arXiv subjects

Jeffrey G. Wang

Publications and source records attributed to Jeffrey G. Wang.

4 recordsLinked to original sources

arXiv2026

CheckMIABench: Firm Foundations For Membership Inference Attacks on Language Models

Membership inference attacks (MIAs) are a canonical way to assess a machine learning model's privacy properties. Although several attempts have been made to evaluate MIAs on language models, the extant literature has suffered numerous difficulties in constructing clean evaluations to test new techniques. In particular, subtle distribution shifts between member and non-member sets can undermine the statistical validity of MIAs; recent work has underscored this by showing that "blind" methods with no access to the underlying model can perform far better than published methods on the same benchmarks. This paper constructs a benchmark for principled evaluation of MIAs against LLMs, by leveraging the insight that training data before and after a fixed point during training are drawn from the same distribution. Therefore, all open-source models with intermediate checkpoints and public training data can be converted into MIA testbeds. We apply our framework to a half-dozen published attacks on the Pythia and OLMo family of models, from 70M to 7B parameters. To facilitate further privacy research, we open-source a modular library for designing and implementing attacks in this setting: https://github.com/safr-ai-lab/pandora_llm.

cs.LG↗

arXiv2025

Learning from Synthetic Labs: Language Models as Auction Participants

This paper investigates the behavior of simulated AI agents (large language models, or LLMs) in auctions, introducing a novel synthetic data-generating process to help facilitate the study and design of auctions. We find that LLMs -- when endowed with chain of thought reasoning capacity -- agree with the experimental literature in auctions across a variety of classic auction formats. In particular, we find that LLM bidders produce results consistent with risk-averse human bidders; that they perform closer to theoretical predictions in obviously strategy-proof auctions; and, that they succumb to the winner's curse in common value settings. On prompting, we find that LLMs are not very sensitive to naive changes in prompts (e.g., language, currency) but can improve dramatically towards theoretical predictions with the right mental model (i.e., the language of Nash deviations). We run 1,000$+$ auctions for less than $\$$400 with GPT-4 models (three orders of magnitude cheaper than modern auction experiments) and develop a framework flexible enough to run auction experiments with any LLM model and a wide range of auction design specifications, facilitating further experimental study by decreasing costs and serving as a proof-of-concept for the use of LLM proxies.

cs.GT↗

arXiv2024

Bias Begets Bias: The Impact of Biased Embeddings on Diffusion Models

With the growing adoption of Text-to-Image (TTI) systems, the social biases of these models have come under increased scrutiny. Herein we conduct a systematic investigation of one such source of bias for diffusion models: embedding spaces. First, because traditional classifier-based fairness definitions require true labels not present in generative modeling, we propose statistical group fairness criteria based on a model's internal representation of the world. Using these definitions, we demonstrate theoretically and empirically that an unbiased text embedding space for input prompts is a necessary condition for representationally balanced diffusion models, meaning the distribution of generated images satisfy diversity requirements with respect to protected attributes. Next, we investigate the impact of biased embeddings on evaluating the alignment between generated images and prompts, a process which is commonly used to assess diffusion models. We find that biased multimodal embeddings like CLIP can result in lower alignment scores for representationally balanced TTI models, thus rewarding unfair behavior. Finally, we develop a theoretical framework through which biases in alignment evaluation can be studied and propose bias mitigation methods. By specifically adapting the perspective of embedding spaces, we establish new fairness conditions for diffusion model development and evaluation.

cs.LG↗

arXiv2024

Pandora's White-Box: Precise Training Data Detection and Extraction in Large Language Models

In this paper we develop state-of-the-art privacy attacks against Large Language Models (LLMs), where an adversary with some access to the model tries to learn something about the underlying training data. Our headline results are new membership inference attacks (MIAs) against pretrained LLMs that perform hundreds of times better than baseline attacks, and a pipeline showing that over 50% (!) of the fine-tuning dataset can be extracted from a fine-tuned LLM in natural settings. We consider varying degrees of access to the underlying model, pretraining and fine-tuning data, and both MIAs and training data extraction. For pretraining data, we propose two new MIAs: a supervised neural network classifier that predicts training data membership on the basis of (dimensionality-reduced) model gradients, as well as a variant of this attack that only requires logit access to the model by leveraging recent model-stealing work on LLMs. To our knowledge this is the first MIA that explicitly incorporates model-stealing information. Both attacks outperform existing black-box baselines, and our supervised attack closes the gap between MIA attack success against LLMs and the strongest known attacks for other machine learning models. In fine-tuning, we find that a simple attack based on the ratio of the loss between the base and fine-tuned models is able to achieve near-perfect MIA performance; we then leverage our MIA to extract a large fraction of the fine-tuning dataset from fine-tuned Pythia and Llama models. Our code is available at github.com/safr-ai-lab/pandora-llm.

cs.CR↗