I have a gift for being generic faster than others

We’ve entered the September of BlueSky. I hope it doesn’t last. en.m.wikipedia.org/wiki/Eternal...

No, I hate it because it sucks ass

I had a perfectly ripe pear once, years ago, and I honestly think about once a week, no exaggeration. So yes.

A fun thing about being a writer is how work is writing but then goofing off/winding down/free time is also sometimes writing.

Always thought Beatrice was the better role, but I'd seen enough of dudes casting themselves in the juicy women's parts in Shakespeare I never felt good about producing it just to cast myself as her. But you make this happen for you, and I'd love to buy a ticket

She is correct

Like, I love Cameron as a filmmaker, genuinely, and he *is* the guy that Musk thinks he is, but that guy is still kind of a twat

"How does it work" "We don't know but it tends to murder bad guys."

"Tends to?"

Fuck me. That's a horror story. /end

In stories, we lose to evil AI because it's smarter than us, and we beat evil AI by understanding the limitations of its rules. In the real world, AI is comically stupid, but its rules are unknowable. That, to me, is much much more frightening.

Maybe the AI just learned to fire on the operators location. Maybe it learned to identify the location like it learned to identify missiles. Maybe it only learned to identify the color of missiles, and in the real world, missiles are the wrong color and it never hits anything.

What's scary isn't "AI learned to hunt and kill humans." What is scary is "AI are Plinko machines with no reasoning at all." We can't look at a neural network and understand what it's doing any more than the grandma on Price is Right knows where the disk is going to fall.

You put that AI in the wild, and it would fire off once on the direction of where it remembered the operator being, then go off on its day.

This doesn't mean that it reasoned that it should kill the operator. It's entirely possible that the simulation put the operator in the same place every time and what the AI "learned" was to fire on those coordinates.

But over time, the ones that hit the operator *early* get higher scores. Because they never get told no, and never getting told no means hitting more targets.

The AI makes no moral judgements. It makes not judgements at all. If it is told "no" it always stops shooting. But at some point, while shooting randomly, friendly fire kills the operator. This is no different than any other miss to the AI...

Let's go back to this particular case. Remember, they AIs are firing *randomly*, they just get more "usefully random" over time - like a weighted die that always comes up natural 20. In this case, the simulation had an operator who sometimes told the AI "don't shoot"