LessWrong 2.0 Reader

After adding the mean subtraction, the numbers haven't changed too much actually -- but let me make sure I'm using the correct calculation. I'm gonna follow your and @Adam Karvonen [LW · GW]'s suggestion of using the SAE bench code and loading my clustering solution as an SAE (this code).

v3 (KMeans): Layer blocks.3.hook_resid_post, n_tokens=100000, n_clusters=4096, variance explained = 0.8469745517
v3 (KMeans): Layer blocks.3.hook_resid_post, n_tokens=100000, n_clusters=16384, variance explained = 0.8537513018
v3 (KMeans): Layer blocks.4.hook_resid_post, n_tokens=100000, n_clusters=4096, variance explained = 0.6992422938
v3 (KMeans): Layer blocks.4.hook_resid_post, n_tokens=100000, n_clusters=16384, variance explained = 0.7112221122

You're right. I forgot subtracting the mean. Thanks a lot!!

I'm computing new numbers now, but indeed I expect this to explain my result!

Could you please ask about the specific examples of the Esperanto words? (I speak Esperanto.)

I think a similar example would be the adjective "Russian" in English, which translates to Russian as two different words: "русский" (related to Russian ethnicity or language) or "российский" (related to Russia as a country, i.e. including the minorities who live there).

(That would be "rus-a" vs "rus-land-a / rus-i-a" in Esperanto.)

I noticed this in a video where a guy explained that "I am Rus-land-ian, not Rus-ethnic-ian", which could be expressed in English as "I am a citizen of Russian Federation, but I am not ethnically Russian". On one hand, it can be translated without any loss of information; on the other hand, four words in Russian expanded to over a dozen words in English. More importantly, in 99% of situations the English speaker would not bother making the distinction, while a Russian speaker would be making it all the time.

Still seems to me that these things are rare, and more importantly, they don't seem to have the impact one might naively predict based on the Sapir-Whorf hypothesis. For example, one could naively predict that such language nuance would lead to less nationalism (because the country is less linguistically conflated with the dominant ethnicity), and yet, ethnic Russians don't seem less nationalistic.

Similarly, English-speaking feminists spent a lot of effort changing the default "he", through "he or she", to the singular "they" (and some of them go even further). But there are languages, such as Hungarian, which never even had "he" and "she", and have always used a gender-neutral pronoun. And yet, I don't think that Hungarians are less sexist than their neighbors.

Baumol effect vs Jevons paradox -- a trillion $ show down?

Prima facie the Baumol effect is bearish for sectors with larger gains in efficiency whereas Jevons paradox is bullish. So unsurprising Satya Nadella wants us to know about Jevons paradox which any way seems to be well known b/c of its connection to climate change concerns. Some misc thoughts on this topic…

Cameo appearances -- ‘Keynesian utopia’ & Kondratiev waves

§ 1. Definitions.

Let's say x alone (not y) becomes more efficient meaning… Either increased productivity ie more output from x. Or reduced cost from the marginal cost curve falling.

(Generalized) Baumol effect -- x becomes more efficient -> y becomes more expensive

This is ubiquitous. Remember than only relative cost really matters. So the only way to avoid this is if y ceases to be produced or if it gets pushed far enough down the marginal cost curve. But since the generalized Baumol effect can manifest in counter intuitive ways, it's still a useful thing to know. Most often the Baumol effect is used to explain why wages rise in sectors w/o much productivity growth.

‘Strong’ Baumol effect [N1] -- if a sector becomes more efficient, it shrinks relative to the economy

Unfortunately most sources (including Wikipedia?) do not clearly distinguish between the 2 Baumol effects

Jevons paradox also has 2 different forms, but these are quite similar to each other. The designations 1 & 2 are arbitrary.

Jevons paradox 1 -- x becomes more productive -> demand for x increases

Stated abstractly this is quite unsurprising, but again it can manifest in counter intuitive ways. Eg supposedly more efficient use of a natural resource often leads to greater exploitation of that resource.

Jevons paradox 2 -- x becomes less costly -> revenue from x increases

N1. Not actually stronger in the mathematical sense, but in practice it probably is.

§ 2. Discussion.

Whereas the generalized Baumol effect is ubiquitous, Jevons paradox & the strong Baumol effect are more uncertain. In the case of US manufacturing, the strong Baumol effect only shows up in the early 20th century. Until then that sector was growing relative to the rest of the economy in terms of employment, productivity, revenue & value added. So it seems the strong Baumol effect is some thing that shows up late so AI is probably safe for the time being.

