Towards Tracing Factual Knowledge in Language Models Back to the Training Data
This addresses the challenge of tracing factual knowledge in language models back to their training data, which is important for transparency and verification, but the work is incremental as it builds on existing methods.
The paper tackles the problem of fact tracing in language models, aiming to identify which training examples taught a model to generate a factual assertion, and finds that current training data attribution methods have lower precision than a baseline retrieval method, with room for improvement.
Language models (LMs) have been shown to memorize a great deal of factual knowledge contained in their training data. But when an LM generates an assertion, it is often difficult to determine where it learned this information and whether it is true. In this paper, we propose the problem of fact tracing: identifying which training examples taught an LM to generate a particular factual assertion. Prior work on training data attribution (TDA) may offer effective tools for identifying such examples, known as "proponents". We present the first quantitative benchmark to evaluate this. We compare two popular families of TDA methods -- gradient-based and embedding-based -- and find that much headroom remains. For example, both methods have lower proponent-retrieval precision than an information retrieval baseline (BM25) that does not have access to the LM at all. We identify key challenges that may be necessary for further improvement such as overcoming the problem of gradient saturation, and also show how several nuanced implementation details of existing neural TDA methods can significantly improve overall fact tracing performance.