Figuring out neural networks thru sparse circuits

OpenAI is exploring mechanistic interpretability to know how neural networks explanation why. Our new sparse fashion means may just make AI programs extra clear and strengthen more secure, extra dependable conduct.