Towards Logically Consistent Language Models via Probabilistic Reasoning

arXiv:2404.12843v1 Announce Type: new
Abstract: Large language models (LLMs) are a promising venue for natural language understanding and generation tasks. However, current LLMs are far from reliable: they are prone to generate non-factual information and, more crucially, to contradict themselves when prompted to reason about beliefs of the world. These problems are currently addressed with large scale fine-tuning or by delegating consistent reasoning to external tools. In this work, we strive for a middle ground and introduce a training objective …

abstract arxiv consistent cs.cl cs.lg current generate however information language language models language understanding large language large language models llms natural natural language reason reasoning tasks type understanding via world