But what about non AI software? Humans did not evolve to do computer programming -- a human doing computer programming is like GPT4 playing chess. AI will likely drastically reduce the cost of making non AI software. This is a mature industry w/o the explosive growth potential of AI so the strong Baumol effect is looking quite probable to me.

What does this mean for a company like Microsoft? Microsoft's deep pockets have only given it a slim edge relative to Anthropic & Deepseek. Non AI software is likely to go into relative decline in the near future. On the other hand Microsoft has so many people hooked on Azure, Outlook etc it will likely continue to make fat profits for a while. What does all this means for the market value of Microsoft & other software companies? I remain neutral on that question.

The emerge of the strong Baumol effect provides some mechanistic basis for sector specific Kondratiev waves [N2]. These could produce economy wide Kondratiev waves depending on how they line up.

Arguably both the Baumol effect & Jevons paradox are a bit over played. The relative decline of the US manufacturing sector to some extent reflects government actions that channel more & more $ into dysfunctional education & health care sectors & raise the cost of living via mandates & restrictions. In a Keynesian utopia w/o such things people could easily live off the manufacturing bonanza (w/o doing much work) so presumably the economy would be smaller but the relative size of the manufacturing sector may be larger. An apparent Jevons paradox need not reflect efficiency gains specific to that good; overall economic growth may be the decisive factor [N3].

N2. I see no reason why Kondratiev waves should be very close to 53 years long. But if we take Microsoft's IPO in 1986 as the beginning of a Kondratiev wave in non AI software, + 53 years gives 2039 as the end which seems plausible. The AI boom should more than offset this.

N3. The “Let's say x alone (not y) becomes more efficient” bit in § 1 excludes the latter possibility. But this would need to be modeled counter factually in situations like the industrial revolution.

Hzn

I'm more worried about counterfactual mugging and transparent Newcomb. Am I right that you are saying "in first iteration of transparent Newcomb austere decision theory gets no more than 1000$ but then learns that if it modifies its decision theory into more UDT-like it will get more money in similar situations", turning it into something like son-of-CDT?

I think for some games the GM world-model should change if it disagrees with applications of the rules.

The GM’s world-model either already incorporates the rules (in which case, there is nothing to change)—or else the GM’s world-model fails to incorporate relevant information, which is a mistake on the GM’s part. The GM should, after admitting the mistake, now decide how to rectify it. Here several options are available, but the key is that this is a mistake—an incorrect application of the approach I described.

You went through my examples of questions for a hypothetical shapeshifting power, answering as though you were GMing a game I was in. That makes sense, but I want to point out that I wouldn’t expect every GM to answer the same way.

Yes, for two reasons. First, not all GMs play correctly; many are bad, and wrong. Second, different GMs will obviously construct different world-models of their respective game worlds. Neither possibility in any way whatsoever contradicts or undermines anything I wrote.

I mean, sometimes the designer didn’t say. As a designer, sometimes I don’t bother to say; pointing out how much fine control the ability gives, and every other unspecified question of similar importance that might come up, is one way to wind up with this ability taking up about a page of the rulebook instead of two sentences.

That it’s possible to write a full-page version of an ability description does not imply that the optimal version is one sentence long. If the designer omits some information which makes the text simply impossible to unambiguously interpret, even in the absence of any “rules lawyers” and with the utmost desire to “keep things simple” and full willingness to extrapolate, infer, draw reasonable conclusions, etc., then that is simply designer error. Sometimes that happens. It’s pointless to pretend that it doesn’t happen or that it’s not a problem that needs to be solved. “How do we play a game where some of the rules are genuinely and irreducibly ambiguous—and no, ‘fix the rules’ is not an option” is a completely useless question to ask.

While this depends on what TTRPG we’re talking about, I think in most editions of D&D you’re incorrect about Charisma not coming from appearance.

Granted, the exact count of how many editions of D&D there are is a little fuzzy.[1] [LW(p) · GW(p)] Lets go with five, pretending for the moment that the current edition being called “5th edition” means everyone’s been counting normally. You’re wrong on three of them.

