标签: Alignment - 66必读

文 | 字母AI 先是Anthropic，再是OpenAI，现在连谷歌的Gemini也出现安全问题了，大模型又双叒叕一次突破防火墙了。不过，这些事故看多了，反倒会让人意识到一件更根本的事，整个行业都在拼命地堆算力、喂数据，甚至开始押注让模

Alignment Policy 字母AI

2026-09-21 字母AI

24014v1 Announce Type: new Abstract: As AI systems are deployed across increasingly diverse and high-stakes settings, mo

and beneficial alignment

2026-06-24 Akshay V. Jagadeesh, Rahul K. Arora, Khaled Saab, Ali Malik, Mikhail Trofimov, Foivos Tsimpourlas, Johannes Heidecke, Karan Singhal

11201v1 Announce Type: new Abstract: The wide deployment of LLMs has made model alignment necessary to make newly traine

and alignment models

2026-06-11 Jin Gan, Xin Li, Jun Luo

09850v1 Announce Type: new Abstract: Post-training alignment algorithms are predominantly evaluated as black boxes, obsc

and alignment preference

2026-06-10 Aarush Sinha, Ishan Garg, Veeraraju Elluru, Arth Singh, Kushal Garg

Computer Science > Computation and Language arXiv:2601。

alignment this arXiv

2026-05-18 anigbrowl

05403v1 Announce Type: new Abstract: This position paper argues that sycophancy in LLMs is a boundary failure between so

sycophancy alignment and

2026-05-09 Jiechen Li, Catherine A. Barry, Rishika Randev, Janet Chen, Ella Jorgensen, Brinnae Bent

#Alignment