Can Performant LLMs Be Ethical? Quantifying the Impact of Web Crawling Opt-Outs

2025-04-08

AI Summary

The “data compliance gap” (DCG)—the performance degradation incurred when training large language models (LLMs) exclusively on copyright-compliant data—lacks formal definition and empirical measurement. Method: We formally define DCG and empirically quantify its impact by training a 1.5B-parameter LLM from scratch and via continued pretraining, comparing fully compliant data (e.g., filtered of major publishers’ copyrighted content) against non-compliant data. Results: On general knowledge benchmarks (MMLU, RobustBench), DCG ≈ 0%, indicating copyright-compliant data suffices for foundational capabilities. In contrast, biomedical reasoning (PubMedQA) exhibits significant performance decline, confirming domain-specific tasks remain dependent on high-quality copyrighted resources. Crucially, targeted reintroduction of curated copyrighted data substantially mitigates this gap. This work establishes the first quantitative framework for assessing data compliance in LLM pretraining, providing empirically grounded guidance for responsible data governance and model development.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: (Large) Language Models
• Computer Vision: Large Vision Models

Application Category

• Web Mining and Content Analysis: Large pretrained models with web data
• Economics, Online Markets and Human Computation: Humans versus LLMs for data annotation and labeling
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions

Abstract

The increasing adoption of web crawling opt-outs by copyright holders of online content raises critical questions about the impact of data compliance on large language model (LLM) performance. However, little is known about how these restrictions (and the resultant filtering of pretraining datasets) affect the capabilities of models trained using these corpora. In this work, we conceptualize this effect as the $ extit{data compliance gap}$ (DCG), which quantifies the performance difference between models trained on datasets that comply with web crawling opt-outs, and those that do not. We measure the data compliance gap in two settings: pretraining models from scratch and continual pretraining from existing compliant models (simulating a setting where copyrighted data could be integrated later in pretraining). Our experiments with 1.5B models show that, as of January 2025, compliance with web data opt-outs does not degrade general knowledge acquisition (close to 0% DCG). However, in specialized domains such as biomedical research, excluding major publishers leads to performance declines. These findings suggest that while general-purpose LLMs can be trained to perform equally well using fully open data, performance in specialized domains may benefit from access to high-quality copyrighted sources later in training. Our study provides empirical insights into the long-debated trade-off between data compliance and downstream model performance, informing future discussions on AI training practices and policy decisions.

Problem

• Impact of web crawling opt-outs on LLM performance
• Data compliance gap between filtered and unfiltered datasets
• Effect of copyrighted data exclusion on specialized domains

Innovation

• Quantifies performance gap via data compliance
• Tests pretraining and continual pretraining impacts
• Evaluates domain-specific effects of data exclusion