Research map / Large language models / Language, safety and interpretability
Common Corpus: The Largest Collection of Ethical Data for LLM Pre-Training
Pierre-Carl Langlais, Pavel Chizhov, Catherine Arnett, Carlos Rosas Hinostroza, Mattia Nee, Eliot Krzysztof Jones, Irène Girard, David Mach, Anastasia Stasenko, Ivan P. Yamshchikov
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| pre-train on Common Corpus (open data) instead of closed-data corpora | 350M model · MultiBLiMP | accuracy: 0.711 -> 0.774 | Improved |
| pre-train on Common Corpus instead of OLMo's open data | 1.2B model · MultiBLiMP | accuracy: 0.699 -> 0.797 | Improved |
| pre-train on Common Corpus instead of Gemma 3 data | 1.2B model · XStoryCloze | accuracy: 0.594 -> 0.526 | Hurt |
| pre-train on Common Corpus instead of XGLM data | 350M model · XCOPA | accuracy: 0.550 -> 0.533 | Hurt |
| pre-train on Common Corpus instead of Gemma 3 data | 1.2B model · MultiBLiMP | accuracy: 0.799 -> 0.797 | No clear effect |
| use custom regex patterns on top of Presidio for phone numbers (ablation) | PII removal · phone number detection | detection rate: 55-60% -> 85% | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.
- BLOOM 560M — BLOOM: A 176B-Parameter Open-Access Multilingual Language Model (JMLR 2024) · better on 1 of 3 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- The Common Pile v0.1: An 8TB Dataset of Public Domain and Openly Licensed Text (NeurIPS 2025 · similarity 0.93)
- It's All Training: A Fully Synthetic Single-Stage Recipe for LLMs (NeurIPS 2026 · similarity 0.91)
- Fixing It in Post: A Comparative Study of LLM Post-Training Data Quality and Model Performance (NeurIPS 2025 · similarity 0.90)
- Dataset Decomposition: Faster LLM Training with Variable Sequence Length Curriculum (NeurIPS 2024 · similarity 0.89)
- ResearchCodeBench: Benchmarking LLMs on Implementing Novel Machine Learning Research Code (NeurIPS 2025 · similarity 0.89)
- DataMan: Data Manager for Pre-training Large Language Models (ICLR 2025 · similarity 0.89)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
