But also: it's not about any one example -- it's about the totality. Little instances of misbehavior do eventually add up. Whenever I peer under one of these rocks I seem to find an AI doing something weird and a human declaring "I don't like this, so it's an instance of my new scientific construct, AIs Doing Things I Personally Dislike When I Totally Wanted Them To Do Other Stuff Instead, How Dare They."
This is why, I think, we should also be focusing on the broader totality that these things are being grown within. Optimization is a very weird thing in t... (read more)
This is why, I think, we should also be focusing on the broader totality that these things are being grown within. Optimization is a very weird thing in t... (read more)