AI Transparency

Understanding neural networks through sparse circuits

OpenAI is exploring mechanistic interpretability to understand how neural networks reason. Our new sparse model approach could make AI systems more transparent and support safer, more reliable behavior.