Intractable Inference

This research paper discusses the limitations of autoregressive large language models (LLMs) in compressing knowledge from training data. The authors propose a solution by using amortized Bayesian inference to sample…