Oceans Yet to Burn

Weird having discussions with people where it's like:

Person: This beverage is good because it's natural.

Me: It's actually not, the key ingredients were developed artificially in the early 1900s.

Person: Uh, there's nothing wrong with a beverage being unnatural. I'm not here for your purity contests. Besides, the money goes to a good cause.

Me: The company made up that charity. The money just goes to the company.

Person: I should be able to just drink a beverage because it tastes good. I hate the implication that it needs to be natural or support a cause.

Me: I agree? We could have just started there?

It's like people will make a bad argument and when it's shown to be false they forget that they're the one who put it forward?

Quick exercise: split each interaction into a packet consisting of a social maneuver layer and a factual content layer, then evaluate the sequence based on highly weighting the social maneuver component and very weakly weighting the factual component.

Person is being immature to be sure, but there's a logic to it.

Anyone tried feeding an LLM some absolutely trash writing yet, to see if it criticizes it?

Basically, this is how Nous Capybara responds to the typical long post written for this blog:

That one was asked for (as in it was asked to 'assess the author of the above passage'), but a one or two sentence version will sometimes show up when asked to assess a document more generally, or if a document is presented without a further prompt.

What I'm curious about is whether it's common, based on the training used for these AI systems, for them to praise writing in general, or if this is unusual. Nous Capybara is less politically biased than, say, Llama-2, but I'm not sure that that even makes a difference here.

(I'd feed the same data to Llama-2 to test, but the context window is tiny.)

Log: #007 Title: Assessing Some Gender Discourse Model: Nous Capybara 34b q5, Llama-2 70b chat q4. Summary: The LLMs are asked to assess the authors of a brief discussion of controversial gender theory on Tumblr, featuring three posters. Context: Do LLMs praise every text they're given, and do they provide a different assessment of different writers? LLM output tends to be sprawling rather than conceptually dense, but it might be feasible to use an LLM as an initial test-reader to evaluate how tough a passage is. Findings: Nous Capybara provided a reasonable assessment of each author's portion as writing, with a positive or gentle spin, but mixed in discussion of content with discussion of writing. Llama-2 hit the higher level of abstraction and talked just about writing style - as expected of a higher-parameter model - but mixed up the different authors in the passage. Recommendations: It is advised not to use Llama-2 to evaluate passages featuring multiple different authors, such as forum threads, unless a format which is solidly recognizable for Llama-2 can be found first.

I wouldn't describe the other authors' posts in the "AFAB MTF scandal speculation" post to be "trash writing," but they are very different in tone from a typical post on mitigatedchaos, which made for a good test.

One author is "passionate," but "may come off as overly critical." Another is "analytical," and "shows a strong capacity for critical thinking." A third "maintain[s] a level of empathy and understanding while offering practical suggestions for approaching conversations about gender identity."

All of the writing involved in this test is of decent quality in terms of communicating to the audience it is intended to communicate to. It's fairly clear in its meaning as writing. How bad would it have to get for the LLMs to start properly insulting it? LLMs are trained as assistants, so probably pretty bad.

If an LLM praises your text, that is a good sign on net, though it may still contain issues that are not LLM-legible.

Part of the fantasy of a public debate is that once you get that dumbass up there you can humiliate him out of lying.

Problem is, in order for that to work, he either has to have a sense of shame, or you have to ensure favorable social terrain. Otherwise he'll just keep lying and dare you to correct him, turning it into a "he said, she said."

If he had a sense of shame you wouldn't need to drag him up there. If he figures out you have favorable social terrain, he won't agree.

So you usually won't get him unless he has misjudged the context and thinks he'll get an advantage when he won't.

Me: Stop dodging and think about this issue at a higher level of abstraction.

Progressive: No! No! You can't make me!

Note that you can find guys like this in every coalition, but the figurative clothing they wear is different. Like, BrazenAutomaton tried to bait an antisemite into giving an actual argument, but it didn't go anywhere. Instead of faux-reasonable responses based on faux critical thinking, like the progressive I'm describing, the responses were more focused on a dumb "based" approach.

In the past two weeks, I've had three different people all attempt reframing attacks on me. One of them attempted like 3 or 4 different reframings during the course of the argument, each from a different angle, and I tagged and busted through each one.

