All of Decius's Comments + Replies

Late information regarding mobility devices:

In the US, federal regulatio defines a wheelchair as “A wheelchair is a manually operated or power-driven device designed primarily for use by an individual with a mobility disability for the main purpose of indoor, or of both indoor and outdoor, locomotion.” and also defines “other power-driven mobility device” (OPDMD), including approximately anything powered that helps someone with a mobility disability that isn’t technically a wheelchair.

There isn’t actually a whole lot of difference: Wheelchairs cannot be pr... (read more)

Long-term information: the title of this article is how I remembered it to reference it five years later regarding a discussion of disability.

If I parse things right, the initial state is something like 1/3 “I’m Luigi” 1/3 “I’m bowser” and 1/3 “I’m waluigi”, and the RLHF eliminates the bowser belief while having no effect on the other beliefs.

If this is based on the narrative prediction trained off of a large number of narrative characters with opposing traits, do all the related jailbreaking methods utterly fail when used on an AI that was trained on a source set that doesn’t include fictional plot lines line that?

My general reply is “if you think you’re spending too much time at the airport now, try missing a connecting flight”.

Different airports vary greatly in how much it sucks to unexpectedly spend the night in the terminal.

3[DEACTIVATED] Duncan Sabien2y
(TBC I have never missed a flight "on purpose" in this way, by just saying "eh, I'd rather spend less time at the airport.")

My experience with that behavior has been: 1: have a desired outcome in mind. 2: consider the largest visible difference between that outcome and the currently expected one 3: propose a change that is expected to alter the world in a way that results in that difference no longer being visible 4: if there are still glaring visible differences between the expected future and the desired one, iterate until there are no visible differences.

For example, people who see homeless encampments in public parks wish that they were not reminded of income inequality. Th... (read more)

It seems like the core issue underlying all of these specific examples is that “gather more information about the expected outcome and seek additional options” choice isn’t considered.

Sometimes the price gouging actually is someone who is making an obscene profit even considering their expenses. Price-fixing in that specific case can just be the socially desired outcome, but the policy maker has to have detailed information about the specifics of that specific case.

So far the idea that an embryo will become immortal if it exits the womb alive has been take... (read more)

Are there better tools for figuring out how much you value things?

There are alternative methods for figuring out how much you value things. Marie Kondo recommends holding an object in your hands and asking "does this thing spark joy?"

Is that also a reason to oppose every other advance?

I'm not saying I oppose making advances here, I just want us to think a bit more carefully about what kind of advances are good and bad.

Why are you characterizing "contraindicated cancer screening" as "healthcare"? In either case, it's not central to the issue where rural specialists have two-month waits for appointments and four-hour waits from the appointment time.

You said "More healtchare isn't always better".

Can you give a central example about a situation where more people receiving healthcare is worse, and why we should characterize that situation as one where more people receive healthcare?

If the government restricts the supply of meat (and food generally is adequately distributed), then the finite supply of meat makes it positional, and the fact that all needs are being met (within the scope of the example, at least) makes the outcome satisfactory.

If you thought that I intended some element of "maximize production even if all needs have already been met", then we have completely failed to communicate.

If you do cancer screeing for a person, you might find things that would go away on their own. A lot of women lost part of their breasts that way. After analysis of the issue the Obama administration restricted the amount of breast cancer screeings because the additional breast cancer screeings were not believed to reduce mortality by breast cancer.  If you reduce the number of surgeries that surgery salesmen can perform then the surgery salesmen don't try to sell surgeries to everyone but focus on selling the surgeries to those who profit from them.  Denmark cut  a third of their hospital beds from 2000 to 2018. The fact that every hospital bed finds it's patient whether or not there's a real need for treatment is healthcare economics 101 and you find plenty of public policy that's controls the supply of healthcare services to prevent overconsumption.

You're speaking of meat being a positional good, not a need. If luxury food or cosmetic surgery are easily obtained by everyone who wants them, then they stop being even positional goods.

Making more things universal is a goal that goes beyond the scope of providing basic needs, although the mechanism is rather similar. But it would be inhuman to eliminate the idea of positional goods and status entirely.

