Haifeng Xu Wins Best Paper Award at Leading AI Conference for Pioneering Research on Mechanism Design for LLMs

As this year’s Web Conference is under way, pioneering research work by Assistant Professor of Computer Science and Data Science Haifeng Xu and his collaborators has been announced as the winner for their prestigious Best Paper Award.

The Web Conference is a premier international conference on AI, Information Retrieval, and Web Technology. Since 1989, the Web Conference has focused on the future direction of the World Wide Web, and serves as a venue to present and discuss progress in research, development, standards, and applications of the topics related to the Web.

Xu’s paper, entitled “Mechanism Design for Large Language Models,” was selected from amongst 2008 submissions.

This paper lays out a newly developed method to aggregate language generations from multiple self-interested LLM agents into a single text generation. It does so by accounting for these LLM agents’ self-interests in an incentive-compatible way. As summarized in the meta review, “the review team unanimously finds the paper novel, well-executed, and … has potential to be a landmark paper sparking a new line of research linking LLMs and mechanism design.”

This paper is a joint work with Google Researchers. The technology Xu and his team developed has been tested on Google’s LLM model Bard and Xu reports that it performs very well. According to Xu, the nice (and often very rare) combination of both strong theoretical development and real-world implementation on Bard is probably a key reason for the paper to be named the Best Paper.

Congratulations, Haifeng!

This article was originally published by the Data Science Institute.