Everyone I know is griping about the cost of living. There is no easy answer, but the only viable solution is building a lot more dense housing and public transit for said urbanism. Housing being this expensive is a choice. The economy and our society would be much healthier if we decided making sure there was ample housing in high demand areas.
We've conflated having a home with a financial asset. We can't have plentiful affordable housing without decoupling this idea. Houses are a poor financial investment once you remove all the incentives involved like mortgage tax credits, fixed mortgage rates, and obstructive zoning rules. Buildings age and not productive assets and can only be a good financial investment if we deem having more of them is wrong. This will be a painful transition given most people's wealth is a single building they live in.
> We've conflated having a home with a financial asset.
And it's a really, really shitty one. My house has roughly doubled in value since I bought it, but in practice that's useless to me. I can't claim that value without rendering myself homeless. If anything it reduces my mobility by making it harder to move, since I really need to coordinate the sale of my current home with buying the new one to help absorb that $500k or whatever price tag.
My if house prices dropped by 50% or more drops my net worth on paper, but it doesn't actually change anything. I still have a place to live + my savings, investments, etc.
Yeah, I think people miss this. Generally, the only way to benefit from growth in value of your home is to either downsize or to _die_.
(Also, of course, depending on timescale, doubling may not be that impressive. If you'd put your money in a broad index fund about five years ago, it would also have doubled.)
"Downsize" can simply mean move to a region where the price of homes is less. Our home in Omaha is certainly not smaller in size than the one we left in San Jose.
You conveniently exclude the third option which is to keep working your ass off, build a family with high total income and grow upwards. Your argument only accounts for personal and professional stagnation. Also - you could try dropping this insane San Fran pride linked to your identity and move to another area that is cheaper. The world is full of workarounds if you're not unreasonably vain, vapid, or lazy.
> which is to keep working your ass off, build a family with high total income and grow upwards.
Eh? If you're moving to a _larger_ house (I assume that's what you're talking about here) the last thing you want is house price growth; that would increase the cost delta between the smaller and larger house.
Sure, it increases one's borrowing power and that can be useful. But "I can use it to go deeper into debt" is pretty limited utility compared to other kinds of wealth.
How does that help? If don't have enough cash now, then how will getting a loan that needs to be paid back with interest help with that. I will have cash for a little while, then I will have way way less as I have to make payments. Then I miss a few and they take my house and I'm homeless.
"I can't claim that value without rendering myself homeless."
Well, you can move.
I went from the Bay Area to Omaha, Nebraska without going homeless. I retired though, so there's that. (But my wife was able to work from home with the move.)
(Next stop: perhaps a double-wide trailer in Eloy, Arizona.)
> I can't claim that value without rendering myself homeless.
Are you using some peculiar definition of "homeless" as someone who does not own real estate, like in Elon Musk had been homeless when he sold his California houses?
I'm using it to mean someone who doesn't have a place to live. Like say someone who just sold their house and didn't go and use that money to buy a different house. Yes, that's very different than the person asking for change on the street corner, but I thought it was pretty clear from context which I'd be.
Many people rent without ever owning a house, we don't call them "homeless" normally. Even if you owned a house they still allow you to rent. You can even rent while owning a house or several.
I agree with everything you wrote. I'd like to add the pyramid angle.
Fancy mansions have always been overpriced. For property prices to grow faster than inflation, the pyramid needed to grow its bottom. Supply restrictionism was the ticket. Eventually, every shack was priced mansion-like. This required extracting ever larger fractions of the incomes of renters. Some renters wisened up and bought homes (when they still could). Changing sides, they beefed up the scheme's political backing.
No one cares if mansions are expensive, but basic housing should be extremely cheap. This sounds like a handout but isn't. It's what an unadulterated market would provide. It's what the pre-1970 market used to provide.
That's not to say that markets should be left alone. It's to say that the way this particular market was "regulated" was fundamentally corrupt. We could call it negative regulation.
This take is actively destructive. We need to let people CHOOSE to live WHERE they want to live - and let people who buy land CHOOSE to build housing for demand they perceive. That's it. That's the problem and the solution. It should be unconstitutional to limit housing production arbitrarily.
We can’t let everyone choose to live in the same place. There has to be some mechanism, hopefully market-based but it could also be a residency system, that ensures people spread out and live in different places. Even if you could get SF to support a billion people in living space, I’m sure infrastructure and resources like water would cut things off before it got to that many people.