It makes no difference whatsoever why the scarcity is created- incompetence, malice, and apathy are all causes of waste. Logistical failures are no more tolerable than intentional genocide of equal total deaths.

Cosmetic surgery has little to do with the arguments I made about healthcare. While you can argue meat being a positional good in some countries, middle-class people in the US consume a lot of it without seeing it as a positional good. The US still overcomsumes meat likely both from a health perspective and an ethical one.  China on the other hand decided that they don't want their population to have as unhealthy diets and thus moved to reduce meat consumptions.  China has more scarcity of meat because the government believes that eating as much meat as the US Americans do is unhealthy and thus bad for China.  If you want to create change it matters a lot why the world is the way it is. 

The number of people killed by scarcity is the measure of scarcity. At least until there's enough that nobody is being killed by it. There's no point in trying thought discussion about whether scarcity changes nature after we stop killing people with it, not until we're a lot closer to that.

This actually wasn't obvious to me from the OP, fwiw.

Food being scarce only because so much is destroyed instead of distributed. While people are starving to death for a lack of enough food at all, it isn't "meat" that is scarce, it is "food".

For a particular person, more medical attention might be harmful, but there's no shortage of examples of cases where people are not getting enough medical care because they can't. Sometimes because they can't afford to, sometimes because doctors simply literally refuse to perform certain procedures.

There was a time where we Europeans thought that the surplus food we produce can easily gifted to Africans. As a result we destroyed a few local food markets. After that mechanism good public criticism we now. You might argue that it's totally worth destroying their local food markets if it means that we Europeans don't destroy food but it's not a straightforward argument.  Distributing food is an issue of being good at logistics. Even within Africa plenty of food gets destroyed because people don't do the logistics well enough for it to reach the people in need of the food.  There are plenty of poor people who have enough food to not starve but who want to eat meat but can't (or only can a few times per month) because it's too expensive.  I'm very curious what you mean with "scarce" if meat isn't.  This basically boils down to "there are examples where more medical care helps and there are examples were less medical care helps". It's no argument against the point I made.

No, and you failed to comprehend what I was saying as soon as you said "For a given amount of scarcity".

Also, the fewer people die, the less scarcity there was. Pretty much linearly.

So you mean as little scarcity as possible? At what point does the number of affected people enter into it?

If the prompt was supposed to be examples of good explanations of puns, I'm sure that we can't agree on what a good explanation of puns looks like. But it appears to treat pun jokes and regular jokes equally. And it understands how to make formulaic jokes, but it's impossible for me to tell if it made any adequate ones or just copied them.

Most of the questions of law will have to be decided by a judge- not just 'ruled on', actually decided by, unless the legislators clear up the uncertainty.

I don't see any place in Somerville code that defines "family" at all, so many things are insufficiently specified (is 'family' even reflexive?).

The existence of forgetting is necessary, given recall speed that scales with amount of memories, but the haphazard nature of remembering and also of unbidden recall is not.

A garbage collector that periodically deallocated memory addresses based solely on when they had been accessed, or a CPU cache that randomly fetched memory from addresses associated with system instability, would be horrible design choices compared to the ones that were made for computer memory; biological memory lacks design choices.

Transhumanist 'perfect memory' isn't "perfect recall of everything", it's "ability to chose what to recall" combined with "offsite backup capability".

I understand that there's certainly an information-theoretical security flaw, but if there is an attacker who could gain net value by seeing your mouse activity, you should be in a secure facility that prevents eavesdropping and none of the computers allowed in that area should be allowed to have bluetooth trancievers.

If a given dongle can be spoofed into providing arbitrary HID input (or just arbitrary keystrokes, in addition to mouse movement and clicks), that would be a more serious vulnerability.

Dongles of bluetooth keyboards certainly can input arbitrary keystrokes. That's already enough to do basically anything on the computer. For example the tab character can be used to switch between different UI elements and exploits are usually carried out in code and not by manually navigating through files or windows.

There's also an element of "past performance is not a guarantee of future results". It's possible that someone correctly confidently predicted one thing for exactly the right reasons, and then confidently makes an error in the next thing for almost exactly the right reasons.