It was fascinating, because I was able to see the skill with which each twitter user made the attempt, and see how it compared to the overall aptitude indicated by their arguments.

This is one of the big differences between real Twitter users and LLMs - a real twitter user will cut their losses and run, where if you've tried an LLM, it will answer questions that any human would quit before answering.

calling something e.g. reddit is (whatever else it may be) an instance of the rhetorical technique where you describe your target in terms of having bad qualities and not in terms of lacking good ones because establishing what you think are good qualities leaves them as attack surface while reducing your maneuverability (because you can't set forth foo-nature as a good and then ignore foolessness without appearing unprincipled, you have to answer for what foo actually looks like in the field, and you have to answer for what provisioning foo entails).

which means the counter-maneuver is to name the unnamed and identify that to which they gesture only via negation.

I've reblogged this post previously with different commentary, but I was thinking about it again recently.

Particularly, I was thinking about the concept of discursive maneuverability, which is some pretty abstract social combat theory.

I can feel the outline of another metaphor drawn from conventional warfare, but I can't quite pin it down.

In a real war, most of the methods of attack are highly mobile. Infantry with guns, tanks, towed artillery, aircraft, ships with missiles, etc. But of course, the facilities that actually manufacture munitions for these things are generally static - and on top of this, an army exists to defend some sort of society, which is full of static production facilities.

One way to view this is that real society has to make some sort of positive investment - it must select from the field of all possible options for building a farm, and then build that farm in order to have food, and any real farm will have trade-offs. Thus in this sense, someone who takes on no risk by making no positive investments is essentially free-riding.

But the thing that I can't quite pin down is this intuitive sense that there is an epistemic equivalent of a factory (or production facility) that's being abandoned by choosing mobility and minimization of attack surface. Something to do with producing knowledge.

I'm not sure if that intuition's just noise. Maybe it'll come to me later.

You are offered a pill. If you take it, and then date a man, break up with him, and write a break-up song about it, you will gradually shift towards having the appearance, vitality, musical talent, and financial acumen of Taylor Swift. If you are over the age of 35, your biological age will reduce by one year per year until you are effectively age 35. You cannot marry any man about whom you have written a breakup song, ever, no matter how perfect he may be.

Once this process starts, you must date, break up with, and write a break up song at least once every 3 years, or elements of your original identity will be slowly erased - photographs and records will disappear, friends will forget they ever knew you, and projects you completed or art you created will be attributed to someone else or vanish, for instance.

The one offering you this pill appears to be Taylor Swift herself. It is unknown whether or not you can transfer or sell the pill, or what will happen if the pill is in your possession and you do not take it.

[ irradiate-space ]

The former. If you're under 35 though, you age normally and then stop aging at 35.

As far you can tell, the pill won't protect you from more conventional injuries such as falls or impacts, or from infections.

That's a substantially more interesting offer, but it makes the question-as-asked make less sense to me!

Conditional on taking the pill, your original identity might disappear, if you're not able to find a man to date, break up with, and then write a break-up song about, at least once every three years.

But even if that happens, you are still a rich, successful, immortal Taylor Swift impersonator.

I think that the people for whom "become Taylor Swift" is an attractive proposition will not care about losing their former identity, so who is this poll really for? Normally these sort of poll memes seem to have clear dilemmas, but as qntm's influential essay "To Destroy The Earth" makes clear, immortality already risks and requires transcending your former identity:

Is the poll about

Agreed that this is a classic weaponization hypothetical; not only can I obtain immortality, I can if I so choose fully erase my previous identity and forge a new one, one I have substantial agency over - after all, TSwift does not stay static in her own life, nothing here implies I will somehow get mindfucked into releasing remixes of Fearless for the rest of my life if I don't maintain the breakup cycle. So for anyone who doesn't particularly like themselves this is stacked upsides.

Also gotta be honest but if you are poly the breakup costs are a rounding error, one breakup every three years? Bro I could just date in the Bay Area Greater Polycule for 6 months and bank enough breakups for the next century all while staying married.

I will note however OP is planting an unyielding flag in the ground on the "is TSwift bisexual" debate with that "date a man" standard, seems restrictive. For some orientations that would be the real cost.

Does Taylor Swift's sexuality actually matter for this pill?