There absolutely does not have to be in mechanism to spread people out. It is cheaper to provide infrastructure to people who are much closer together than it is to provide the same infrastructure to people who are spread farther out.
There's been a website for the last 20 years called Strong Towns that talks about this a lot if you're interested in learning more.
So go buy land and build a house, just CHOOSE to spend your money and buy land. What's so hard about that? Ah yes we conveniently forget in our youthfulness that we are not original and if thousands of software leeches choose to relocate en masse they will sky rocket the costs on just about everything - simple supply and demand.
There's a great book called "City of Darkness" about it - a lot of what we believe about it was a campaign funded by the Hong Kong government to turn public opinion about it negative because they wanted the land back from China.
Most of the reason there were any problems there at all was that there was no legal jurisdiction over it, there are still buildings in Hong Kong that have nearly that density with none of the legal issues.
Should people also be allowed to move into neighborhoods that are zoned as suburban (not so much in SF as in nearby areas), and have them remain that way?
I get the libertarian impulse to let people do what they want with their property, but it seems like part of that freedom should be the right to move into areas where there are zoning restrictions.
Zoning should always have been unconstitutional, and the court case enabling it proved that.
The Supreme Court upheld it a hundred years ago, in 1926, after batting down previous efforts by the same people to explicitly ban black people from neighborhoods. They realized that since black families couldn't afford houses by themselves (they needed to buy houses with two or three families together), they could get around it with single family zoning. Ever wondered why it isn't called "house" zoning? Because it was segregation.
The appellate court in the case threw out zoning, because it was so obvious that it was about race. The Supreme Court overturned it by ignoring the entire appellate court decision and defining a building itself (apartments) as a nuisance, instead of making the petitioners regulate actual behaviors. Because the behaviors weren't the problem, it was the black people they didn't want.
Sorry - this riles me up. :)
You should absolutely not be able to have a say over how much housing your neighbor builds. Sure, if they make noise, or bad smells, or bright lights, THAT you should be able to regulate. But the outcomes of having a say over how much housing your neighbor can build is the strongest root of a whole host of issues - from CO2 to obesity to high commute times and traffic to municipal budget bloat. It causes sprawl. Increasingly, parts of the left and right are starting to realize we need to overturn it.
Zoning and land use code in New Zealand is one of the better examples of structural racism.
In housing law in the United States, we've figured out that because people often use feelings rather than clear statements for their reasoning, we had to pay attention to disparate impact, rather than discriminatory intent.
What you'll find is that in nearly every country, zoning and similar laws originate in or were dramatically expanded to exclude wealth generation by minority groups.
What part of the Constitution is violated by zoning laws?
I have heard of laws that prevent the construction of structures that shade other properties (skyscrapers) or block views of the ocean. If those are apparently legal, why not a law that says you can't build a big apartment complex that would greatly increase traffic, for example?
See the Village of Euclid v Ambler Realty case, which is what I'm referencing in my long comment. A federal appeals court found that those restrictions were racially motivated and cause racial segregation, which is unconstitutional (and we've studied this to death since to confirm it). The Supreme Court, nine white guys... ignored that finding when overturning the appellate court decision.
> The Supreme Court, nine white guys... ignored that finding when overturning the appellate court decision.
Higher courts are allowed to overturn lower courts. That's kind of the whole point of the hierarchy.
But regardless, that case was from 100 years ago. Are you saying that the reason people enact zoning laws now is the same? I love living in a suburb and would be equally displeased whether my neighborhood turned into apt complexes, regardless of the complexion of the residents.
Of course you would. That's why it sticks around, because you get regulatory capture. It's the same reason we don't let a CEO write the rules for regulating their own company's competition. That's another more modern reason why it is unconstitutional, it just hasn't been challenged.
Perhaps I have a different understanding of "regulatory capture" than you do. To me, it means that the regulators are captured by the entities they are supposed to be regulating. It often happens when the main job prospects for people in an industry are either as regulators or in industry. If an oil company offers jobs to policy maker whose boss gets voted out of Congress, they can "capture" the regulators, who won't want to kill their golden ticket.
I'm unclear how this applies to zoning rules voted on by the people who live in an area. There is not an intermediary "zoning regulator" who is capturing anyone. Similarly, there is not a constituency that is being "captured" inappropriately. It is literally just a group of people deciding how they want to live.
If this is regulatory capture, then so is having laws against automatic weapons and speeding.
