Could you please clarify what parts of the making of the above comment were done by a human being, and what parts by an AI?

Sure, but plausibly that's Scott being unusually good at admitting error, rather than Tyler being unusually bad.

It's still pretty interesting if it turns out that the only clear example to be found of T.C. admitting to error is in a context where everyone involved is describing errors they've made: he'll admit to concrete mistakes, but apparently only when admitting mistakes makes him look good rather than bad.

(Though I kinda agree with one thing Joseph Miller says, or more precisely implies: perhaps it's just really rare for people to say publicly that they were badly wrong about anything of substance, in which case it could be that T.C. has seldom done that but that this shouldn't much change our opinion of him.)

The language used by some of the LLMs in answering the question seems like pretty good evidence for the "they one-box at least partly because Less Wrong is in their training data" theory. E.g., if you asked a random philosopher for their thoughts on the Newcomb problem, I don't think most of them would call the predictor "Omega" and (less confidently) I don't think most of them would frame the question in terms of "CDT" and "EDT".

1. Unfortunately "love" means a lot of different things. If you answer "what is the best thing about romantic relationships?" with "love" then you haven't done anything to distinguish e.g. "feelings of gooey happiness in one another's presence" from "fierce commitment to giving the other person as good a life as possible" from "accepting the other person no matter what they do or what happens to them" from etc., etc.
2. I think it's highly debatable whether what John is describing is "love" in any of the usual senses of that word. It's clearly related to it, just as it's related to (say) trust, but it's not the same thing.
3. Even if it is, John has identified (or at least claims to have identified) a very specific thing about the phenomenon that is much more specific than "love".

Pedantic note: there are many instances of "syncopathy" that I am fairly sure should be "sycophancy".

(It's an understandable mistake -- "syncopathy" is composed of familiar components, which could plausibly be put together to mean something like "the disease of agreeing too much" which is, at least in the context of AI, not far off what sycophancy in fact means. Whereas if you can parse "sycophancy" at all you might work out that it means "fig-showing" which obviously has nothing to do with anything. So far as I can tell, no one actually knows how "fig-showing" came to be the term for servile flattery.)

The Additional Questions Elephant (first image in article, "image credit: Planecrash") is definitely older than Planecrash; see e.g. https://knowyourmeme.com/photos/1036583-reaction-images for an instance from 2015.

They're present on the original for which this is a linkpost. I don't know what the mechanism was by which the text was imported here from the original, but presumably whatever it was it didn't preserve the images.

Yes, that sounds much more normal to me.

Though in the particular case here, something else seems off: when you write $f\left(x\right)$ you would normally italicize both the "f" and the "x", as you can see in the rendering in this very paragraph. I can't think of any situation in actual mathematical writing where you would italicize one and not the other in order to make some distinction between function-names and variable names.

For that matter, I'm not wild about making a distinction between "variables" and "functions". If you write $f\left(x\right)$ and also $\sin \left(x\right)$ then it would be normal for "f" and "x" to be italicized and not "sin". I was going to say that the reason is that f and x are in fact both variables, and it just happens that one of them takes values that are functions, whereas sin is a fixed function and you'll never see anything like "let sin = 3" or "let sin = cos" -- but actually that isn't quite right either, because named mathematical constants like e are usually italicized. I think the actual distinction is that single-letter names-of-things get italicized and multiple-letter ones usually don't.

(Brief self-review for LW 2023 review.)

Obviously there's nothing original in my writeup as opposed to the paper it's about. The paper still seems like an important one, though I haven't particularly followed the literature and wouldn't know if it's been refuted or built upon by other later work. In particular, in popular AI discourse one constantly hears things along the lines of "LLMs are just pushing symbols around and don't have any sort of model of the actual world in them", and this paper seems to me to be good evidence that transformer networks, even quite small ones, can build internal models that aren't just symbol-pushing.

There's a substantial error in the post as it stands (corrected in comments, but I never edited the post): I claim that the different legal-move-prediction abilities of the "network trained on a smallish number of good games" and "network trained on a much larger number of random games" cases is because the network isn't big enough to capture both "legal" and "good strategy" well, when in fact it seems more likely that the difference is mostly because the random-game training set is so much larger. I'm not sure what the etiquette is around making edits now in such cases.