SELF-[IN]CORRECT: LLMs Struggle with Discriminating Self-Generated Responses
Proceedings of the AAAI Conference on Artificial Intelligence · 11.04.2025 · 39 (23) · 24266-24275
DOI: 10.1609/aaai.v39i23.34603 ↗Kurz gefasst
Fachartikel in der Zeitschrift Proceedings of the AAAI Conference on Artificial Intelligence, online erschienen am 11.04.2025. Verfasst von Dongwei Jiang und 5 weiteren Autor:innen. Die Arbeit wurde bisher 6-mal zitiert (Crossref).
Abstract (Originalsprache)
Can LLMs consistently improve their previous outputs for better results? For this to be true, LLMs would need to be better at discriminating among previously-generated alternatives, than generating initial responses. We explore the validity of this hypothesis in practice. We first formulate a unified framework that allows us to compare the generative and discriminative capability of any model on any task. In our resulting experimental analysis of several open-source and industrial LLMs, we observe that model’s are not reliably better at discriminating among previously-generated alternatives than generating initial responses. This finding challenges the notion that LLMs may be able to enhance their performance only through their own judgment.
Zitieren
APA
Jiang, D., Zhang, J., Weller, O., Weir, N., Durme, B. V., & Khashabi, D. (2025). SELF-[IN]CORRECT: LLMs Struggle with Discriminating Self-Generated Responses. Proceedings of the AAAI Conference on Artificial Intelligence, 39(23), 24266-24275. https://doi.org/10.1609/aaai.v39i23.34603
Vancouver
Jiang D, Zhang J, Weller O, Weir N, Durme BV, Khashabi D. SELF-[IN]CORRECT: LLMs Struggle with Discriminating Self-Generated Responses. Proceedings of the AAAI Conference on Artificial Intelligence. 2025;39(23):24266-24275. doi:10.1609/aaai.v39i23.34603
BibTeX
@article{Jiang2025,
author = {Jiang, Dongwei and Zhang, Jingyu and Weller, Orion and Weir, Nathaniel and Durme, Benjamin Van and Khashabi, Daniel},
title = {SELF-[IN]CORRECT: LLMs Struggle with Discriminating Self-Generated Responses},
journal = {Proceedings of the AAAI Conference on Artificial Intelligence},
year = {2025},
volume = {39},
number = {23},
pages = {24266--24275},
doi = {10.1609/aaai.v39i23.34603}
}
EUVIMED