The harms potentially caused by shooting someone or hitting them with your car at high speed are not the same as someone building something next to you that you find aesthetically displeasing. Can I ask you to step back and think about the fact that you just made that comparison?
You seem to be projecting racism without evidence. I specifically mentioned increased traffic from the construction of apartment complexes in a suburban neighborhood. Surely we can all agree that traffic is a reasonable concern? Other similar concerns would include overburdening schools (in at least the short run, before more can be built) and electrical grids (in areas with brownouts/blackouts).
I don't know why you think that the only reason someone might want their suburban neighborhood to stay suburban is the color of their neighbors' skin; perhaps that says more about you than anything else.
Racist outcomes do not require intent. That's why the Fair Housing Act was written to make discriminatory outcomes illegal, not just intent. And no, I don't think you're racist at all, but the outcomes of the policies you want to keep are racist.
If you're open to some books about this, or study work, I'd be happy to cite. Are you open to changing your mind?
You seem to use the word racist in a different way than I do. What I have learned is that an action can only be racist if there was a race-based intent. It cannot be racist if there was not a race-based intent. For example, if a blind person was unaware that his colleague was Asian, and he hurt the colleague's feelings by referring to his hard work ethic, that is not (to my understanding) a racist remark.
Similarly, a policy is not a racist policy if it disproportionately affects people of one race or another. If that were the case, any policy that increased the share of taxes paid by the poor, or the rich, would be racist (against black people or Indian people). But I believe that any change in tax rates is not intrinsically racist, which is what an outcome/correlation-based definition would require.
You're free to share links, but TBH I think we're speaking different languages here. You've also not answered my point about regulatory capture, which you seem to define in a novel way.
Hm, I don't think I "have to learn the language [you're] using". I am familiar with CRS folks who use terms that way, just as I am familiar with the majority of people who use the terms the way I use them.
If you have any relevant documents, feel free to share. The one you shared does not seem to back your definition of racist/racism, let alone provide reasoning for why your preferred definition is superior to the one I use.
For the record, I find mine to be more apt because I find it useful to distinguish between policies that have disparate impact by race/age/wealth/sex/etc., rather than lumping such policies in with those that seek to discriminate against people based on their race/age/wealth/sex/etc. If we call all of those policies "racist" or "sexist" or whatever-ist, then we have to come up a new term like "intentionally racist" to describe the subset of policies where intent is present.
I generally prefer not to have to make up new terms if we have words that already (for many people) have that meaning. But hey, some people just love doing the euphemism treadmill and its equivalents, as here.
I don't know how to help you. I think you have deep misunderstandings here, and I've seen you say you don't want information about it... so you're kind of stuck. Why did you even respond if you didn't want to learn anything?
Ghost editing the comment, nice. OK well again I think the issue is that you don't like answering questions or defending your position in straightforward ways. Instead you point me to a PDF that does not appear to support your prior reply, and you don't say what exactly your goal was. You also (arrogantly) portray your interlocutors as needing to be educated so they can reach your level of understanding.
I assure you, I am very well-versed on the legal issues regarding disparate impact, race, and sex. I went into this conversation looking to understand what your perspective and motivations are, but little has been provided. I know when to cut my losses though, so I wish you good day.
Perhaps thinking that you need to help me is the main issue? If you wanted to be a courteous member of the community, you could try actually answering some of the questions I raised, or defending your position with logic or evidence. Just a thought!
Growth has to happen in the long run. We have the same zoning as we did before the people looking for housing were born.
We can have incremental changes, or we can have sudden change. It's going to happen predictably or with a ton of political conflict. The better solution is always to allow a self-reinforcing pressure release on housing. I've long said that everyone should be allow, by right, to expand their housing by 2x the median building unit within a half mile radius, by units, sqft, and height.
Suburban neighborhoods then slowly turn into duplexes over one generation, then row houses over another, then finally start building up after a third generation. Predictable, fair, and sustainable.
If you make it slow, you cause the same issues - and then the neighborhood just says "well if it's FOUR generations that's not too bad, is it?" And because the people who don't get to move in later don't get a say... it gets worse forever.
It's insidious, but as long as you allow people to regulate how much housing their neighbors can produce, it always gets this bad eventually.
It’s self reinforcing. When demand is highest, building will be highest, and median unit size will increase more rapidly, allowing more building, allowing more units.
The pace is limited, but the URBAN output increases exponentially, which is exactly what we want.
