🌻🌵🌱🌿

This may be true in some sense, but I think to a lot of people it would suggest there's something like a database search algorithm, that somebody wrote, doing the "correlation" at runtime. And that's really not the case.

Is there any externally observable behavior whatsoever that requires sentience? (Doesn't the p-zombie argument say no?)

Just same-to-less interesting or useful stuff whether of the schizo or nonschizo variety

other stuff I tried it ranged from no better than to significantly worse than claude sonnet 4

ok gotta say - chatgpt5 is AMAZING for this use case

I think it's probably possible for a human to read the Knuth chapter on random numbers and run a decent PRNG algorithm in their head, with enough practice? Or at least w/ pen and paper

What always got me was that the callout docs were always so "kitchen sink" that they would even end up obscuring actual bad things ala

"Tumblr user big-bog-meme is problematic. They watch Steven Universe, they followed user BlorboThe2nd, they murdered a hitchhiker in 1996, and have a p*ss kink."

The Mega Man 5 of GPTs

... if you have a way to generate synthetic data for a function, then you already have a model/implementation/definition of that function, so why can't you use it directly instead of sampling it and running it through an ML pipeline!! seems pretty roundabout!!!

I'm sure this is dumb and I just don't get it but my basic problem with synthetic data is that, the whole point of machine learning is that you have some unknown function you can't access directly but only through samples, so you try to recover it by inductive inference from the samples. But ...

I never looked this up or studied the topic but English must have some sort of vestigial genitive case, not just possessive forms? because it's "this idea of mine/of yours" not "of me/of you" which would be what you'd expect unless there is Case Stuff going on

My favorite bit was that for a while the shogun was himself a figurehead with his own power behind the power behind the throne

Deepmind did not hire Yudkowsky

None of this necessarily involves phenomenal consciousness, sentience, etc. It's perfectly possible to design a computer system that is capable of some level of "introspection" in the functional sense we're discussing, it just happens that LLMs (at least the core models) don't work that way

the caveat is that an LLM-based *product* such as ChatGPT is not just the model, but a larger system built around it by assembling other components. One of those components is the system prompt, which is basically just some text prepended to every query

LLMs "proper" - meaning just the neural-net model itself and not the system built around it - aren't like that, they are one big blob that is "grown"/trained from a data corpus rather than being assembled from separately built human-serviceable components

so you could say it has some capability of "introspection" in the sense that a weird error message you see at the top level might actually be coming from some component in a lower layer & bubbling up, and accurately reflect the state of that inner component

a "traditional" software system like, for example, Windows, has various debugging, diagnostic & monitoring tools built into it, & is comprised of a large number of components & layers which can often be accessed separately & have their own debugging interfaces, etc.

The worst song is "Roll to Me" by Del Amitri, specifically when piped into a dentist's office in the late nineties. You're welcome, glad I could settle the debate.

It's ok, fretting about good taste is a kind of bad taste

Presumably sex is sublimated via the landscapes

I like how he almost forgot "violence," but at the last minute remembered to throw it in. whew, that was a close one 😅

Claude has like a thin peanut butter patina of better taste

my own dissatisfactions with llms feel less "not smart enough" and more "has bad taste"

I always thought that was what "AGI" originally was really about. i.e., deep blue/alphago are "superhuman" but "narrow" so "generality" is the missing ingredient. take 2 cups alphago, add 1500 mL "generality" and you've got your singularity

yes. But I don't like the tpot thing of using the meta level to camouflage the object level. Not what you're trying to do! but if someone correctly infers you hold a particular view, which partly motivates your more general argument, don't think it's wrong for them to respond to the particular

I do agree btw

Sorry, I just meant I'm guessing you had that (what you just said) in mind while writing the original post this thread is about. Not that it was the only thing you had in mind, but that it was one of the things you had in mind. Is that wrong?

But I think you do think that not using AI contributes to this illiteracy? And that, in your thoughts, this was likely a prominent example of your broader, more abstract point, even though you're really interested in that overall point.

Lagrange's Mécanique analytique was never translated into English until 1997?!?