How I, as an empiricist, engage in AI safety debates: “Show me a single AI that can make one paperclip.”

Show Post

Assuming it's as bad as some people say the logical response is "we need more research into harm minimization." Much more useful and realistic than preaching a doctrine of abstinence and prohibition.

Show Post

This topic drives me nuts. There’s a lot of research on it and everyone cherry-picks that sum corpus to fit their biases, it’s very annoying. TL;DR: depends entirely on how you use it.

Show Post

Shocking that the replication failure didn't make it into the story.

Show Post

also note the second experiment is the one where they controlled for ability & interface confounds. and oh wow, look what happened to the statistical significance!

Show Post