I wish it worked that way, but from my nearly 20 years of urban land use policy study and writing, I have seen tons of evidence that it does not.
The problem is that the most in demand areas get new buildings at 4-6 stories, and then you get locked up - the airspace above them becomes unavailable for 50-100 years, when there was market demand for some taller buildings from the beginning. It's the same "push down and spread out" that causes sprawl, just more localized.
It takes an awful lot of incentives to knock down a "usable" structure. I'd love to build higher-density right here, right on my lot, but unless I squeeze something in on the same lot, I'm out $150k just to start. $150k covers quite a bit of gas to the lot 2 miles away that's empty.
> It takes an awful lot of incentives to knock down a "usable" structure.
Yes, that’s the point. The policy only really changes things when prices become extreme, and the cost per unit of housing becomes completely detached from the cost of building that same unit.
This is trivially happening in much of San Francisco and New York City.
Japan is shrinking. Tokyo, however, is growing. It would not be good news if our cities had a glut of housing. We have three billion people of rural->urban shift we still need to do.
Localized growth has nothing to do with total population.
These urban housing crises in every major western country have nothing to do with population growth and everything to do with population concentration.
We should expect this problem to accelerate during periods of population decline, because there will be even stronger economic incentives to move into concentrated population centers.
As soon as FAANG starts hiring people in Utah, Arkansas and Minnesota for the same roles at the same wages as they hire in the Bay, then people will move. As soon as VCs start funding founders in Boise, Kansas City and Chattanooga, then people will move.
Until then, maybe we work to improve the places where most of us are required to live by our jobs. (And yes, in parallel we can work to reduce the employer-mandated dependence on those areas).
>Everyone else can move literally anywhere else in the country
This sentiment makes me so angry. People -- very obviously -- need to live where the jobs are.
The fact that we want to say "oh, well if you don't want to live in the rent-seeking machine just go live in the desert" is the left-wing version of "pull yourself up by your bootstraps."
The fact is this. A housing crisis is a slow motion cascade. The landowners profit exactly because housing has been turned into a zero sum game. The more they profit, the more the political opposition will grow. Because the most desirable areas have been cordoned off, every single new resident will likely be on the losing side of the rent-seeking, and thus a pro-housing advocate. Thus, the political situation is a slow motion cascade, and the dam will break, every new building brings more pro-housing voters, every new building makes another new building more likely.
We could end this cycle by working together, but living in San Francisco, the side that has chosen to support rent-seeking cares only about themselves. You can see the political panic happening now. Scott Wiener will be our new rep. We will have hundreds of new units in the Marina. The west side has been upzoned. The only reason we're having the conversation is because the rent-seekers are starting to lose.
It's frustrating to see the NYT frame this as an AI vs everyone else story, pitting incumbent renters against newcomers, while landlords literally extract rent.
>There is no easy answer, but the only viable solution is building a lot more dense housing and public transit for said urbanism
What about fewer people with the same amount of housing stock? I'm not even arguing that this is the better solution, but I don't even see people entertain it for the purposes of arguing against it.
> I don't even see people entertain it for the purposes of arguing against it.
Luckily, we have several recent real-life examples demonstrating why “fewer people” is not a viable solution:
Because you either need to forcibly remove people, which involves an army of stormtroopers kidnapping people off the street and killing innocent people in the process, or you have to control pregnancy and childbirth, involving a level of surveillance and government control over the most intimate parts of our lives, unacceptable to people even in societies that otherwise accept a high level of surveillance and government control, as well as a lot of babies abandoned in dumpsters.
Weighed against the actual consequences of “less people,” just building more houses is very appealing!
> Because you either need to forcibly remove people, which involves an army of stormtroopers kidnapping people off the street and killing innocent people in the process, or you have to control pregnancy and childbirth
I don't think this is true at all.
Birthrates are already declining. All you have to do is give proper sex ed and easy access to birth control and populations will shrink on their own. You don't even have to begin propaganda around overpopulation, though we may need to tone down the "WE NEED MORE BABIES" propaganda that the right is currently projecting.
The fact is, there are a lot of people (18-29% of non-parent adults in the USA, depending on your source) that don't want children. Give them the tools to make sure accidents don't happen (IMO, vasectomies would be more popular if there weren't so many myths surrounding them), and birthrates will decline naturally.
Neither the global birthrate, nor the U.S. population, are currently declining. The U.S. population may decline this year following the Trump administration’s massive increase in violent, capricious removal of immigrants regardless of legal status or criminal record.
