The Techno Optimist Libertarian Take on AI Risk

The problem is not that AI is too dumb, the problem is that it's too dumb. What really matters is the amount of compute that you invest in the amount of data that you invest to kind of actually update those parameters so that they have values that are good. The real risk is losing the geopolitical competition to adversarial states like Russia and China. And I would say the number of parameters is up there, just like very quickly,. Like this is the Jeremy nitpick, but  you could instantiate a version of GPT for today on your computer with like 500 trillion parameters, assuming your computer had big enough memory to store that with all random numbers.