analiza aiqbench
This research introduces CitePretrainBench, a benchmark designed to evaluate whether large language models can attribute citations directly from their pretraining data without relying on external retrieval systems. By training with active indexing techniques that bind facts to persistent identifiers, the study demonstrates significant improvements in citation precision and robustness against retrieval noise for Qwen-2.5 models. This approach addresses the latency and infrastructure dependencies inherent in current retrieval-augmented generation pipelines while mitigating hallucinated citations.
Cite Pretrain: Retrieval-Free Knowledge Attribution for Large Language Models
arXiv:2506.17585v4 Announce Type: replace Abstract: Large language models can answer questions correctly and still hallucinate their citations. Most citation systems address this by retrieving documents at inference time. Retrieval adds latency and depends on external infrastructure. Retrieval noise can also affect the answer. We study whether LLMs can cite documents from their (continual) pretraining without test-time retrieval. We introduce CitePretrainBench to evaluate this setting. It includes Wikipedia, Common Crawl, and arXiv as real-world corpora and also includes novel, unseen documents. The benchmark includes short-form citation tasks for a single fact and long-form tasks for multiple facts. Continual-pretraining first binds factual knowledge to a persistent identifier for each document. Instruction tuning then teaches the model to put these identifiers in its answers as citations. Passive Indexing appends an identifier at the end of each document. Synthetic data used in Active Indexing restate facts in diverse, compositional forms. We train two mappings together. These mappings are source$\to$fact and fact$\to$source. The first direction involves the model generating content given a cited source. In the second, it learns to attribute its own answers when facts are paraphrased or combined. On Qwen-2.5-7B and 3B, Active Indexing improves citation precision over Passive Indexing on every evaluated citation task, with gains of up to 30.2 percentage points. In the scaling ablation, citation performance continues to improve as we add augmented data, including at 16 times the original token count. Combining internal and external citations also improves robustness to retrieval noise.
Abstract:Large language models can answer questions correctly and still hallucinate their citations. Most citation systems address this by retrieving documents at inference time. Retrieval adds latency and depends on external infrastructure. Retrieval noise can also affect the answer. We study whether LLMs can cite documents from their (continual) pretraining without test-time retrieval. We introduce CitePretrainBench to evaluate this setting. It includes Wikipedia, Common Crawl, and arXiv as real-world corpora and also includes novel, unseen documents. The benchmark includes short-form citation tasks for a single fact and long-form tasks for multiple facts. Continual-pretraining first binds factual knowledge to a persistent identifier for each document. Instruction tuning then teaches the model to put these identifiers in its answers as citations. Passive Indexing appends an identifier at the end of each document. Synthetic data used in Active Indexing restate facts in diverse, compositional forms. We train two mappings together. These mappings are source$\to$fact and fact$\to$source. The first direction involves the model generating content given a cited source. In the second, it learns to attribute its own answers when facts are paraphrased or combined. On Qwen-2.5-7B and 3B, Active Indexing improves citation precision over Passive Indexing on every evaluated citation task, with gains of up to 30.2 percentage points. In the scaling ablation, citation performance continues to improve as we add augmented data, including at 16 times the original token count. Combining internal and external citations also improves robustness to retrieval noise.
Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Machine Learning (cs.LG)
arXiv:2506.17585 [cs.AI]
(or arXiv:2506.17585v4 [cs.AI] for this version)
https://doi.org/10.48550/arXiv.2506.17585 Focus to learn more arXiv-issued DOI via DataCite
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Text integral sub licența deschisă a publicației · 337 cuvinte
Subiecte conexe
[AINews] not much happened today
Latent Space · 10 sept. 2026
[AINews] AI Engineer NYC: Gemini 4 Argon, Vals AI Benchmarks
Latent Space · 14 oct. 2026
[AINews] The Future of Latent Space
Latent Space · 25 sept. 2026
Models for minimalist RAG: B1ade 335M Embedding and 1B Parameter Small Language Models
arXiv · 31 iul. 2026