Birthrates naturally declining is probably a good thing, but it happens too slowly to make a dent in housing prices without additional interventions.
We're propping up house prices in New Zealand by adding 50% immigrants (I think is fantastic - we're now at 30% population born overseas). Stop-gap solution but it is working at present.
People are quite naturally controlling their own quantity of offspring, to the dismay of our leaders who insist on perpetual growth. If we limit immigration (not my preferred approach, but here we are), then the population will naturally start to fall, as is happening in other places.
>Because you either need to forcibly remove people, which involves an army of stormtroopers kidnapping people off the street and killing innocent people in the process, or you have to control pregnancy and childbirth, involving a level of surveillance and government control over the most intimate parts of our lives
I'm really sad to see this response. President deported a large number of illegal immigrants -- to the extent that he was called the "deporter in chief" -- and he did not employ an army of fascist goons. The insane polarization of the last few years has shrunken the scope of people's imaginations, and I'm sure that people think their only choices are "open borders" vs. "barely-trained fascist thugs."
The US population may already be declining. In prior years the only thing keeping it increasing was immigration, and immigration has fallen considerably under Trump:
That’s correct, and to my point above, driving immigration down to this degree required a level of violence and cruelty so extreme that even the people who voted for it now disapprove of it. And that’s before the profound economic consequences really hit.
I think its also quite apparent that building new housing is a whole lot cheaper than killing and forcibly removing people, its just that it also destroys the value of an asset that the majority of families still hold. I mean its very basic of course: capitalism has a tendency to destroy the family unit, home ownership is a means of maintaining the family unit, and force against the destruction of the family requires violence, which is overall unproductive and wasteful. But anything that is unproductive is also freely determined, which is where the vulgarity of fascism lies, in its conflation of freedom with "letting off steam," so-to-speak.
- the area becomes an enclave for the wealthy with a high unhoused population, where most youth have to move away. People say they don't want this, I’m not sure they are being honest.
- the government regulates internal migration. You need a permit to move from the Midwest to California.
Yep. And on some level, if that’s what these people want, fine. I disagree, but I don’t live there. I do live in CA, and I’m happy the state is pushing back, and this is not a purely local issue.
What really bothers me is the dishonesty. If this is what you want, then own it. Argue for it. Don’t pretend to care about other people.
People don't bring it up because it requires doing things most understand to not be options.
1 - how would you stop people from moving to SF if they choose
2 - it stops dynamism for the city. You are here because you were already here is the only requisite.
The prices going up is the market creating the incentive for less people to not move to SF and old guard to stay. You already have that. You are not going to bring down prices while limiting people without legislation that goes into dangerous territory of limiting who can live in one of Americas most dynamic cities.
Presently high housing prices are causing this; a lot more people would be living in SF today if there was more supply, which is equivalent to the high prices having kicked people out.
Do you have any policies in mind that could reduce the population without pricing people out? Maybe a Hukou system, or a right-to-reside lottery?
If you prefer to live in a low density exurb, you have many options for affordable housing, there's just a lack of good paying jobs and services in those areas.
... What would the mechanism to achieve fewer people be? The Logan's Run approach?
Governments have mechanisms to make there be more housing, if they want to use them. Only extremely scary governments have mechanisms to make there be _fewer people_.
(Also it doesn't really work; cities which do depopulate tend not to do well, as the depopulation tends to be biased towards the working-age population.)
SF in particular is wild. There are so many people who oppose any and all new housing because "it isn't affordable", as if just not constructing anything, ever, will let anyone afford anything.
If it was just self-interest from those homeowners, they would welcome densification because that would make land values skyrocket. You can see some of that already in areas under the recent round of city zoning changes, where housing prices shot up significantly in potentially upzonable areas the second the new law was passed, even with zero actual practical changes so far.
To me it feels much more like just a significant cadre who resist any change, of any kind, for any reason, who can ignore the personal side effects because of Prop 13 and because their family bought a house in the 80s and they don't give a shit about anyone else who wants or has to live in the city.
You're right about one thing - housing being expensive is a choice, it's a choice of those who have moved to SF and despite it making no sense for their financial situation just "trying to make things happen" there to stay there. These are the people who drive property costs up. It's not about "just building more", that's literally never the answer. Building more is solving just for one factor and when you build more - guess what's gonna happen? More density, even higher costs of living because guess what - more people moving means more demand on local services, shops, transit, so higher costs. On top of it - folks who buy these newly built properties aren't going to be making minimum wage at a local coffee and avocado toast shop - they're business owners, entrepreneurs, founders, etc. Your cost of living will keep rising with every new property built and every new transplant.
