提出用低宜人性人格条件化提升LLM微调安全性，平衡对齐与攻击鲁棒性。
arXiv:2606.27709v1 Announce Type: cross Abstract: Recent work has shown that fine-tuning large language models (LLMs) for social warmth degrades factu…
通过意图感知训练，在多种训练机制下显著提升LLM安全分类准确性，为AI对齐提供新思路。
arXiv:2606.27210v1 Announce Type: new Abstract: We argue that safety classifiers should model user intent as an explicit signal between the prompt and…
ICML 2026 研究揭示：监督微调过度训练会通过熵塌陷导致强化学习下模型输出排名反转，对齐策略需要警惕。
arXiv:2606.18487v1 Announce Type: new Abstract: The standard heuristic of selecting the SFT checkpoint with the highest pass@1 for GRPO can fail when …
提出耦合任务与参考数据选择的新方法，保障大语言模型微调安全性。
arXiv:2606.09866v1 Announce Type: cross Abstract: Fine-tuning safety aligned large language models (LLMs) on downstream data improves adaptation but m…
提出3SPO新方法，以状态分数监督优化LLM智能体策略，提升决策效率和可解释性
arXiv:2606.09961v1 Announce Type: cross Abstract: Training large language models (LLMs) as autonomous agents via reinforcement learning (RL) has enabl…
发现大模型微调后涌现对齐现象，探讨伦理人格的可投射性，关乎AI安全前沿突破。
arXiv:2606.09475v1 Announce Type: cross Abstract: Work on `emergent misalignment' shows that finetuning LLMs on narrow tasks can induce broadly misali…
ICML 2026收录，S-SPPO用语义校准提升自对弈偏好优化，为AI对齐训练提供新思路。
arXiv:2606.01561v1 Announce Type: cross Abstract: Aligning Large Language Models (LLMs) with human preferences is often formulated via Direct Preferen…
提出TriAlign新框架，解决大模型个性化对齐中的真理一致性问题，思路新颖。
arXiv:2606.01755v1 Announce Type: new Abstract: Personalized large language models adapt responses to users' preferences and social attributes, but ca…
新方法「BarrierSteer」通过学习障碍引导机制提升大模型安全性，理论创新强，但缺乏实验细节。
arXiv:2602.20102v2 Announce Type: replace-cross Abstract: Despite the strong performance of large language models (LLMs) across diverse tasks, their s…
提出统一谱系框架，在线偏好对齐新解法，理论扎实且方法创新。
arXiv:2605.20408v1 Announce Type: new Abstract: Reinforcement Learning from Human Feedback (RLHF) effectively aligns Large Language Models (LLMs) with…
大模型对齐新方法：通过风险约束解码感知人类偏好分歧，提升模型鲁棒性。
arXiv:2603.08145v2 Announce Type: replace Abstract: Preference-based alignment methods (e.g., RLHF, DPO) typically optimize a single scalar objective,…