Abstract:
A common view is that a normal human assertion is successful only if (i) the speaker utters words which express what they intend to express, and (ii) the hearer comes to believe what is expressed. We examine how this picture looks on the hypothesis that LLMs can make assertions. In particular, we ask whether LLM word-choice can be understood to be intentional, and if not, what this might suggest about human word-choice in normal cases.
