mursepg.bsky.social @mursepg.bsky.social · 24/09/2026He's saying that AI biases can't be explained by training data. Other people contend that just because HE can't explain or doesn't believe training data bias explains model bias, doesn't make it so. 250
mursepg.bsky.social @mursepg.bsky.social · 24/09/2026The further context is that Wil thinks we should use anthropomorphic language to describe what generative AI does. And that there's some mysterious quality in AI that might as well be sentience because he can't explain how AI works therefore it's the same as the inability to explain human sentience. 150
NewAccountWhoDis @newaccountwhothis.bsky.social · 25/09/2026It's the same principle, though. Complex systems have emergent properties. The actual AI models are for most intents and purposes a black box. They can get biases directly from training data, but they can also create their own biases in the process of training. 000
