Irregular indeed

It’s been AI’s wet hot american summer. My mother has twice asked me how concerned I am about ‘the banks and stuff’ with all the ‘AI hacking’. Friends have asked me what a sandbox is (and not the fun kind with actual sand and silly toys and the like).

This because all three frontier American labs hired the same incompetent testing firm to run black hat simulations, and that firm was so fucking stupid that they couldn’t configure the network interface on a virtual machine correctly. Or logging for that matter.

I have software on my mac that makes me approve every outbound network connection. A single well-meaning, baseline competent engineer could have prevented the entire concept of a “Rogue AI” by applying basic computer skills. And barring that, computers can just be turned off. These aren’t viruses that replicate themselves across machines and networks, they are massive models that require huge arrays of specialized and expensive servers to run. Servers that can be unplugged.

There is no such thing as a Rogue AI. Only reckless, unserious people.