User Behavior Prediction as a Generic, Robust, Scalable, and Low-Cost Evaluation Strategy for Estimating Generalization in LLMs
This addresses the problem of evaluating LLM generalization for researchers and practitioners, offering a scalable and robust method, though it is incremental as it adapts existing personalization concepts to a new evaluation context.
The paper tackles the challenge of measuring generalization in LLMs due to data contamination by proposing user behavior prediction as an alternative evaluation strategy, showing that GPT-4o outperforms GPT-4o-mini and Llama-3.1-8B-Instruct on movie and music recommendation datasets, with all models having significant room for improvement.
Measuring the generalization ability of Large Language Models (LLMs) is challenging due to data contamination. As models grow and computation becomes cheaper, ensuring tasks and test cases are unseen during training phases will become nearly impossible. We argue that knowledge-retrieval and reasoning tasks are not ideal for measuring generalization, as LLMs are not trained for specific tasks. Instead, we propose user behavior prediction, also a key aspect of personalization, as a theoretically sound, scalable, and robust alternative. We introduce a novel framework for this approach and test it on movie and music recommendation datasets for GPT-4o, GPT-4o-mini, and Llama-3.1-8B-Instruct. Results align with our framework's predictions, showing GPT-4o outperforms GPT-4o-mini and Llama, though all models have much room for improvement, especially Llama.