Likely, even, because the people who are confident about hard questions are more likely to be overconfident than have superpowers.

"Undevelopable" does not mean "utterly without use". An area that can't be paved over and built up because it would cause watershed damage might still be usable for grazing cattle. A city block surrounded by blocks that have variances from the building height code is worth less, not worthless.

Suffice it to say that there are epicycles that negate the specific problems that you were pointing at. They almost certainly invoke problems that I'm not capable of identifying.

Bidding on tax bonds would set equally efficient prices on them, since it would be the insurance companies' people bidding on them.

Treasury bonds pay back their principal; deducted improvements would not be added to basis price at the time of sale.

I don't know all these words.

That's entirely incompatible with the idea of selling the tax liability separately.

Right. That's a silly and unnecessary idea. Your liability is someone else's asset. In this case, the government's. There's absolutely no way that the asset holder (be it the tax authority, or a lender) should or would let you turn their secured debt (you pay or we take the asset) into an unsecured one (you pay or we ... ask again?). You can't generally transfer liability and you certainly can't transfer or discharge FUTURE TAX liability.

If the insurers go insolvent, does the government default?

I would suggest a system where the government sells bonds linked to specific future tax revenues, and allow a property owner to buy the bond for their own property many years in advance, essentially prepaying the taxes at the current bond rate. If property values and taxes, or even just collections, crash, the bond holders take the loss.

If I were designing the system, the final fallback would STILL be that the tax is secured by the asset. Insurance is just a side-contract, and if the insurer doesn't pay, you sue them. If they still don't pay, then the owner pays. If nobody pays, the government takes the land. It'll make a nice park, if they can't auction it off for enough :) This way, the tax rules are simple, and any insurance or smoothing deals are optional ways to make it more predictable for owners (and more costly overall).
I guess I'd trust competition between private insurers to set better prices than a government agency. I don't know much about how insurers are regulated, but I think this system works alright?

Should undevelopable land adjacent to a formerly rural developed area pay taxes as though it were developed, while giving the developed adjacent land two tax breaks?

If the land is undevelopable, it doesn't really matter who does what with it. If the tax exceeds the value anyone can get out of it, it will default to the government (who will always buy land at $0). The government may not be a great land manager, but there's nothing to be done with this land anyway. If there's rural land nearby that is developable, maybe the land is actually a bit more valuable than the way it is currently being used, so it's not such a problem if the property tax is higher.
  • If someone has a lot of wealth stored in land that has been significantly improved, they could bear a larger tax burden than they would be paying under an unimproved land value tax.

I'm not sure why this is considered a weakness, other than the effect is has on incentives; it is easier to offset that effect on incentives than it is to eliminate it.

For example, tax land at its improved value, and allow the cost of any improvements to be deducted against up to half the tax on the land for the entity that made the improvement.

Improved value can be easily... (read more)

I think I disagree with the principle that it's easier to offset tax incentives than to eliminate them. But I wouldn't take the other 100% of the time either. If the tax is ever above twice the rate of returns on US Treasury bonds, no property-tax-payer will ever buy US treasury bonds. Instead, they'd buy a huge chunk of marble for a counter-top and hide it under wood. Then they deduct half the cost of the marble, and the rate of return on this risk-free investment is half the value of the property tax. Plus, if you can only deduct half, then improvements are dis-incentivized.

Yes, other orders which are conditional on the price can also result in the market clearing price being undefined.

How do you determine what price to trade at? Is it the market clearing price with lowest/highest volume?

I don't see how you've explained how batching transactions is positive-sum; would it reduce transaction costs? Would it somehow provide net benefit to both buyers and sellers as compared to executing trades as they can be, rather than randomly benefiting some buyers at identical cost to the sellers?