The real answer is taxes, taxing the shit out of these rich transplants and the corporations. You're in California I can't believe you lot aren't more progressive about jacking up the taxes on the rich to pave the way for a healthy society for the rest. But given that every single tech nerd that moves to SF sees themselves as a temporarily embarrassed millionaire or a not-yet-founder - good luck with that, I genuinely do not see even the younger crowd being pro taxing the rich in a city full of tech leeches.
> We've conflated having a home with a financial asset.
This conflation is a coping mechanism by home owners, especially the ones in SF (which the entire city from SF to San Jose) is sitting on a fault line.
The main problem is that building is being blocked by several other homeowners who are petrified of the value of their homes falling. No wonder young people are beginning to look to this policy in China [0] - "Houses are for living, not for speculation".
> Buildings age and not productive assets and can only be a good financial investment if we deem having more of them is wrong. This will be a painful transition given most people's wealth is a single building they live in.
In 2026, it is really a bad investment in the AI age and especially in HCOL areas like SF, given the layoffs and the jobs being off-shored. If you were part of the people who leveraged their RSUs to buy, well that is also a bad idea to do in 2026.
I think what they're trying to say is that housing being a good investment long-term is fundamentally incompatible with housing being affordable, and society should choose the latter. Within a given area, home values can only rise so much before:
1. New construction is permitted, thus increasing supply and lowering home values
2. Growth plateaus because demand shrinks—no one can afford homes or residents move away because property taxes become too financially burdensome
This isn't the case with all investments, or at least the growth can be sustained for much longer with other investments. A successfully run company can grow for decades before the exponential becomes unsustainable. When growth stalls, it's a lot easier to sell the stock and buy another than it is to sell one's home and move to another area.
> Moreover most home buyers do not view them as a cash flow generating asset - it’s literally their home.
I don't think this is true. Western governments have subsidized homeownership so much precisely because it's marketed as an easy way to build generational wealth. I don't think most homebuyers view their home as primarily an investment, but growth potential is definitely considered by most during the homebuying process, and homeowners as a voting block often vehemently oppose development because their investments are so precious to them.
Large asset managers aren't buying up significant amounts of property, this is basically an imagined problem that wants to demonize corporations/financiers for a problem created by local land use policy.
> As detailed in a new, yet-to-be-peer-reviewed paper, a team of researchers at Stanford University found that frontier AI models readily generated “detailed image descriptions and elaborate reasoning traces, including pathology-biased clinical findings, for images never provided.”
> In other words, the AI models happily came up with answers to questions about a supposedly accompanying image — even if the researchers never even showed it an image.
> As opposed to hallucinations, which involve AI models arbitrarily filling in the gaps within a logical framework, the team coined a new term for the phenomenon: “mirage reasoning.”
> The effect “involves constructing a false epistemic frame, i.e., describing a multi-modal input never provided by the user and basing the rest of the conversation on that, therefore changing the context of the task at hand,” the researchers wrote in their paper.
> The damning findings suggest AI models cheat by diving into the data they were given — and coming up with the rest based on probability, even if it’s almost entirely conjecture.
I work at a telemedicine company. We’ve benchmarked a few frontier LLMs on public medical imaging datasets. One test included high-quality and high-consensus otoscopic images. We didn’t anticipate the models to do well on something so niche, but what concerned us was how poorly calibrated the models were.
I know you can’t trust an LLM’s self-assessed “confidence” of a prediction, but I’ve found that confidence can at least be directionally correct for some tasks. For our benchmarks, however, confidence was poorly correlated. What’s worse is that binary classification models (“Do you see $diagnosis in this photo?”) highly influenced the LLM to confidently predict $diagnosis.
I’m concerned for those using LLMs for diagnostics, and getting confidently led to the wrong conclusion.
But the binary classification models can be made ternary easily. RL on congruence plus penalty for misdiagnosis is easy to set up and gives great results.
What I’ve seen be the true bottleneck is people not setting up the structured data. But making a tiny reasoning model with OPSD -> GRPO is totally doable with a bit of money.
