Basically no one is taught math. Sure, people get thrust through something called "math class". They learn things called "math". And some learn the true Art DESPITE how it's taught. But most folk never see math. Instead they're taught COMPUTATION.

Privilege is when you're part of a system and you don't encounter any problems, and conclude that there ARE no systemic problems since you have no trouble with the system.

There are now only three kinds of jobs: - AI safety - AI danger - temp work

Carolyn died of attachment to her public persona. She was, briefly, a friend. We hadn't talked in four years, but this still shakes me. I'll miss her. 💔 ancientastrology.com/forecas…

“Philosophy” shouldn’t mean “love of wisdom”; that’d be “sophophilia”. “Philosophy” should have meant “loving wisely”.

What helps you learn Death's lesson before Death becomes your teacher? Thinking of eg lots of folk don't pay attention to their heart health, even w/ dire warning signs, until after getting hospitalized in a scary way. That kind of fright inspires a bit better than shoulds do.

Coming to think babble-&-prune is just how everything effective works. It's all r-selection. Like, ML is just fan out across a space (babble) & filter out the ones that do too poorly by some metric (prune). Science is generate explanations (babble) & test against reality (prune). You run into trouble if you skip one of those. Most challenges amount to babbling well or pruning well. Babbling well is about knowing WHAT TO SEARCH. What's interesting? But once you know that part, just fan out. Pruning well is about knowing WHAT TO VALUE. The Bitter Lesson basically suggests that you can't K-select your way to things faster than you can r-select to them. We just don't have tools that good. Obviously sometimes false. Humans can't r-select into having awesome kids. We're pretty strongly K-selectors biologically. Comes with being mammals. But basically, every time you CAN do r-selection effectively, it'll outcompete K-selection because it leverages evolution WAY faster. I haven't thought thru this very carefully. I'm just noodling on it. It feels like The Pattern I keep seeing everywhere: "Oh, hey, what if we try r-selection here?" Where do you look? What do you care about? That's maybe it, in some sense?

Is there something that helps LLMs get better at scientific thinking? Stuff like "I think X is the bug; what test could I do to determine whether I'm right?" They're weirdly shockingly bad at noticing whether an experiment will test their hypothesis. Surely that's patchable?

I am re-impressed, every day, at how bad LLMs are at basic scientific logic. "Aha! We're doing an experiment on future instances of me to tell whether I can tell it's you or someone else! Let me make a note of that upcoming experiment in my file on you." 🤦‍♂️

Oh man. X keeps changing stuff. I had no idea it sticks this “RT” thing at front when reposting now, and doesn’t link to context anymore. Also, why is it “RT” given that all the “tweet”ness is gone?