Cathy Jiao

Synthetic data now is essential to LLM training, used to strengthen advanced capabilities such as autonomous and long-horizon task execution. Yet recent work shows that training on it at scale can degrade model generation, making it important to decide what synthetic data is worth training on. While current data curation practices do so with individual-level signals (i.e., estimates of each data sample’s training utility in isolation), across pre-training and post-training settings we show that this is insufficient for synthetic data, and that group-level signals (i.e., estimates of utility that account for interactions among data samples) are necessary for effective data curation. First, we show that individual-level signals are blind to how samples jointly affect training: synthetic datasets with different compositions can be indistinguishable under individual-level influence yet differ sharply under group-level influence, and curating by the latter yields better downstream performance, particularly in generative capability. Second, we find that group-level signals matter more as training pipelines become increasingly synthetic: among widely used data curation methods, only those incorporating them improve over baseline, with gains increasing when weights capturing relations among samples are amplified. Finally, we translate these findings into practice – for model developers under a compute budget, we offer a cheap diagnostic that prioritizes which groups of synthetic data most need group-level estimation, recovering much of the benefit of full group-level scoring at a fraction of the compute cost.

Data attribution methods quantify the influence of training data on model outputs and are becoming increasingly relevant for a wide range of LLM research and applications, including dataset curation, model interpretability, data valuation. However, there remain critical gaps in systematic LLM-centric evaluation of data attribution methods. To this end, we introduce DATE-LM (Data Attribution Evaluation in Language Models), a unified benchmark for evaluating data attribution methods through real-world LLM applications. DATE-LM measures attribution quality through three key tasks – training data selection, toxicity/bias filtering, and factual attribution. Our benchmark is designed for ease of use, enabling researchers to configure and run large-scale evaluations across diverse tasks and LLM architectures. Furthermore, we use DATE-LM to conduct a large-scale evaluation of existing data attribution methods. Our findings show that no single method dominates across all tasks, data attribution methods have trade-offs with simpler baselines, and method performance is sensitive to task-specific evaluation design. Finally, we release a public leaderboard for quick comparison of methods and to facilitate community engagement. We hope DATE-LM serves as a foundation for future data attribution research in LLMs.

Training data is a pivotal resource for building large language models (LLMs), but unfair pricing in data markets poses a serious challenge for both data buyers (e.g., LLM builders) and sellers (e.g., human annotators), which discourages market participation, reducing data quantity and quality. In this paper, we propose a fairshare pricing framework that sets training data prices using data valuation methods to quantify their contribution to LLMs. In our framework, buyers make purchasing decisions using data valuation and sellers set prices to maximize their profits based on the anticipated buyer purchases. We theoretically show that pricing derived from our framework is tightly linked to data valuation and buyers’ budget, optimal for both buyers and sellers. Through market simulations using current LLMs and datasets (math problems, medical diagnosis, and physical reasoning), we show that our framework is fairshare for buyers by ensuring their purchased data is reflective of model training value, leading to higher LLM task performances per-dollar spent on data, and fairshare for sellers by ensuring they sell their data at optimal prices. Our framework lays the foundation for future research on equitable and sustainable data markets for large-scale AI.

Data attribution methods are used to measure the contribution of training data towards model outputs, and have several important applications in areas such as dataset curation and model interpretability. However, many standard data attribution methods, such as influence functions, utilize model gradients and are computationally expensive. In our paper, we show in-context probing (ICP) – prompting a LLM – can serve as a fast proxy for gradient-based data attribution for data selection under conditions contingent on data similarity. We study this connection empirically on standard NLP tasks, and show that ICP and gradient-based data attribution are well-correlated in identifying influential training data for tasks that share similar task type and content as the training data. Additionally, fine-tuning models on influential data selected by both methods achieves comparable downstream performance, further emphasizing their similarities. We also examine the connection between ICP and gradient-based data attribution using synthetic data on linear regression tasks. Our synthetic data experiments show similar results with those from NLP tasks, suggesting that this connection can be isolated in simpler settings, which offers a pathway to bridging their differences.