How do you watermark AI text when there's barely any randomness? 🤔
It is unclear how to effectively watermark low-entropy text— until now. Meet SeqMark at #COLM2026: a watermarking algorithm for constrained generation
Paper: arxiv.org/abs/2601.11629
Blog: nle18.github.io/blog/seqmark…
Work in collaboration with my advisors Alan Ritter and Kartik Goyal.
Having trouble selecting the best prompt for in-context learning? Why not ensemble hundreds of prompts?
Check out our #EMNLP2022 Findings paper Few-shot Anaphora Resolution in Scientific Protocols via Mixtures of In-Context Experts (arxiv.org/abs/2210.03690).
[1/5]
MICE + knowledge distillation is a promising solution for few-shot anaphora resolution, an important task for deep understanding of scientific protocols, since labeled data in this domain may be costly.
[4/5]
