Language Models May Verbatim Complete Text They Were Not Explicitly Trained On

• URL: http://arxiv.org/abs/2503.17514v2
• Date: Tue, 25 Mar 2025 04:43:33 GMT
• Title: Language Models May Verbatim Complete Text They Were Not Explicitly Trained On
• Authors: Ken Ziyu Liu, Christopher A. Choquette-Choo, Matthew Jagielski, Peter Kairouz, Sanmi Koyejo, Percy Liang, Nicolas Papernot,
• Abstract summary: We show that a $n$-gram based membership definition can be effectively gamed.<n>We show that it is difficult to find a single viable choice of $n$ for membership definitions.<n>Our findings highlight the inadequacy of $n$-gram membership, suggesting membership definitions fail to account for auxiliary information.
• Score: 97.3414396208613
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: An important question today is whether a given text was used to train a large language model (LLM). A \emph{completion} test is often employed: check if the LLM completes a sufficiently complex text. This, however, requires a ground-truth definition of membership; most commonly, it is defined as a member based on the $n$-gram overlap between the target text and any text in the dataset. In this work, we demonstrate that this $n$-gram based membership definition can be effectively gamed. We study scenarios where sequences are \emph{non-members} for a given $n$ and we find that completion tests still succeed. We find many natural cases of this phenomenon by retraining LLMs from scratch after removing all training samples that were completed; these cases include exact duplicates, near-duplicates, and even short overlaps. They showcase that it is difficult to find a single viable choice of $n$ for membership definitions. Using these insights, we design adversarial datasets that can cause a given target sequence to be completed without containing it, for any reasonable choice of $n$. Our findings highlight the inadequacy of $n$-gram membership, suggesting membership definitions fail to account for auxiliary information available to the training algorithm.

Related papers

• $\texttt{SEM-CTRL}$: Semantically Controlled Decoding [53.86639808659575]
  $texttSEM-CTRL$ is a unified approach that enforces rich context-sensitive constraints and task- and instance-specific semantics directly on an LLM decoder.<n>texttSEM-CTRL$ allows small pre-trained LLMs to efficiently outperform larger variants and state-of-the-art reasoning models.
  arXiv  Detail & Related papers  (2025-03-03T18:33:46Z)
• Emergent Response Planning in LLM [6.969916576925677]
  We show that large language models (LLMs) plan ahead for the future in their hidden representations.<n>The findings suggest potential applications for improving transparency and generation control.
  arXiv  Detail & Related papers  (2025-02-10T08:48:10Z)
• Zero-Shot Statistical Tests for LLM-Generated Text Detection using Finite Sample Concentration Inequalities [13.657259851747126]
  We show that type I and type II errors for our tests decrease exponentially in the text length.<n>We show that if the string is generated by $A$, the log-perplexity of the string under $A$ converges to the average entropy of the string under $A$, except with an exponentially small probability in string length.
  arXiv  Detail & Related papers  (2025-01-04T23:51:43Z)
• Reasoning to Attend: Try to Understand How <SEG> Token Works [44.33848900059659]
  We show that the $texttSEG>$ token contributes to semantic similarity within image-text pairs. We present READ, which facilitates LMMs' resilient $textbfREA$soning capability of where to atten$textbfD$ under the guidance of highly activated points.
  arXiv  Detail & Related papers  (2024-12-23T17:44:05Z)
• FLARE: Faithful Logic-Aided Reasoning and Exploration [50.9814063216852]
  We introduce a novel approach for traversing the problem space using task decompositions.<n>We use the Large Language Models to plan a solution, soft-formalise the query into facts and predicates using a logic programming code.<n>Our method allows us to compute the faithfulness of the reasoning process w.r.t. the generated code and analyse the steps of the multi-hop search without relying on external solvers.
  arXiv  Detail & Related papers  (2024-10-14T19:39:11Z)
• Evaluating $n$-Gram Novelty of Language Models Using Rusty-DAWG [57.14250086701313]
  We investigate the extent to which modern LMs generate $n$-grams from their training data. We develop Rusty-DAWG, a novel search tool inspired by indexing of genomic data.
  arXiv  Detail & Related papers  (2024-06-18T21:31:19Z)
• Protecting Copyrighted Material with Unique Identifiers in Large Language Model Training [55.321010757641524]
  A major public concern regarding the training of large language models (LLMs) is whether they abusing copyrighted online text. Previous membership inference methods may be misled by similar examples in vast amounts of training data. We propose an alternative textitinsert-and-detection methodology, advocating that web users and content platforms employ textbftextitunique identifiers.
  arXiv  Detail & Related papers  (2024-03-23T06:36:32Z)
• Blessing of Class Diversity in Pre-training [54.335530406959435]
  We prove that when the classes of the pre-training task are sufficiently diverse, pre-training can significantly improve the sample efficiency of downstream tasks. Our proof relies on a vector-form Rademacher complexity chain rule for composite function classes and a modified self-concordance condition.
  arXiv  Detail & Related papers  (2022-09-07T20:10:12Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.