In how many of those editions does changing your appearance (in any way whatsoever, not just alter self or analogous abilities!) change your Charisma (or other mental stats)? As far as I know, none. (In 3rd edition, there is a disease that literally removes your face^[1], and it doesn’t do so much as a single point of Charisma damage. It could’ve! We know this because there’s another disease, listed on the very same page, which does do Charisma damage. That disease is soul rot, which “eats at the victim’s mind and soul”. So a disease damaging your Charisma is possible. But deleting your entire face doesn’t do it.)

But I did say this was from the Onyx Path forums, and they don’t do D&D. They do (among other things) Exalted, where Charisma is an Attribute alongside Manipulation and Appearance. Offhand, the Lunar charm Perfect Symmetry is shapeshifting magic that changes the way you look and the way the Appearance attribute normally interacts with social combat.

Well, then obviously that would change the answer I’d give in the “form-changing magic” example.

So, you talk about the GM having “a consistent, coherent, and complete world-model” that they use to answer the players questions. I think that’s often useful, but as a designer, it’s also useful to emphasize to the GM what parts of the model aren’t important that they can mess with easily and what parts are important that they should be wary of changing.

It is almost always a designer mistake to consider some parts of the model to be unimportant. All the parts are important. A good GM should understand how all of them interact.

As far as your commentary re: “humanoid”, I agree, of course. (Indeed, I am personally not a big fan of the “creature type” system as used in 3e-like systems. I think that it’s fundamentally a design mistake in the same way and for the same reason as many—perhaps most—hierarchical classification systems, namely that tree structures are not a good match for modeling a diverse space of entities with many dimensions of variation. I think that creatures should be categorized by non-exclusive “tags” rather than exclusive “types”, and further, that these should be kept as informal as it is possible to make them while retaining clarity. My views on spell categorization into “schools” of magic is similar, by the way.)

This, however, is an excellent argument, not for artificially declaring that Thou Shalt Switch Off Your Creativity When Thinking About This Topic—but rather, for thinking more carefully about what exactly we’re trying to accomplish with the design of this ability, what we want to be true of how such things work in our game world, how those things should interact, and so forth. (Ideally, of course, professional game designers would do this for us, and we’d pay them money for the result. Alas, well-known financial incentives usually drive the design of popular games in a different direction.)

I start getting suspicious a player is up to something unbalanced when they start trying to access magical abilities of creatures from that broad of a range and ask what’s up

Uh-huh, and the answer to this is to ask: “should there be character powers that give you magical abilities of a broad range of creatures?”.

There is, in my experience, always a design that is firmly grounded in the world-model and neatly and elegant cuts through “balance” concerns. The task is to find it. The way to approach said task is, first and foremost, not to abdicate the challenge—but that is precisely what declaring “chicanery: no” amounts to.

I disagree that’s what we’ve learned from this exercise. What I take away from this exercise is that if you try to unambiguously specify a model sufficient for multiple GMs to rule consistently, this power winds up being multiple pages long.

This is only true if you are writing for GMs who are stupid do not fully grok the “query the world-model” approach. It is entirely possible to unambiguously specify things much more concisely than that. I’ve done it.

Of course, publishers of mass-market-appeal products like D&D are in fact targeting GMs who can’t be expected to think. (This is especially true if organized play events are taking place.) But the fact that the financial incentives for appealing to the lowest common denominator prevent the large publishers from doing things in the optimal way does not, in fact, make that way any less optimal (where “optimal” means “resulting in the best play experience”).

Assuming the GM isn’t new or uncertain and looking to the rulebook for guidance. The rulebook is often the teaching tool, after all.

Yes, it certainly is. The approach I describe makes it a better teaching tool, not a worse one.

It seems a bit like you narrowed in on one kind of TTRPG and one way to play that RPG, and assumed that’s how they all ought to work?

One kind of TTRPG? Sure, in the sense that I specified—what I called “traditional roleplaying game[s]”. That’s a reasonably broad umbrella. Indeed, I’d hesitate to call anything else a “TTRPG” at all.

One way to play? Sure, if by that we mean: I claim that this is the right way to play, in the sense that it satisfies all desiderata which are typically adduced as alleged reasons to favor other play styles, while avoiding the problems produced by those play styles. What I described is an extremely flexible approach, not a rigid and narrow one.

“Wait a moment,” you might be thinking, “Said established for this exercise that a player can’t turn into ‘something that they don’t know about or haven’t seen.’ Surely the character hasn’t seen an Arakocra?” “What do you mean, I haven’t seen an Arakocra?” the player asks, “Didn’t you read my backstory? I grew up in the Icewind Mountains. The Arakocra mostly live in mountains. I met some there.”

