Truth-value judgment in language models: belief directions are context sensitive

arXiv:2404.18865v1 Announce Type: new
Abstract: Recent work has demonstrated that the latent spaces of large language models (LLMs) contain directions predictive of the truth of sentences. Multiple methods recover such directions and build probes that are described as getting at a model's "knowledge" or "beliefs". We investigate this phenomenon, looking closely at the impact of context on the probes. Our experiments establish where in the LLM the probe's predictions can be described as being conditional on the preceding (related) sentences. …

abstract arxiv belief build context cs.cl judgment knowledge language language models large language large language models llms multiple predictive spaces truth type value work