Large language models (LLMs) have demonstrated strong performance across a wide range of reasoning tasks, yet their understanding of joint and conditional probability distributions remains largely unexplored. In this work, we present a comprehensive evaluation of LLMs’ ability to reason about discrete joint and conditional probability distributions involving multiple random variables. Given observations drawn from a distribution, we evaluate models on three carefully designed tasks—mode identification, maximum likelihood estimation, and sample generation—which together capture fundamental skills required for probabilistic reasoning. Our study reveals both capabilities and limitations of current LLMs. In sample generation task, some models produce empirical distributions that closely match the target distribution despite exhibiting limited sample independence. We further observe that model performance is highly sensitive to representation: replacing explicit frequency information with raw samples leads to large accuracy drops, and even minor changes to outcome labels can further degrade performance. Together, these findings provide new insights into how LLMs reason about discrete joint and conditional probability distributions and highlight their strengths and limitations in certain probabilistic reasoning settings.

In this work, we study how the finetuning stage in the pretrain-finetune framework changes the behavior of a pretrained neural language generator. We focus on the transformer encoder-decoder model for the open-domain dialogue response generation task. Our major finding is that after standard finetuning, the model forgets some of the important language generation skills acquired during large-scale pretraining. We demonstrate the forgetting phenomenon through a set of detailed behavior analysis from the perspectives of knowledge transfer, context sensitivity, and function space projection. As a preliminary attempt to alleviate the forgetting problem, we propose an intuitive finetuning strategy named “mix-review”. We find that mix-review effectively regularizes the finetuning process, and the forgetting problem is alleviated to some extent. Finally, we discuss interesting behavior of the resulting dialogue model and its implications.