Haichuan Wang

New methods enhance LLM alignment during inference

Researchers have developed new methods for improving the alignment of large language models during inference. One approach, BlendIn, uses probabilistic model blending to integrate knowledge from multiple models, stabili…