Robustness of LLMs with Xiangliang Zhang and Pin-Yu Chen скачать в хорошем качестве

Robustness of LLMs with Xiangliang Zhang and Pin-Yu Chen

Our ND–IBM Research collaboration team is pushing the frontier of safe and trustworthy AI. From lab safety benchmarks for LLMs (LabSafety Bench), to multi-dimensional trust assessments (TrustGen), bias detection in LLM-as-a-Judge, and controllable synthetic dataset generation (DataGen), our work provides critical tools for responsible AI deployment. These efforts ensure that AI systems are not only powerful—but also reliable, fair, and safe in real-world scientific and societal applications. We have the following papers as a collaboration outcome, between ND (Zhang’s group) and IBM Research (Pin-Yu Chen, Tian Gao). Two PhD students, Yujun Zhou and Yue Huang, are leading these works. 1. LabSafety Bench: Homepage: https://yujunzhou.github.io/LabSafety... Paper: https://arxiv.org/abs/2410.14182  2. TrustGen: Homepage: https://trustgen.github.io/ Paper: https://arxiv.org/pdf/2502.14296 3. Justice or Prejudice? Quantifying Biases in LLM-as-a-Judge Homepage: https://llm-judge-bias.github.io/ Paper: https://arxiv.org/abs/2410.02736 4. DataGen: Homepage: https://github.com/HowieHwong/DataGen Paper: https://arxiv.org/abs/2406.18966