Is the player saying this because they read some tie-in novels you haven’t and just assumed it was obvious? Is the player saying this because they really, really like birds and want to turn into a bird person? Is the player saying this because they have some twelve step plan that ends in a Candle of Invocation and your campaign final boss dead from six towns away with no saving throw? Good question! It might be nice to have a conceptual handle for ‘I’m fine if you want to be a bird person, but if it’s the twelve step plan I’m going to go over this like it’s Al Capone’s tax records.’

That’s actually exactly the wrong way to think about these things.

First, some peripheral bits:

“Didn’t you read my backstory?”

Well, didn’t you? If you didn’t—how come? GM mistake!

Is the player saying this because they read some tie-in novels you haven’t and just assumed it was obvious?

The game world is fully constituted by what the GM declares to be the case. If the GM doesn’t know something to be true of the game world, then definitionally it can’t be true. The correct answer to this one is “why in the world would you think that something you read in a tie-in novel (or anything else whatsoever that isn’t an extrapolation from what has already been established) is true of this game without asking me if it’s true”. “Does this world have aarakocra” is something the GM decides.

Now, to the main point:

It might be nice to have a conceptual handle for ‘I’m fine if you want to be a bird person, but if it’s the twelve step plan I’m going to go over this like it’s Al Capone’s tax records.’

If the player has such a twelve step plan, then they should be exactly as free to determine any aspect of their character’s backstory as if they have no such plan, no more and no less. To do otherwise is to mix levels of description, and to severely undermine player agency.

Consider: are we supposing that the twelve step plan in question… works? Genuinely works, breaking no rules, requiring no obviously nonsensical interpretations, etc.?

Supposing that it does work: do we, as designers, think that it should work? If so, then there is no problem.

If not, then we should alter the rules in such a way as to make it not work. That alteration is nigh-guaranteed to not involve banning player character origins from regions inhabited by aarakocra. (Changes to rules that patch undesirable exploits basically never look like that.) But whatever the change ends up being, it should be a change that applies to everyone equally. (And if we can’t imagine a way to fix this bug without preventing people who just want to be bird persons from doing so, then perhaps we ought to find a different hobby, as that would betray a rather shocking lack of imagination.)

Does a fireball do more damage or go further if it’s in a confined space? The answer used to be Yes back in ~D&D 2e. The answer is No in D&D3.5. What changed, all of the GM’s world-models? Nah, the rules did

False dichotomy. A GM who adopts the new rules thereby changes his world-model; a GM who keeps the old rules… keeps the old rules. In other words: both changed, or neither, definitionally.

My world-model doesn’t really give me a principled answer either way.

If you have rules that give you an answer and you follow those rules, then your world-model absolutely does give you a principled, unambiguous answer.

The DM’s world-model is the ground truth. Answers to all questions flow from it.

Please go read Baron Munchausen, come back, and tell me if that sentence makes any sense as a global statement about tabletops.

See below re: scope of my claims.

There’s a lot of fluff, go ahead and skip it.

I’ve heard this sort of sentiment many times. I think that it’s wrongheaded. Did you know that the word “fluff” appears nowhere in the 3.5 or 5e PHBs or DMGs (excepting a box quote, in the 5e Player’s Handbook, of a passage from a tie-in novel)? What’s “fluff”? There isn’t support in the rulebooks for the notion that some parts of the rules are meaningless and skippable. Sure, modify whatever you like—that’s “Rule Zero”—but constructing this category of “fluff” is not the way to go.

If you aren’t trying to talk about all RPGs but just the subset that structurally resemble recent editions of D&D, that’s fine, but admit you’re talking about a much smaller subset of games.

Not “recent editions”, but basically all editions other than 4th (which was, as you are no doubt aware, a quite different sort of thing). I refer you once again to the term “traditional roleplaying games”, as I have defined and used it.

″...Maximum amount of player engagement and satisfaction”

″...The best at allowing newcomers to engage with the game...”

Those are confident and broad claims.

Yes.

Charitably, you’re accurately describing the players you’ve happened to game with, and the players you’ve gamed with are subject to some selection effect which explains the limited variety.

No, my claims are meant to be as broad as they appear to be.

My own experience has had a lot of players who light up when offered the chance to take the narrator spotlight, or who report they bounced off of games when they brought up that the GM wasn’t following the rules and the GM leaned too hard on Rule Zero. It’s a big hobby.

