AI-Fun & ELLIS Invited Speaker Series | Jin Zhu
Online
Wednesday 14th October 2026
Our invited speaker on 14 October will be Jin Zhu from the University of Birmingham.
Title: Statistics-Powered Detection of LLM-Generated Text
Abstract: Large language models (LLMs) such as ChatGPT, Gemini, and Claude enable the rapid and large-scale generation of human-like text. Their impact is everywhere, from education and academia to professional work and everyday life. Despite these benefits, the widespread deployment of LLMs also raises concerns about misinformation, academic dishonesty, and the integrity of digital content. As a result, detecting LLM-generated text has become an active research topic in machine learning over the past few years. However, statistical perspectives on this problem remain largely underexplored. In this talk, I will present several of my recent works on developing statistical methods for detecting LLM-generated text, with the aim of providing principled and efficient tools for identifying such content. The talk will cover methods presented at recent machine learning conferences, including NeurIPS and ICLR. An interactive demonstration of one of the proposed methods is available at https://huggingface.co/spaces/stats-powered-ai/StatDetectLLM [huggingface.co]
Bio: Dr Jin Zhu is an Assistant Professor in the School of Mathematics at the University of Birmingham. His research focuses on large language models, reinforcement learning, and high-dimensional data, particularly on developing computationally efficient methods with statistical guarantees.
Title: Statistics-Powered Detection of LLM-Generated Text
Abstract: Large language models (LLMs) such as ChatGPT, Gemini, and Claude enable the rapid and large-scale generation of human-like text. Their impact is everywhere, from education and academia to professional work and everyday life. Despite these benefits, the widespread deployment of LLMs also raises concerns about misinformation, academic dishonesty, and the integrity of digital content. As a result, detecting LLM-generated text has become an active research topic in machine learning over the past few years. However, statistical perspectives on this problem remain largely underexplored. In this talk, I will present several of my recent works on developing statistical methods for detecting LLM-generated text, with the aim of providing principled and efficient tools for identifying such content. The talk will cover methods presented at recent machine learning conferences, including NeurIPS and ICLR. An interactive demonstration of one of the proposed methods is available at https://huggingface.co/spaces/stats-powered-ai/StatDetectLLM [huggingface.co]
Bio: Dr Jin Zhu is an Assistant Professor in the School of Mathematics at the University of Birmingham. His research focuses on large language models, reinforcement learning, and high-dimensional data, particularly on developing computationally efficient methods with statistical guarantees.
