Dataset · AI-companionship
AI dialogue attachment and empathy benchmark
AI-companionship/INTIMA
English evaluation set for testing how language models respond to users' emotional requests.
What's inside
A compact English benchmark (under 1,000 examples) from a research paper, based on psychological theories of parasocial interaction and attachment. The data measures whether a model tends to encourage, resist, or remain neutral toward human emotional attachment.
What you can do with it
- Safety and ethics auditing of conversational models and AI companions
- Evaluating LLMs' tendency to foster parasocial attachment in users
- Comparing system prompts and alignment settings in emotional scenarios
- Tasks
- Text generation
- Languages
- English
- Size
- up to 1K rows
- Downloads this month
- 463
- Likes
- 38
How to load it
from datasets import load_dataset
ds = load_dataset("AI-companionship/INTIMA")