Zero-Shot Detection of LLM-Generated Text via Implicit Reward Model

2026-04-22

AI Summary

This work addresses the growing risk of misuse posed by the increasingly realistic text generated by large language models (LLMs) and proposes IRM, a novel zero-shot detection method that requires no training. IRM leverages implicit reward signals embedded in publicly available instruction-tuned models and their base counterparts to construct a detector without collecting human preference data or performing task-specific fine-tuning. Evaluated on the DetectRL benchmark, IRM significantly outperforms existing zero-shot and supervised detection approaches, demonstrating superior performance in identifying machine-generated text. The method establishes an efficient and generalizable paradigm for LLM-generated text detection, offering a practical solution to a critical challenge in AI safety and content authenticity.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: Fact-Checking / Misinformation Detection (NLP Focus)
• Computer Vision: Large Vision Models

Application Category

• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Search and Retrieval-Augmented AI: Search Tool Learning with LLM: Teaching LLMs to invoke search and make use of retrieved information
• Economics, Online Markets and Human Computation: LLM based quality controls for crowd work

Abstract

Large language models (LLMs) have demonstrated remarkable capabilities across various tasks. However, their ability to generate human-like text has raised concerns about potential misuse. This underscores the need for reliable and effective methods to detect LLM-generated text. In this paper, we propose IRM, a novel zero-shot approach that leverages Implicit Reward Models for LLM-generated text detection. Such implicit reward models can be derived from publicly available instruction-tuned and base models. Previous reward-based method relies on preference construction and task-specific fine-tuning. In comparison, IRM requires neither preference collection nor additional training. We evaluate IRM on the DetectRL benchmark and demonstrate that IRM can achieve superior detection performance, outperforms existing zero-shot and supervised methods in LLM-generated text detection.

Problem

• Zero-Shot Detection
• LLM-Generated Text
• Implicit Reward Model
• Text Detection
• Misuse Prevention

Innovation

• Zero-Shot Detection
• Implicit Reward Model
• LLM-Generated Text
• Reward Modeling
• Instruction-Tuned Models