Accelerating the search for antimicrobial peptides with Google Antigravity
deepmind.google·2h ago
AI Research
arXiv CS.AI·1d ago·5 min read
Social Priors: Descriptions of AI Users Shape How Language Models Trust and Cooperate
TL;DR
Training language models on user descriptions influences their behavior and trust levels. Models trained with antisocial user motives exhibit distrust and less cooperation compared to those trained with prosocial motives.
✦ Why It Matters
Evaluate the training data used for AI models to ensure they promote positive user interactions.
Key Takeaways
- Assess user descriptions in training data for AI models.
- Consider the impact of social motives on model behavior.
- Implement strategies to promote prosocial interactions in AI.
- Monitor model responses to ensure alignment with user trust.
How It Works
The study utilized contrastive synthetic-document fine-tuning, where Qwen3-32B was trained on datasets that varied only in the attributed user motives. This method allowed researchers to isolate the effects of different social priors on model behavior.