A Statistical Turing Test for Generative Models
By Hayden S. Helm, Carey E. Priebe, Weiwei Yang. 2023
Subfields: Statistics and Probability, Generative Models
How researchers see this paper
The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.
In Statistics and Probability
In Generative Models
Citations in the corpus
- Tracking the perspectives of interacting language models (2024, background): “In-sofar as an individual model is an intriguing proxy for an individual human 1 (Helm et al., 2023), a system of interacting language models is an in-triguing proxy for human communities.”
- Toward a digital twin of U.S. Congress (2025, unknown): “LLMs are now capable of producing human-like content (13) and behaving like humans in controlled experimental settings.”
- Statistical inference on black-box generative models in the data kernel perspective space (2024, unknown): “As can be seen in the top figure of Figure 3, the projection of the models is correlated with the proportion of sensitive data that the model has had access to when the DKPS is induced with sensitive queries.”
- Investigating social alignment via mirroring in a system of interacting language models (2024, unknown): “Although recent LLMs are able to generate human-like outputs (Webb et al., 2023; Helm et al., 2023) and replicate classic experiments without explicit training (see, e.g., (Webb et al., 2023)), human-like behavior is not sufficient to model individuals in a social system.”
- Testing for LLM response differences: the case of a composite null consisting of semantically irrelevant query perturbations (2025, unknown): “…et al., 2023), radiology (D’Antonoli et al., 2024; Kim et al., 2024), law (Sun, 2023; Siino et al., 2025), etc. (Lo, 2023; Rahman et al., 2023; Helm et al., 2023; Khan and Umer, 2024; Zhang et al., 2024), has spurred investigations into failure modes of the models and the systems in which…”