Large language models (LLMs) have immense potential in the field of general intelligence but come with significant risks. As a research team at Peking University, we actively focus on alignment techniques for LLMs, such as safety alignment, to enhance the model's safety and reduce toxicity.
Welcome to follow our AI Safety project:
VLA-Arena is an open-source benchmark for systematic evaluation of Vision-Language-Action (VLA) models.
Safe RLHF: Constrained Value Alignment via Safe Reinforcement Learning from Human Feedback
NeurIPS 2023: Safety-Gymnasium: A Unified Safe Reinforcement Learning Benchmark
Alignment with a millennium of moral progress. Spotlight@NeurIPS 2024 Track on Datasets and Benchmarks.
S1-M: Simple Test-time Scaling in Multimodal Reasoning
Loading…
