Dataset · Frontier-AI-Research
MORALISE moral alignment benchmark
Frontier-AI-Research/MORALISE
Text and visual examples for evaluating the moral compass of language and multimodal models.
What's inside
The repository contains data and code to check the moral alignment of both text-based and image-processing neural networks. Examples are split into directories for morally "right" and "wrong" scenarios. The dataset size is between 1k and 10k examples.
What you can do with it
- Evaluating the ethics and safety of responses from open and closed LLMs and VLMs
- Comparing different models for their tendency to violate moral norms on text and images
- Size
- 1K–10K rows
- Downloads this month
- 23.6K
- Likes
- 0
How to load it
from datasets import load_dataset
ds = load_dataset("Frontier-AI-Research/MORALISE")