Would the magic surplus be great... (read more)

We'd have to look into the specific rules here. The only thing I do know is limit orders trade before market orders do and orders are time stamped. In other words, the current institutional structure is more about managing order flow than establishing the clearing price when supply and demand are considered in the larger context -- such as buyers and sellers who want to participate today but will be sending the orders at much different times during the period the market is open. Draw a simply Supply and Demand picture, and assume those curves do represent the true price quantities of the market participants that exist during the given period. The intersections would then be the standard economic model clearing prices. Let those curves represent one day. Currently, under the order flow regime, the when ever the orders come in at attempt to find a matching order on the other side occurs. But that would allow a supplier that has a price in the supply curve that is above the intersection to be pairs with buyer on the demand curve that is at or above the ask price. The difference between the transaction price and that implied market clearing price when considering the days actual supply and demand characteristics seems to represent the loss of value the buyer would have saved if they could have bought at the theoretical market clearing price, which is only known when the day closes. This also represents a transfer to the seller with the above market price. This seller, on some pretty standard market logic, really should not have been able to get that trade as the price asked was too high. Changing the rule about matching trades (supply and demand) at the end of the day rather than as the orders come in prevents that type of inefficient pairing of buyer and seller. To my knowledge that is what happens in the morning for all trades, and why sometimes when unusual events occur the opening for a security is delayed while the market or market makers try to figure out jus

If the end-of-day price is above my bid or below my ask, the stock couldn't trade, and the transaction never happens. For that reason there is a moderate incentive not to trade FCOJ futures right before the crop report is revealed, because your offers will have to stand but other parties can wait until after the news is out in order to meet them, and the future actors have an insurmountable information advantage.

It also results in some stop-loss order situations being metaconsistent; the trade price is below the stop-loss trigger point IFF the stop-losses trigger, resulting in multiple market clearing prices.

I don't understand either of these points. Stop-loss is merely a special form of limit price and would certainly fit into a end of day clearing/settlement process for supply and demand. Here I don't see where the multiple clearing prices can emerge due to stop-loss orders but not from other limit orders. And, even if there is a spread between the limit bid/offers that might produce a gap, so indeterminate price in that range, it would be rare where the market price orders were insufficient to close the gap. The fact that the end of day price might be above or below the price you will trade at is not different if it's end of day only or tick-by-tick. If no one was willing to sell at your bid or buy at your offer you are not part of the market in that period of time. That is now we think markets should work. The concern here seems related to my next paragraph. One thing that has motivated my musing here is the market argument about maximizing surplus generated by markets -- the area between the supply and demand curves. End of day trading allows those two curves to be known and match the trades that allows that maximum surplus to emerge. Random trades as they show up throughout the day does not. Some of the trades my well be pure transfers generating zero surplus. I do think it's fair to ask is the lost of surplus is meaningful and I'm not 100% sure. But if not then there is a lot of welfare economics that needs to be rethought and a certain amount of market philosophy to reconsider. Is a society just as well of under the standard free market outcome with producer and consumer surplus as it is under a regime of perfect price discrimination (i.e., only producer surplus)?

Sometimes the market is wrong but the barrier to competition is so high that if you try to take advantage of it you run of money.

In my experience, cafeterias are more likely trying to be inoffensive to everybody than to be trying to reduce costs, and in any case their costs per person are lower than the per-person costs of preparing comparable food at most, if not all, qualities of food.

The line between cafeteria and buffet restaurant isn't perfectly defined, and I think that there are already businesses that operate on a cash-per-eater basis that are substantially high-end cafeterias. Shifting to a pay-per-month basis for them seems plausible, but I'm not sure if they ge... (read more)

Daily seems arbitrary, at least as compared to hourly and weekly settling. Hourly settling would seem to provide the same fairness to small investors while still allowing the market to transact during the entire day, instead of only once per day.

If responding to new information during the day is not desired, then less frequent settling is indicated.

Well pricing can be assessed though out the day, and reassessed as well via the submitted bids and offers. The change his is about the actual price the security trade at. In this approach everyone selling or buying on a given day gets the same price (very much like the standard economic auctioneer model). While I would be more interested in hearing what thoughts others might have, one though might be that such an approach eliminates a bunch of little welfare loss triangles that are the result of the individual trades occurring throughout the day by requiring the day's supply and demand curves to be fully reveled.

Human architecture means that the best humans do not always lose to Alphago, even with their massively inferior computational resources.

