Factuality Beyond Reference in LLMs

Thierry Poibeau

Keywords: Factuality, Reference, Grounding, Inferentialism

TL;DR: LLMs may refer successfully to the world while still failing to produce genuinely accountable assertions, because factuality depends on inferential and normative commitment structures rather than reference alone.

Abstract: Large language models are often discussed in terms of grounding and reference. This paper argues that such debates leave largely unresolved the question of factuality: LLMs may refer successfully while still producing false or unstable propositions. Drawing on inferentialist approaches associated with Brandom, the paper shifts attention from reference to assertion, commitment, and epistemic accountability.

Email Sharing: We authorize the sharing of all author emails with Program Chairs.

Data Release: We authorize the release of our submission and author names to the public in the event of acceptance.

Submission Number: 38