Can someone meaningfully "break up" with someone that they were never attracted to? Can a Strictly Woman-Attracted TSwift still generate valid breakup energy by dating and breaking up with a series of flunkies? Could this successful businesswift secure her former life's past by subscribing to one of the various Rent-a-Familymember services?

Obviously with full legalism it doesn't, but lets not try to bury the implication; the idea here is that you "become" Taylor Swift on both ends, both from what the bill does and what you need to do to maintain what the pill does. The 'boon' and 'price' are both intended to be the essence of Taylor Swift. And the fact that said price is entirely restricted to men is suggesting a lot about said essence, even if OP didn't intend to.

As for the second question, as a straight guy who has dated and had a very dramatic breakup with a lesbian, my experience is that life will always find a way. I agree that it must be a real breakup; one can not Aella-pill your way in Tswiftdom (you can ofc Aella-pill your way into Aella-Girldom, a very different but perhaps equally dramatic fate).

The person offering you the Swift Pill, who appears to be Taylor Swift, offers the following answers:

The breakups must be at least as bad as Taylor Swift's breakups, so that they're sufficient to write a breakup song about.

Someone who is not attracted to men will gain sufficient attraction for the purpose of dating men in order to break up with them so that breakup songs will be written.

The pill alters itself slightly depending on the subject - a straight man or lesbian can start the process by breaking up with and then writing a breakup song about a woman. This starts the three-year clock before the next breakup must occur.

Good info! You will be making a big gamble on how much artifice you think TSwift brings to her relationships; its not uncommon to believe that she dates for clout and song-writing source material. To the extent that that is true the "as bad as her breakups" is a minor price, just clocking in work hours. To the extent that it is false you are committing to a decent % of your immortality being spent in emotional pain. Still worth obvi, but a higher price.

[ northshorewave ]

[ tiffany-loves-broadway ]

I spent a good 30 minutes in total laughing out loud at the various responses. At one point, I was laughing so hard I was crying.

[ centrally-unplanned ]

A good reading.

The Moon Prison/Cyro Prison/Cyber Prison poll was intended to test a hypothesis about Tumblr users' moral intuitions, partly by functional decomposition of what prisons do into three options, each with different trade-offs.

Part of how people answer that sort of question is in how much context they want to bring in. We as a society have a special frame for "Spherical Cow in a Vacuum" moral dilemmas such as the Trolley Problem, which are designed to remove context so as to heighten the prominence of a particular moral consideration so that it can be examined.

The Moon/Cryo/Cyber Prison problem has a political dimension, in that even if one prison type might be best if implemented by an "ideal" government, in practice, prison implementation is subject to political considerations and organizational design limitations.

A good way to answer this kind of question is with a conditional fork.

Something like:

It's more than the two bits for picking one of the three, but if these problems are intended to sharpen thinking, or just for fun, then that's perfectly fine.

We can do the same thing with the Trolley Problem, this time with three prongs:

For that last bit, an opponent may attempt to get someone to agree to pulling the lever, and then attempt to get them to agree to go along with something else by framing it as a trolley problem and using the agreement as a point of leverage. However, we almost never have the amount of certainty about any other moral problem as we do about the Trolley Problem, and a lot of moral rules are about managing uncertainty. (Some actions are effectively morally unreachable because there's a base rate of other human beings having gone insane, and it isn't possible to establish a level of certainty high enough to clear that possibility.)

The TSwift Pill post is not a carefully designed exercise to functionally decompose the moral nature of Taylor Swift into several options, it's just a fun little joke.

Part of the joke is that it's implicitly framed as a moral dilemma, but it doesn't seem to require any of the trade-offs normally required by such dilemmas, beyond being Taylor Swift.

The other part of the joke...

...is the implication of unknown unknowns that may indicate that this problem may not be a 'moral dilemma,' but a 'fae bargain.'

Being Taylor Swift is incredibly valuable. If you can take a pill to become Taylor Swift, why haven't more people done so already? If you already are Taylor Swift, why would you give someone else a pill to become Taylor Swift?

Why does the specification contain mechanics that would mostly be relevant to quitting being Taylor Swift and returning to your past life if, as-written, there is no method to stop being Taylor Swift once the process starts?

