Large language models are trained on vast amounts of internet data, prompting concerns and speculation that they have memorized public benchmarks. Going from speculation to proof of contamination is challenging, as the pretraining data used by proprietary models are often not publicly accessible. We show that it is possible to provide provable guarantees of test set contamination in language models without access to pretraining data or model weights. Our approach leverages the fact that when there is no data contamination, all orderings of an exchangeable benchmark should be equally likely. In contrast, the tendency for language models to memorize example order means that a contaminated language model will find certain canonical orderings to be much more likely than others. Our test flags potential contamination whenever the likelihood of a canonically ordered benchmark dataset is significantly higher than the likelihood after shuffling the examples. We demonstrate that our procedure is sensitive enough to reliably prove test set contamination in challenging situations, including models as small as 1.4 billion parameters, on small test sets of only 1000 examples, and datasets that appear only a few times in the pretraining corpus. Using our test, we audit five popular publicly accessible language models for test set contamination and find little evidence for pervasive contamination.
Language models (LMs) are increasingly used as simulacra for people, yet their ability to match the distribution of views of a specific demographic group and be \textit{distributionally aligned} remains uncertain. This notion of distributional alignment is complex, as there is significant variation in the types of attributes that are simulated. Prior works have underexplored the role of three critical variables -- the question domain, steering method, and distribution expression method -- which motivates our contribution of a benchmark explicitly addressing these dimensions. We construct a dataset expanding beyond political values, create human baselines for this task, and evaluate the extent to which an LM can align with a particular group's opinion distribution to inform design choices of such simulation systems. Our analysis reveals open problems regarding if, and how, LMs can be used to simulate humans, and that LLMs can more accurately describe the opinion distribution than simulate such distributions.
As businesses, products, and services spring up around large language models, the trustworthiness of these models hinges on the verifiability of their outputs. However, methods for explaining language model outputs largely fall across two distinct fields of study which both use the term "attribution" to refer to entirely separate techniques: citation generation and training data attribution. In many modern applications, such as legal document generation and medical question answering, both types of attributions are important. In this work, we argue for and present a unified framework of large language model attributions. We show how existing methods of different types of attribution fall under the unified framework. We also use the framework to discuss real-world use cases where one or both types of attributions are required. We believe that this unified framework will guide the use case driven development of systems that leverage both types of attribution, as well as the standardization of their evaluation.
Singh et al. (2020) point out the dangers of contextual bias in visual recognition datasets. They propose two methods, CAM-based and feature-split, that better recognize an object or attribute in the absence of its typical context while maintaining competitive within-context accuracy. To verify their performance, we attempted to reproduce all 12 tables in the original paper, including those in the appendix. We also conducted additional experiments to better understand the proposed methods, including increasing the regularization in CAM-based and removing the weighted loss in feature-split. As the original code was not made available, we implemented the entire pipeline from scratch in PyTorch 1.7.0. Our implementation is based on the paper and email exchanges with the authors. We found that both proposed methods in the original paper help mitigate contextual bias, although for some methods, we could not completely replicate the quantitative results in the paper even after completing an extensive hyperparameter search. For example, on COCO-Stuff, DeepFashion, and UnRel, our feature-split model achieved an increase in accuracy on out-of-context images over the standard baseline, whereas on AwA, we saw a drop in performance. For the proposed CAM-based method, we were able to reproduce the original paper's results to within 0.5$\%$ mAP. Our implementation can be found at this https URL
As businesses, products, and services spring up around large language models, the trustworthiness of these models hinges on the verifiability of their outputs. However, methods for explaining language model outputs largely fall across two distinct fields of study which both use the term "attribution" to refer to entirely separate techniques: citation generation and training data attribution. In many modern applications, such as legal document generation and medical question answering, both types of attributions are important. In this work, we argue for and present a unified framework of large language model attributions. We show how existing methods of different types of attribution fall under the unified framework. We also use the framework to discuss real-world use cases where one or both types of attributions are required. We believe that this unified framework will guide the use case driven development of systems that leverage both types of attribution, as well as the standardization of their evaluation.
We introduce HARMONI, a three-dimensional (3D) computer vision and audio processing method for analyzing caregiver-child behavior and interaction from observational videos. HARMONI operates at subsecond resolution, estimating 3D mesh representations and spatial interactions of humans, and adapts to challenging natural environments using an environment-targeted synthetic data generation module. Deployed on 500 hours from the SEEDLingS dataset, HARMONI generates detailed quantitative measurements of 3D human behavior previously unattainable through manual efforts or 2D methods. HARMONI identifies longitudinal trends in child-caregiver interaction, including child movement, body pose, dyadic touch, visibility, and conversational turns. The integrated visual and audio analysis further reveals multimodal trends, including associations between child conversational turns and movement. Open-sourced for large-scale analysis, HARMONI facilitates advancements in human development research. HARMONI achieves 63 to 80% consistency on key attributes with human annotators on SEEDLingS and 84 to 93% consistency on videos taken from a laboratory setting while achieving >100 times savings in time.
Singh et al. (2020) point out the dangers of contextual bias in visual recognition datasets. They propose two methods, CAM-based and feature-split, that better recognize an object or attribute in the absence of its typical context while maintaining competitive within-context accuracy. To verify their performance, we attempted to reproduce all 12 tables in the original paper, including those in the appendix. We also conducted additional experiments to better understand the proposed methods, including increasing the regularization in CAM-based and removing the weighted loss in feature-split. As the original code was not made available, we implemented the entire pipeline from scratch in PyTorch 1.7.0. Our implementation is based on the paper and email exchanges with the authors. We found that both proposed methods in the original paper help mitigate contextual bias, although for some methods, we could not completely replicate the quantitative results in the paper even after completing an extensive hyperparameter search. For example, on COCO-Stuff, DeepFashion, and UnRel, our feature-split model achieved an increase in accuracy on out-of-context images over the standard baseline, whereas on AwA, we saw a drop in performance. For the proposed CAM-based method, we were able to reproduce the original paper's results to within 0.5$\%$ mAP. Our implementation can be found at https://github.com/princetonvisualai/ContextualBias.
