Value Over Language Model: Detecting Original Contribution in Writing

Summary

Reframes AI-text detection away from surface style toward informational contribution: VOLM measures how much value a human added over a replacement-level document an LLM could produce from the task description alone, by reconstructing the document from progressively coarser content representations and comparing against task-description-only reconstructions. It needs no training or labels and never scores surface text, so it resists humanization and paraphrase. One of the three evaluation domains is ICLR peer reviews, making it a methodologically distinct approach to the detection-of-AI-reviews problem.

Contribution

Queued — not yet read.

Method

Queued — not yet read.

Relevance to RISE

Surfaced by the 2026-09-07 literature sweep as a medium-priority candidate for the RISE knowledge base (angle: AI in peer review). Verified against: Opened https://arxiv.org/abs/2609.00700 on 2026-09-07; verbatim abstract read (VOLM framework, three domains incl. ICLR peer reviews, invariance to round-trip translation), three authors, submitted 1 Sep 2026, 39 pages. Not in dedupe_titles.txt.

Critique / open questions

Queued — not yet assessed.

Key quotes

LLMs have been rapidly adopted across writing tasks, prompting the development of tools for detecting LLM-generated text. (abstract)

Yet, these tools largely measure how much of a document's surface text was written by an LLM and aren't fundamentally designed to measure how much of the information content or ideas originated from the LLM itself rather than being supplied by the user in the prompt. (abstract)