3 papers
cs.CL2026
Safe Evolution with Circuit Anchors
Yan Liu, Jie Fu, Tsung-Yi Ho
In biological evolution, unconstrained mutation can lead to catastrophic outcomes: organisms may evolve enhanced capabilities while losing essential functions for survival. Nature'…
cs.CL2024
Elephant in the Room: Unveiling the Impact of Reward Model Quality in Alignment
Yan Liu, Xiaoyuan Yi, Xiaokang Chen +6
The demand for regulating potentially risky behaviors of large language models (LLMs) has ignited research on alignment methods. Since LLM alignment heavily relies on reward models…
cs.CL2024
Negating Negatives: Alignment with Human Negative Samples via Distributional Dispreference Optimization
Shitong Duan, Xiaoyuan Yi, Peng Zhang +5
Large language models (LLMs) have revolutionized the role of AI, yet pose potential social risks. To steer LLMs towards human preference, alignment technologies have been introduce…