Training Language Models for Controlled Stochasticity
by Sruthi Kuriakose and Davide Baldelli
Pseudo-random generation solved the problem of sampling from mathematical distributions on computers. Faced with a similar need in natural language, it feels natural to simply ask an LLM: "Name a random city" but this leads to the disappointing discovery that language models are heavily biased and suffer from mode collapse....
May 2617