The Moral Gap of Large Language Models
Fuente:
arXiv
Saved in:
| Main Authors: | , |
|---|---|
| Format: | Preprint |
| Published: |
2025
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866916861726162944 |
|---|---|
| author | Skorski, Maciej Landowska, Alina |
| author_facet | Skorski, Maciej Landowska, Alina |
| contents | Moral foundation detection is crucial for analyzing social discourse and developing ethically-aligned AI systems. While large language models excel across diverse tasks, their performance on specialized moral reasoning remains unclear.
This study provides the first comprehensive comparison between state-of-the-art LLMs and fine-tuned transformers across Twitter and Reddit datasets using ROC, PR, and DET curve analysis.
Results reveal substantial performance gaps, with LLMs exhibiting high false negative rates and systematic under-detection of moral content despite prompt engineering efforts. These findings demonstrate that task-specific fine-tuning remains superior to prompting for moral reasoning applications. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2507_18523 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | The Moral Gap of Large Language Models Skorski, Maciej Landowska, Alina Computation and Language Computers and Society Human-Computer Interaction Machine Learning Moral foundation detection is crucial for analyzing social discourse and developing ethically-aligned AI systems. While large language models excel across diverse tasks, their performance on specialized moral reasoning remains unclear. This study provides the first comprehensive comparison between state-of-the-art LLMs and fine-tuned transformers across Twitter and Reddit datasets using ROC, PR, and DET curve analysis. Results reveal substantial performance gaps, with LLMs exhibiting high false negative rates and systematic under-detection of moral content despite prompt engineering efforts. These findings demonstrate that task-specific fine-tuning remains superior to prompting for moral reasoning applications. |
| title | The Moral Gap of Large Language Models |
| topic | Computation and Language Computers and Society Human-Computer Interaction Machine Learning |
| url | https://arxiv.org/abs/2507.18523 |