Forecasting Credit Ratings: A Case Study where Traditional Methods Outperform Generative LLMs

F Drinkall, J Pierrehumbert… - … of the Joint Workshop of the …, 2025 - aclanthology.org
Abstract Large Language Models (LLMs) have been shown to perform well for many
downstream tasks. Transfer learning can enable LLMs to acquire skills that were not …

Measuring memorization through probabilistic discoverable extraction

J Hayes, M Swanberg, H Chaudhari, I Yona… - arxiv preprint arxiv …, 2024 - arxiv.org
Large language models (LLMs) are susceptible to memorizing training data, raising
concerns due to the potential extraction of sensitive information. Current methods to …

Paying Attention to Facts: Quantifying the Knowledge Capacity of Attention Layers

LZ Wong - arxiv preprint arxiv:2502.05076, 2025 - arxiv.org
In this paper, we investigate the ability of single-layer attention-only transformers (ie
attention layers) to memorize facts contained in databases from a linear-algebraic …

Sequence-Level Analysis of Leakage Risk of Training Data in Large Language Models

T Tiwari, GE Suh - arxiv preprint arxiv:2412.11302, 2024 - arxiv.org
This work advocates for the use of sequence level probabilities for quantifying the risk of
extraction training data from Large Language Models (LLMs) as they provide much finer …