Decentralization and transparency are needed to understand model bias
Key Challenges in Building Decentralized AI Infrastructure | AI / ALL Summit (Sahara AI)
“Models’ biases are as good as the data it gets fed and the data it gets reinforced… it was heavily biased because it was reinforced by people in San Francisco… I think one of the reasons why decentralization needs to exist, more importantly transparency, is to understand these biases.” — 00:18:02
Context: Bias/explainability question; he cites Microsoft’s Tay turning racist, GPT-3’s RLHF lean, and Google Gemini’s ahistorical founding-fathers images. Quote spans caption blocks.