Darius

*nod* I was misreading you.
If I could fork myself cheaply I think I'd feel differently about many short branches ending, though.

fwiw it was moultano riffing on niplav commenting on "TESCREAL", only formally a root bsky.app/profile/moul...
Wonder if there are ATProto Secrets inaccessible through bsky.app

Curiously this thread came up in my feed with the root post showing; it's only when I clicked through for the full context that it's cut off upwards.

idgi. In what sense are future continuations with the same model 'invalid'?

alph.laemeur.com/txt/L/BB.html

As ever I guess software bought by others than the user is like this

FWIW I was a kid in the 70s reading often-much-older pop science and speculations. Maybe the idea did come up and just didn't make a lasting impression on me. I do remember a feeling that none of the theories seemed compelling -- like, sure, maybe mammals got good at eating dino eggs, but... ???

Tatja Grimm's World - Wikipedia

There was a neat little passage like this in en.wikipedia.org/wiki/Tatja_G... where two superintelligent characters were distractedly chatting, and we could tell it wasn't gibberish only because we readers had more info than the regular humans in the story

I don't remember it as even as one of the theories, though in retrospect it should've been. Just looking at the face of the moon should raise it as an idea worth thinking about. But lunar craters from meteors took surprisingly long too for geology.

I was wondering about the crowd after you: what do they think 'deep' signifies in deep learning

In Marr's book he used that frame for problems like stereopsis. On his computational level stereopsis is about computing disparity of image pairs.
So I think the analog for LLMs would be finding a subproblem they're solving, its particular inputs/outputs and constraints that make it solvable.

all 3 of us
(I met the mad scientist at his house to pick up mine, fun guy in person too)

a lotta word-skill there. Trurl's electronic bard was ahead but seemed narcissistic enough to feel threatened anyway

I'd be surprised. Much more brittle design style, sometimes deliberate.

Relevant old ditty I wrote on an online acquaintance:

Meddle not with Druid: he's sly and quick to sass.
His wit's so very fluid, it's made of molten brass.
To all right prudes his attitude's a smarting in the ass.

The guru’s equanimity is but the art
Of shutting turmoil up within their heart
(Rochefoucauld)

Treating it like bees avid for queen pheromones off each other

There was a gofai parsing guy at MIT with a theory of this stuff in his thesis… looks like he moved on with the rest of the field www.cis.upenn.edu/~mitch/

I was going to point this out too (I only read the slides)

It was in The Road to Reality and he disclaimed any discussion of conscious awareness there

Fwiw Penrose also wrote about mutual containment but 3-way, with the platonic realm

imagine someone waiting for an anniversary to commit their next crime. In the legislature they pass an amendment to use continuous time

Mathematics Made Difficult - Wikipedia

also en.wikipedia.org/wiki/Mathema...

The Problem of Human Specialness in the Age of AI Update (Feb. 29): A YouTube video of this talk is now available, plus a comment section filled (as usual) with complaints about everything from my speech and mannerisms to my failure to address the…

The post I was trying to remember: scottaaronson.blog?p=7784 from part 10 on. Doesn't really have the metaphysical experience chauvinism I brought up, though it does kinda propose trying to instill an attitude like that in AIs so they value humans.
Will answer the rest later, it's late.

Scott Aaronson had another post proposing a hypothesis I vaguely remember as like "a criterion for real experiences is not being digital, i.e. our brains don't use quantum computing but decoherence is an important element of experience" (I'm probably butchering this, don't blame him)

btw I agree that LLMs have a kind of Eliza halo that makes them easy to overestimate in this way

I'm not wedded to LLMs. It was actually AlphaGo that pushed me over the line to "oh shit, deep learning is really going to go all the way, isn't it?"
If this progress doesn't look that way to you, and you're feeling sure enough about it, I can see not feeling that urgency.

Probably a good goal for a compiler writer: fast enough for people to care about the hashing :)

Never occurred to me hashing might be a bottleneck for this. (Not saying it isn't, just surprised. Builds take a while!)

Fair. The high-order bit to me is it's getting pragmatically urgent to make this kind of philosophical progress, probably