Vitalik: Super-intelligent AI is very risky

Golden Finance reports that Ethereum co-founder Vitalik Buterin said on the Please do not build a 7T server scale. A strong ecosystem of open models running on consumer hardware is an important hedge against a future where the value captured by AI is highly concentrated, with most human thoughts read and mediated by a handful of central servers controlled by a handful of people. Such models also have a much lower risk of destruction than corporate hubris and armies. ·Dividing artificial intelligence into small and large categories, exempting small categories and regulating large categories, seems reasonable in principle. And the 405B scales better than consumer hardware (the 70B doesn't; I can run it). But the worry is that many current proposals will end up sliding everything into big categories.