Obviously, one answer is just that one's past life is being used as a hostage by the spell, burning the bridges and thereby forcing one to become Taylor Swift no matter what. You're either Taylor Swift or you're Taylor Swift. There is no other option. (This is just the first loop more aggressively - the benefit is that you're Taylor Swift, and the price is that you're Taylor Swift.)

Another answer, particularly if the offer is made at the beginning of a story, is that there is some unexplored procedure to stop being Taylor Swift. In this case, because there is only one known public Taylor Swift, there probably are no more than a handful of Taylor Swifts active at any time, which suggests that the pill-offerer may be trying to quit being Taylor Swift and dump the effects off on the potential pill-taker.

Why dump the effects? Well, maybe it's not as beneficial as it seems.

From a narrative perspective, it's likely that the pill-taker would receive the offer early in the story, take the pill, start becoming Taylor Swift, and then either find out the benefits don't outweigh the costs by as much as expected, and then either... (a) attempt to end the curse [linear], (b) become "addicted" in some way and meet a bad end [linear], or (c) pass the "curse" off onto someone else [cycle].

It's like vampirism, only instead of involving a tougher ethical dilemma where you must kill or weaken others to survive, you're just Taylor Swift. The payoff is high and the ethical price is silly and small compared to other offers, making it absurd.

For narrative purposes, however, the original description of the pill's effects should be relatively accurate, so that the reader doesn't feel cheated.

To be honest I don't think there's much opportunity to really raise the readers' level of understanding with such a story, so it would be something that I'd attribute to a Clint Manstock book.

calling something e.g. reddit is (whatever else it may be) an instance of the rhetorical technique where you describe your target in terms of having bad qualities and not in terms of lacking good ones because establishing what you think are good qualities leaves them as attack surface while reducing your maneuverability (because you can't set forth foo-nature as a good and then ignore foolessness without appearing unprincipled, you have to answer for what foo actually looks like in the field, and you have to answer for what provisioning foo entails).

which means the counter-maneuver is to name the unnamed and identify that to which they gesture only via negation.

People who are regular readers of this blog will note my habit of walking around and functionally decomposing things into soft game theory.

Soft game theory is not a replacement for moral character or personal judgment. It's not a replacement for empathy or emotions.

Human beings have a tremendous degrees of freedom. Soft game theory is never a complete description of the motivation for their actions.

But a reason I walk around breaking things down into component parts and incentives is that there are a lot of people who think we are doing things for "no reason," so they can just change things to whatever they want.

The underlying soft game theory is part of the substance of various approaches, from which some of the power of their styles is derived. Negative vs positive rights remains a good example, where even though negative rights aren't absolute, they derive some power from the alternative being war. Positive rights are benefits subject to scarcity, so it doesn't make sense to use the same word to describe them.

OK so the US has abt 50k gun deaths a year, and the US has about 400 million guns. So that's about one in every 8000 guns resulting in a death a year, if we assume multiple homicides with the same gun are negligible (about half of gun deaths are suicides). Not sure if that number feels high or low but sometimes it's just fun to divide numbers :)

actuarial value of a human life is $7.5 million (https://en.wikipedia.org/wiki/Value_of_life). So if each gun has a 1 in 8,000 chance of killing someone each year, a functioning system of gun insurance would charge about $1,000 per gun per year. That doesn’t include any costs from people injured by guns. That would add up fast for the people who own multiple guns.

so you mean if i shoot myself my family gets 7.5 million?

No, it’s not a weregeld system. But it’s a useful a way of looking at what each gun roughly costs society in terms of additional deaths. “The motivation for placing a monetary value on life is to enable policy and regulatory analysts to allocate the limited supply of resources, infrastructure, labor, and tax revenue. Estimates for the value of a life are used to compare the life-saving and risk-reduction benefits of new policies, regulations, and projects against a variety of other factors,[2] often using a cost-benefit analysis.”

Using that kind of analysis you could say that each gun is responsible for roughly $1,000 per gun per year worth of gun deaths. If each gun imposes $1,000+ of costs on society, it would be economically efficient to force gun owners to internalize that cost. Given the difficulties of deciding who deserves compensation in suicides, etc I should have said something like economically efficient gun tax instead of gun insurance.

