # Comments: 0
顶会KDD 2026提出HyperPatch，解决大模型在n元结构漂移下的顺序知识编辑难题，新范式值得关注
arXiv:2606.03179v1 Announce Type: new Abstract: Large Language Models (LLMs) rely on Knowledge Editing (KE) to maintain temporal validity, yet real-wo…
递归AI系统意义漂移的专业术语表，揭示语义忠诚度评估框架的关键概念。
Article URL: https://github.com/therealitydrift/semantic-fidelity-lab/blob/main/01_Canonical_Frameworks/Glossaries/semantic-fidelity-canonical-lexicon…
大模型交易代理会“预判”崩盘？研究发现失败前有独特信号模式，如规划嵌入漂移和有效秩收缩。
arXiv:2605.28850v1 Announce Type: new Abstract: We study behavioral alignment and representation dynamics of large language model (LLM) agents in fina…
IT之家 5 月 29 日消息，由独立开发者 RewindApp 打造的街机风漂移竞速游戏 Drift86 目前正在 Steam 平台开启限时免费领取“喜加一”活动， 本作在国区定价为 11 元 ，玩家可以在 6 月 4 日前免费领取，IT之家附游戏商品页（ https://store.steamp…
用结构抽象作为归纳偏置，解决非平稳语言模型训练中的分布漂移问题，为AI训练提供新思路。
arXiv:2603.17198v2 Announce Type: replace Abstract: A foundational principle in cognitive science holds that intelligent agents do not learn by storin…
大模型代理中记忆偏见导致工具调用偏离，揭示生产系统潜在新陷阱
arXiv:2605.24941v1 Announce Type: cross Abstract: Modern LLM agents combine long-term memory for personalization with tool-calling interfaces for taki…
量化并强制LLM代理在长期任务中保持目标一致性，防止“目标漂移”的新方法。
arXiv:2605.23574v1 Announce Type: new Abstract: Long-horizon language agents can make many plausible local tool calls yet fail to persist until a requ…
Qwen3.5-14B 微调模型专攻 MCP 工具路由决策，BFCL 基准目标≥90% 但遭遇数据集缺失，现场漂移修复实录。
Article URL: https://github.com/dcostenco/prism-coder Comments URL: https://news.ycombinator.com/item?id=48210577 Points: 2 # Comments: 0