Extending Epistemic Uncertainty Beyond Parameters Would Assist in Designing Reliable LLMs
T. Duy Nguyen-Hien, Desi R. Ivanova, Yee Whye Teh, Wee Sun Lee · 2025
Read Extending Epistemic Uncertainty Beyond Parameters Would Assist in Designing Reliable LLMs by T. Duy Nguyen-Hien, Desi R. Ivanova, Yee Whye Teh, Wee Sun Lee (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2506-07448.txt.
Description
Nguyen-Hien and co-authors argue that language models need a richer account of their own uncertainty. Today's safety practice mostly rejects answers flagged as unsure. That blunt rule treats all doubt alike, though some doubt can be cleared up and some cannot. The team urges a Bayesian frame that sorts doubt by its source.
With doubt sorted, a model and its user can act in fitting ways. They can ask a clarifying question, fetch outside facts, or sharpen the prompt. Doubt that cannot shrink can still be flagged plainly. The authors say this active stance would beat passive refusal in high stakes uses, and make systems more open about their limits.
Catalog record
| CREATOR | T. Duy Nguyen-Hien, Desi R. Ivanova, Yee Whye Teh, Wee Sun Lee |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 49 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2506-07448