Papers2

#Wasserstein distance

Humans and LLMs Diverge on Probabilistic Inferences

Gaurav Kamath, Sreenath Madathil et al.Feb 26arXiv

Humans often make guesses about the world that are likely but not certain, and this paper studies how humans and AI compare at doing that.

#probabilistic reasoning#uncertainty calibration#natural language inference

Beyond Token-level Supervision: Unlocking the Potential of Decoding-based Regression via Reinforcement Learning

Intermediate

Ming Chen, Sheng Tang et al.Dec 6arXiv

The paper shows that making a model write a number as a sequence of digits and then grading the whole number at the end works better than grading each digit separately.

#decoding-based regression#sequence-level reward#reinforcement learning