What Can Large Language Models Do for Theorem Proving and Formal Methods?

Paper i proceeding, 2024

With the introduction of large language models, AI for natural language have taken a leap. These systems are now also being used for tasks that has previously been dominated by symbolic methods, such as program synthesis and even to support formalising mathematics and assist theorem provers. We survey some recent applications in theorem proving, focusing on how they combine neural networks with symbolic systems, and report on a case-study of using GPT-4 for the task of automated conjecturing a.k.a. theory exploration.