Adnan KhanAdnanProduct & AI
San Francisco Bay Area
Get in touch

Source note2 min read

Detecting LLM-Generated Tokens in Human--LLM Coauthored Text

Source

arXiv

arxiv.org/abs/2607.21458v1

Read the source

Source

Plain-Language Brief

The paper presents a new method for identifying parts of a text generated by Language Model (LLM) at the token level, improving on existing document-level methods. The method is based on smoothing adjacent token scores and using an adaptive Lepski-type rule to select the bandwidth according to local authorship structure.

What the Document Says

  • The paper discusses a problem: detecting LLM-generated content in mixed-authorship documents.
  • Existing methods focus on document-level classification but cannot identify parts of text generated by LLMs.
  • The proposed method operates at the token level and builds on existing token-level detection scores.
  • The key idea is to smooth adjacent token scores and use an adaptive Lepski-type rule for bandwidth selection based on local authorship structure.
  • The method is simple to implement and does not require token-level labeled data for training.
  • Theoretical analysis shows the proposed method achieves good mean square error performance in estimating underlying signal.
  • Empirical results demonstrate strong performance against various baselines in synthetic datasets and a realistic dataset.

Key Concepts Mentioned

  • human-AI collaborative writing
  • LLM-generated content
  • token-level detection
  • Lepski-type rule
  • mean square error

Claims Made by the Document

  • The proposed method addresses the need for fine-grained LLM-generated content detection.
  • The method performs favorably in estimating underlying signal theoretically and empirically.

Evidence Used

  • The paper introduces a new method for detecting LLM-generated content
  • The theoretical analysis shows mean square error performance
  • Empirical results demonstrate strong performance

Methods / Approach

UNKNOWN

Limitations Stated by the Authors

UNKNOWN

Limitations Noted by the Agent

Agent interpretation — not from the source.

UNKNOWN

Open Questions

UNKNOWN

Notes for Future Q&A

May include interpretation. Labeled where it does.

UNKNOWN

Raw Abstract / Excerpt

The rise of human-AI collaborative writing has created a growing need for fine-grained detection methods that support localizing likely LLM-generated content in mixed-authorship documents. Existing methods for detecting LLM-generated text mainly focus on document-level classification and cannot identify which parts of the text are generated by LLMs. This paper introduces a new method to address this urgent need. Our method operates at the token level, the natural unit of modern language models, and builds on existing token-level detection scores. The key idea is to smooth adjacent token scores to reduce their variability, while using an adaptive Lepski-type rule to select the bandwidth according to the local authorship structure. Our method is simple to implement and does not require token-level labeled data for training. Theoretically, we characterize this trade-off and show that the proposed method achieves favorable mean square error performance in estimating the underlying signal. Empirically, we demonstrate strong performance of our method against a wide range of baselines in both synthetic datasets and a realistic dataset. We deploy a publicly accessible website that implements the methods as well.