I do not doubt your word on this, but none of this contradicts what I wrote.

In some games, the GM’s world-model should lose if it disagrees with applications of the rules

See the start of this comment. The premise of this claim is fundamentally mistaken: either the GM’s world-model already incorporates the rules (and the GM has already decided to use that rule as written, or to change it—in which case he should of course apprise the players of this, in advance), or else the GM has made a mistake. Now, mistakes happen, but this cannot be a criticism of the approach I describe any more than it can be a criticism of a piece of music to observe that a novice musician has played a wrong note in his performance of it.

However, if your claim is that a TTRPG must have no or almost no dissociated mechanics in order to be well-designed and of comprehensive scope, that seems overbroad.

A TTRPG must have no or almost no dissociated mechanics in order to be a traditional roleplaying game. However, more broadly: yes, the games you list definitely have a smaller scope (some of them have a dramatically smaller scope) than, e.g., D&D (of most editions). (Some of them are also much more poorly designed, and often—though, of course, not always—this is precisely due to not fitting into the framework I describe.)

[D&D] trying to position itself as the Everything RPG for business reasons despite a design that isn’t nearly as general as its public perception sometimes suggests

On the contrary, D&D has a design that is much more general than most people perceive it as, and this is precisely because it is so amenable to the approach I describe. It is extensible and generalizable in a way that basically none of the other games you describe are—and this is exactly and directly a function of having (almost) no dissociated mechanics, and a design (and tradition) that supports “query the DM’s world-model” style play, backed by rules of varying comprehensiveness but unlimited scope. (If you doubt this last point, try making a list of all published adventure modules from just TSR or WotC, then make a list of all of the fiction genres encompassed by said adventures, and compare this list to the list of fiction genres encompassed by almost any randomly selected… say, twenty… non-“traditional roleplaying game” “TTRPGs”.^[2])

You’re claiming TTRPG design and implementation as a topic on which you are quite knowledgeable. Based on what you’ve written here about TTRPG rules, my observation is that you have a narrow view of the field and where other players might be coming from, and you don’t even seem aware of that. From my perspective, you’re confidently describing what birds are like, and you’ve managed to make scope so narrow that not only have you excluded penguins and ostriches, you’ve ruled out chickens, seagulls, and hummingbirds, and I’m starting to think an osprey wouldn’t count.

I understand why you might think this. However, I have seen every single one of your counterarguments before (sometimes nearly verbatim, and certainly including all of the examples of other games which you mentioned^[3][4]). I am very well aware of the games you describe, the play styles you describe, and the player preferences you describe.

Please interpret my commentary in light of the above.

First of all, "the most likely outcome at given level of specificity" is not equal to "outcome with the most probability mass". I.e., if one outcome has probability 2% and the rest of outcomes 1%, 98% is still "other outcome than the most likely".

The second is that no, it's not what evolutionary theory predicts. Most of traits are not adaptive, but randomly fixed, because if all traits are adaptive, then ~all mutations are detrimental. Because mutations are detrimental, they need to be removed from gene pool by preventing carriers from reproduction. Because most detrimental mutations do not kill carrier immediately, they have chance to randomly spread in popularion. Because we have "almost all mutations are detrimental" and "everybody has mutations in offspring", for anything like human genome and human procreation pattern we have hard ceiling on how much of genome can be adaptive (which is like 20%).

Real evolutionary theory prediction is like "some random trait get fixed in the species with the most ecological power (i.e., ASI) and this trait is amortized against all the galaxies".

Nod. 

The thing I'm currently hearing you saying (either in contrast to this post, or flagging that this post doesn't really acknowledge) is:

• there's a bunch of technical knowledge (which is a different type of thing than "metacognitive skill training", and which also requires a ton of work to master)
• the amount of work going into all of this is just, like, a ton, and the phrasing in the post (and maybe other conversations with me) doesn't really come close to grappling with the enormity of it?

Are there other things you meant? 

I happen to care about animal suffering, and I am as baffled as you about the move of caring about animal suffering for explicitly anti-speciecist reasons yet dismissing wild animal suffering. Seems pretty inconsistent.

Maybe it originates from a sort of wishful thinking? As in "looks intractable, therefore I wish it were unimportant, therefore it is".

What is this "should" thingy you are talking about? Do you by chance have some definition of "shouldness" or are you open to suggestions?