Researchers Lezhi Tan and Tijana Zrnic propose a new approach to using synthetic data in scientific research. They argue that synthetic data can be biased, noisy, and misspecified, but suggest that their method can provide provable validity guarantees. The key insight is 'task exchangeability', which allows researchers to identify historical tasks with available real data that are equivalent to the current task of interest. This enables valid inference under certain conditions. The authors demonstrate their framework using public opinion surveys and AI evaluation. --- Why it matters: This work matters because it addresses a fundamental concern about synthetic data: its potential for bias, noise, and misspecification. By providing provable validity guarantees, Tan and Zrnic's method can help researchers trust the results of studies that use synthetic data, which is increasingly common in fields like AI evaluation and proteomics. Source: https://arxiv.org/abs/2606.13629