Emergent World Representations: Exploring a Sequence Model Trained on a Synthetic Task

arXiv:2210.13382v5 Announce Type: replace-cross
Abstract: Language models show a surprising range of capabilities, but the source of their apparent competence is unclear. Do these networks just memorize a collection of surface statistics, or do they rely on internal representations of the process that generates the sequences they see? We investigate this question by applying a variant of the GPT model to the task of predicting legal moves in a simple board game, Othello. Although the network has no a priori …

abstract arxiv capabilities collection cs.ai cs.cl cs.lg language language models networks process replace sequence model show statistics surface synthetic type world