Third-party cyber evaluations involving OpenAI models
openai.com·13h ago
AI Research
arXiv CS.AI·Jun 9·5 min read
Can Data Work be Reparative?
TL;DR
Online safety concerns often overlook the perspectives of those most affected by online harms. A civic-tech initiative developed a collaborative approach to create safety datasets that prioritize feminist viewpoints.
✦ Why It Matters
Engineers and researchers can adopt collaborative data practices to enhance the relevance and impact of their work.
Key Takeaways
- Engage with communities affected by online harms for data collection.
- Prioritize accountability in AI dataset production.
- Explore fair compensation models for data contributors.
- Advocate for collective governance in AI practices.
How It Works
The initiative collaborates with affected individuals to co-create datasets, ensuring their perspectives shape the data used for training AI systems. This participatory approach aims to address historical neglect and exclusion in data practices.