Understanding Fact Recall in Language Models: Why Two-Stage Training Encourages Memorization but Mixed Training Teaches Knowledge

Fact recall, the ability of language models (LMs) to retrieve specific factual knowledge, remains a challenging task despite their impressive general capabilities. Common training strategies often struggle to promote robust recall behavior with two-stage training, which first trains a model with fact-storing examples (e.g., factual statements) and then with fact-recalling examples (question-answer pairs), tending to encourage rote memorization rather than generalizable fact retrieval. In contrast, mixed training, which jointly uses both types of examples, has been empirically shown to improve the ability to recall facts, but the underlying mechanisms are still poorly understood. In this work, we investigate how these training strategies affect how model parameters are shaped during training and how these differences relate to their ability to recall facts. We introduce cross-task gradient trace to identify shared parameters, those strongly influenced by both fact-storing and fact-recalling examples. Our analysis on synthetic fact recall datasets with the Llama-3.2B and Pythia-2.8B models reveals that mixed training encouraging a larger and more centralized set of shared parameters. These findings suggest that the emergence of parameters may play a key role in enabling LMs to generalize factual knowledge across task formulations.

Meta-control of Dialogue Systems Using Large Language Models

Utilizing Large Language Models (LLMs) facilitates the creation of flexible and natural dialogues, a task that has been challenging with traditional rule-based dialogue systems. However, LLMs also have the potential to produce unexpected responses, which may not align with the intentions of dialogue system designers. To address this issue, this paper introduces a meta-control method that employs LLMs to develop more stable and adaptable dialogue systems. The method includes dialogue flow control to ensure that utterances conform to predefined scenarios and turn-taking control to foster natural dialogues. Furthermore, we have implemented a dialogue system that utilizes this meta-control strategy and verified that the dialogue system utilizing meta-control operates as intended.