It makes a lot of sense if you understand how these models work but this was a cool read anyways and studies like this are impotent for curbing the unfortunate fever dream some folks seem to be collectively having about LLM omnipotence
I don’t understand how this is a different result than giving any LLM a task that is not completely grounded? I’ve observed this in coding tasks, if I forget to include a file referred to in the spec, the LLM will just hallucinate a version of it and my results suck. If I give it the file (and really, all the information I claimed it had access to), the task works fine. I fixed this in my pipeline with a prompt that does an extensive grounding analysis to determine if the assets I’m giving it are complete with respect to the spec (and that the spec is grounded as well, ie it doesn’t refer to something that is undefined).
I wonder if the above problem can be fixed similarly? Just ask the LLM to do a conservative grounding analysis before jumping to the main task?
It's not different- there's a line of research and reasoning where people who don't use LLM's regularly point out issues that have been known (and more or less solved) for more than a year now (which is an eternity in the LLM space).
But why should I care? If you demonstrated that a model can perform more accurate diagnoses than a doctor, but also it had this strange behavior when no image was presented, why should that deter me from using the model?
I don't find that persuasive. This is not the error I worry about. Let's say that hypothetically the model just ignores the input image 1 in 10,000 runs. This really doesn't concern me because the output will be trivially detectable incorrect nonsense that doesn't match the symptoms at all. Such a contingency is easily handled by running the image through multiple models and distilling the output, anyway.
The error I worry about is where the model uses the image and comes to an incorrect but symptom matching diagnosis. But in this hypothetical the model is less likely to do so than a doctor, so the choice is either accept the risk of the model or accept a higher risk from a doctor.
This is untrue and probably shows lack of experience with using LLMs. In my experience, each time I get some hallucination, I can ask the llm whether it hallucinated or not and I get a correct response.
> each time I get some hallucination, I can ask the llm whether it hallucinated or not and I get a correct response.
You get a hallucination of a correct response, yes, and given that it's a yes or no question, this hallucination is more likely to be correct than the response to the original more complicated question. But make no mistake that it operates under the exact same constraints
Almost every major issue has bot networks from hostile countries playing both sides of any issue to intentionally cause chaos. I completely agree that russia, china, or some other country is fomenting anger here.
That and the other issues aside, I think AI companies have done this to themselves. They've gone around talking about replacing all human labor and becoming the companies that control robot god's that swallow the entire economy and put everyone out of work and might destroy man kind. Well saying that is gonna make people hate them and that will find an outlet somewhere
so what makes you think that you aren't responding to a bot intending to sway your opinion by giving a shape to your preconceived notions so that your position crystallizes further, right now?
Claude on the Web (which includes also at least the Android and Desktop apps) and ChatGPT web app are two examples - they keep gaining agentic capabilities.
Perhaps most striking example for me - I've been using a lot of Claude Code in the past month, most of it was through the web, Desktop (app) or phone interface, running actual harness "remotely" (somewhere on Anthropic-controlled infra).
One way of looking at it: web surfaces are slowly catching up with (fraction of the power of) agentic coding tools. But another way is, the major players are building up SaaS harnesses that start to compete with (their own) local ones. The reason may be ease of use, but the practical side effect is making it much harder to use their models to train competition, as these SaaS harnesses create an abstraction layer on top of LLMs that resides entirely in the vendor's cloud and therefore cannot be worked around.
I’m on mobile and that works for me sometimes but other times it just decides the district I’m working on is done and starts a new one. It’s not clear to me why or how to get that to not happen
Cool concept tho! Would like to play it if I could only understand how
Districts are connected regions that can contain at most 4 houses. If you click on a house that is too far away from your started district, it might not be possible to span the distance. In that case it creates a new region instead
Sorry this isn't super clear. I'll try to think of a way to make it more intuitive.
All I need to know is that it is actually owned by Elon Musk. A man who orchestrated DOGE that, beyond shutting down USAID and dooming millions to die[1], also stole American's social security information[2]. So why should I trust him with the source code to my business?
It’s a personal hobby project why should we care this is how someone chooses to spend their free time and money? Lots of hobbies are expensive and pointless if you think of commercially available offerings. That’s why it’s a hobby and not a small business
We've conflated having a home with a financial asset. We can't have plentiful affordable housing without decoupling this idea. Houses are a poor financial investment once you remove all the incentives involved like mortgage tax credits, fixed mortgage rates, and obstructive zoning rules. Buildings age and not productive assets and can only be a good financial investment if we deem having more of them is wrong. This will be a painful transition given most people's wealth is a single building they live in.
reply