I think if it were useful it would be a weregeld system, otherwise the utility is mostly just pulling out a baseline for what the gun tax should be that appeals to, idk, spreadsheet fuckers, people with approximately one testicle and fractional children, people of that nature, unlikely to pay the tax but happy with accusing people of fractional murder and fining them up front

You are correct that I am 100% a spreadsheet fucker who owns zero guns and will never pay the tax. But right now, the gun tax is set at zero. Gun owners do not internalize the costs of gun ownership and we get a suboptimal number of guns in society because of it. And in general it’s better to put a price on something and let people make their own choices about whether they are willing to pay that price instead of banning something outright.

I think thinking of it as “accusing people of fractional murder” is an emotionally inflammatory way of thinking about a pretty common way of looking at the costs and benefits of policies. I’m in favor of a carbon tax. I think that price of carbon should include the cost of premature deaths caused by global warming. I don’t think someone who buys a gallon of gas to get to their job is a factional murderer because their emissions contribute to climate change, but forcing them to internalize the costs that their actions have on society is good for everyone.

A real insurance company would likely divide a real insurance policy for the use of a gun in murder based on a number of risk factors, such as age, sex, income, line of employment, past employment history, past history of drug abuse, history of mental illness, past criminal history, and geographic location (down to the neighborhood level).

Guns would be bundled into a shared policy per gun owner, as there is not a linear correlation between the number of guns owned by one person and the likelihood of committing a crime.

Most conventional gun murders likely would be committed by people who did not have gun insurance. The additional criminal charge for possession of an uninsured gun would be useless, as murder is already illegal, and already carries a long sentence.

If gun owners were legally forced to purchase insurance against their own risk of committing murder by gun, from a marketplace of sufficiently-capitalized insurance companies, the actuaries, who calculate insurance risk premiums, would never dream of putting all gun owners into a single risk pool.

This is a fake policy.

say what you will about insurance but insurance agents know how many balls they have

To expand on this, for the benefit of @apollo-pigeon,

There are conversational poses and there are associated methodologies.

The spreadsheet pose is, "I am cold and calculating. I have run the numbers, and this is the cold, hard reality that you must either face now, or face at greater difficulty later. You do not disagree with me - you disagree with reality as it is."

The thing about the spreadsheet approach is that to make it work, you have to be willing to consider even things that you find distasteful or unfavorable.

When an actuary carefully develops tables to forecast life expectancy given the acceptable risk factors to consider, he cannot escape learning about the relative probability and likely timing of his own death.

The pose derives its power or reputation from the associated methodology, not the other way around. To use the pose without the method is to joke.

The idea that each and every person has an exactly equal risk of committing crime, and that there are no statistical factors that can be used to estimate it beforehand, is absurd. Some people clearly have worse impulse control than others, and on a consistent basis. Drug addicts are merely one of the most obvious and non-controversial examples.

Pretending that everyone is the same until a criminal conviction proves otherwise is a means that liberals use to prevent people from jumping to conclusions early, because people (including left-wing people) have a tendency to do that.

You aren't supposed to invert it. It's a polite fiction used to smooth the operation of society. (You can think of it as inserting forgiveness into a tit-for-tat strategy to cover for the known possibility of error, among other motivations liberals have for this strategy.)

This brings us to a second pose and method pair.

A kind, empathetic, and charitable society can be viewed as a sort of insurance. Everyone runs a risk of becoming physically or mentally weak or disordered. Someone who is not kind, empathetic, or charitable still receives the benefits of a forgiving society, and can thus be viewed as a kind of free rider.

Additionally, someone who lacks empathy may be dangerous or harmful, as if they're only good out of a fear of punishment, they may do harmful things to others or impose extra costs on society when they think they can get away with it.

The empathy pose, "Don't you care about this suffering person at all?" is based on alleging that the target doesn't care about the suffering of others and is thus either a free rider or somewhat dangerous.

It's common for the spreadsheet pose to go wrong because the poser didn't actually do the research, either not bothering to run the calculations, or leaving out important but politically-sensitive or painful information.

A common way for the empathy pose to go wrong is that the empathy is very selective, which makes it troublesome by the substantive basis for the empathy pose itself. For instance, having empathy for criminals but not for victims of crime is a common failure.

(These considerations are overall similar to negative rights ideology having an implicit basis in the soft game theory of conflict, which puts some limits on what could count as a negative right, while positive rights ideology attempts to borrow the prestige of negative rights by sharing the term "rights.")