Many meals have prep work which scales extremely well, but are incompatible with small sizes.

A way to take advantage of such a property would be to have a central facility produce lots of high-quality meals for lots of people, serving them hot and fresh at the appropriate time.

Having invented the cafeteria/mess hall model of dining, the problem is in implementing such a result over a large enough scale to be viable. That scale is going to be roughly the number of people that will eat a batch of bread in a sitting while it is still hot and fresh, where the... (read more)

When I worked at a summer camp with a reputation for good food, one of the meals they would make involved two baguettes per table, still warm from the oven, and it was super popular. Cafeterias seem like they would be a good candidate for this with a lot of people and predictable demand, though they're often trying to feed people as cheaply as possible, and with a captive audience they don't have much incentive to improve. Fancy tech office cafeterias seem like they could do this if they wanted to?

Yes, and to fit the lawsuit against the manufacturer of a faulty microwave or the journal retracting a flawed paper into GEM requires adding epicycles.

I think you've correctly identified that lots of people believe themselves when they say lots of wrong things, and did a fair job of explaining one part of the ways to reduce that error.

A different alternative worth considering is summarized by "Say enough wrong things that you know for sure that most of them are wrong." In practical terms that means developing three or more mutually contradictory plausible-sounding reasons for what and why. From there, instead of sharing them in a manner where politeness requires that they not be challenged... (read more)

I believe so, but I lack the requisite domain-specific knowledge to extract it, or even to evaluate those reasons once they have been extracted.

The thing about the size of the federal government is that there was a team of people with domain-specific knowledge integrating and responsive to public comments and suggestions from people with and without domain-specific knowledge. Their records *ARE* available, if you can figure out what to ask for.

The general summary is probably fairly accurate, but it would be a major error to think that the actual policy w... (read more)

No, and yes.

For mousetraps, it has the implicit problems with ignoring the effects of marketing on customer decision.

For academic theories, it has the implicit problem where the *consumers* (the scientific community, which is presented with multiple theories and forms a consensus around zero or more of them) is being misstated as the producer.

You can get paid for seeing an error in someone's mousetrap design and making a design that lacks that error, even if the error is "bad marketing". You can't get paid for seeing an error in a mous... (read more)

Is someone has a theory that's badly marketed in academia, you can earn in prestige for appropriating the theory and pushing it with better marketing/framing.

Yes. That is a limitation of the model.

No, it's mot *my* model.

But within the model, marketing is one of the factors that determines the quality of a mousetrap.

Phrase your predictions in the manner such that if you said "I bet [statement] at the odds implied by N% certainty", there would be more money bet against you than money offered at better odds.

The odds that there's some serious side effect that isn't extinction-level are many orders of magnitude higher, and the approval system was made in advance with the full knowledge and careful consideration of the potential of epidemics.

3Timothy Underwood4y
Based on my priors about how groups like the FDA and CDC work that seems unlikely to be true. My strong impression is that the system is designed to minimize the odds of things going wrong in a way that will generate headlines based on errors of commission. Is there any source showing how the decided on this trial duration as something that would balance the risks involved with a deadly epidemic?

Yes, GEM does not permit there to be money on the ground. It's one of the limitations of the model.

Which is a great reason to not use it to model situations where there is money on the ground.

2Matt Goldenberg4y
Mhmmm, I think we're both in agreement that GEM is a good model neither for new mousetraps nor new scientific discoveries... The interesting question for me is: Does the model break down in the same way for both?

There's no way to profit off of people preferring to buy mousetraps that are lest effective at trapping mice because, in that framing, the best mousetrap is *not* the one that is most effective at trapping mice.

You also can't profit off of people buying fake 0-day exploits, unless you're selling the best 0-days.

