Representations of Fact, Fiction and Forecast in Large Language Models: Epistemics and Attitudes
This addresses the challenge of unreliable uncertainty expression in LLMs for applications requiring rational communication, but it is incremental as it builds on existing typological frameworks.
The paper tackled the problem of large language models (LLMs) generating unreliable expressions of uncertainty by evaluating their knowledge of epistemic modality using controlled stories, finding that their performance is limited and not robust. The result indicates that LLMs' expressions of uncertainty are not always reliable, necessitating enriched semantic knowledge for building uncertainty-aware models.
Rational speakers are supposed to know what they know and what they do not know, and to generate expressions matching the strength of evidence. In contrast, it is still a challenge for current large language models to generate corresponding utterances based on the assessment of facts and confidence in an uncertain real-world environment. While it has recently become popular to estimate and calibrate confidence of LLMs with verbalized uncertainty, what is lacking is a careful examination of the linguistic knowledge of uncertainty encoded in the latent space of LLMs. In this paper, we draw on typological frameworks of epistemic expressions to evaluate LLMs' knowledge of epistemic modality, using controlled stories. Our experiments show that the performance of LLMs in generating epistemic expressions is limited and not robust, and hence the expressions of uncertainty generated by LLMs are not always reliable. To build uncertainty-aware LLMs, it is necessary to enrich semantic knowledge of epistemic modality in LLMs.