Your tautologically efficient market is not what economists or anyone else in the world is talking about when they talk about a market being efficient. In the trivial sense you're posing the problem you would describe your market as satisfying people's needs even if people had parasitic worms in their brains that tricked them into thinking that mouse traps cured cancer. In the real world, people buy mousetraps because they *expect* those mouse traps to be more valuable than the price, for some definition of value that does not include "they bought the mouse trap". A pillar of one such value system might be "effectiveness at catching mice". In this scenario, people might be mistaken - they could have bounded rationality, they could have exploitable cognitive biases abused by advertisers, etc. The thing that makes financial markets resilient against this sort of problem is the nice property that rational actors are incentivized to take advantage of other peoples' missteps. Mouse trap buyers cannot do this - you can't buy a hundred mouse traps from an underrated dealer to correct the market, and you can't short a mouse trap dealer's traps because they are running deceptive advertising.

Naive algorithmic anything-optimization will not make those subtle trade-offs. Metric maximization run on humans is already a major failure point of large businesses, and the best an AI that uses metrics can do is draw awareness to the fact that the metrics that don't start out bad become bad over time.

e.g. microprocessor research, psychology as applied to advertising.

Within the constraints of GEM, a 'better mousetrap' is one that customers agree is better.

The best theory isn't the one that most accurately predicts reality, it's the one that the academics agree is best.

Just as situations change and the best mousetrap for securing a clay granary is not the best mousetrap for a nursery, the best theory is not constant over time- but when a new theory becomes best, it is coincident with it becoming the academic standard.

That they agree to be better before they actually use it. Plenty of customers buy microwaves with a lot more programs then they end up using. It's one of many examples of Norman's Design of Everyday Things were companies produce devices that have poor useability. If a microwave is built faulty in a way were it radiates the room around it and causes health damage, that's a bad microwave but customers might not know about the problem. The microwave that radiates it's customers is similar to the flawed scientific theory that gets believed because an academic field lacks the mathematical abilities to see the theory as flawed. Buying an academic theory by spending attention and reputation on it is just like buying a product with money. The buying decision depends in both cases on the qualities of the product that the recipients can perceive.
8Matt Goldenberg4y
This is similar to saying "Within the constraints of GEM, it only counts as a $10 bill lying on the ground if people know it's there. " The EMH, in theory, applies to the value people get from a thing, whether or not they know the value. It may be that there's a mousetrap that makes it less likely other mice would come, and saves me money in the long run. If I understood and believed that, I would buy the mousetrap no question, even though it's more expensive. However, there's a cost to educating me on top of the cost of just building the mousetrap. The EMH has all sorts of weaknesses and holes, but the biggest weaknesses are around "innovation" -- both finding new innovations, and propagating them through the marketplace. See the issues of asymmetric information, the innovator's dilemma, etc. In other words, I think @johnwentsworth was being too optimistic in his assessment: innovations of any kind will run into this issue, whether they have a market attached or not.
Maybe we could remove some confusion by calling it "a more addictive mousetrap" instead? :D
The problem with the mousetrap market, even under your framing of the problem, is that people are merely incentivized not to make incorrect decisions; there's no way to profit off of other peoples' poor choice of moustrap. Mousetraps are to financial markets as metaculus is to predictit. A better example would be the market for zero-day exploits; there not only are people incentivized to find exploitable bugs, but also to search under rocks other people are leaving unturned. Academia is simply that much worse than the mousetrap market, because in that case the real "consumers" are the engineers/randos who read and use the scientific literature to inform decisions, and scientists are poorly aligned with their needs.

There's a middle ground between having an organization be profitable, and an organization optimizing for profitability.

Well said. And this middle ground is exactly what I am worried about losing as companies add more AI to their operations -- human managers can and do make many subtle choices that trade profit against other values, but naive algorithmic profit maximization will not. This is why my research is on metrics that may help align commercial AI to pro-social outcomes.

I'll go one step further: Anything that can be a for-profit loop already is.

> a product to certain beneficiaries, whether clean water for poor rural areas, or a non-rival good such as research for the world at large.

If those rural areas would be able to purchase clean water, or a way of producing their own clean water, in a manner profitable to investors in the rural water service area, they would.

Where the world at large pays for research results, those fields are privately funded.

I think “already is” is correct, except where there are barriers: legal, technological, cultural, etc. Remove the barriers (change the law, invent new technology, etc.) and you could open up opportunities for profit. Not sure what you mean by this, examples?
Load More