> I know not everyone wants this mental overhead, though.
I’m curious how to even do it. I have no idea how to choose which model to use in advance of a given task, regardless of the mental overhead.
And unless you can predict perfectly what you need, there’s going to be some overuse due to choosing the wrong model and having to redo some work with a better model, I assume?
You got some strangely negative reactions, but I agree; the article has not accounted for the safety net effect of a kinship society. It’s a glass half empty view, and there is a glass half full view too. The article is also not considering the country’s economics or the government or geopolitcal history, which others here are pointing out. It’s an interesting thought, but seems premature (and a bit sad) to jump to the conclusion that tight bonds are the cause of poverty, when there are clearly more forces involved.
Help me out, I don’t understand the scorched earth perspective. You want to eliminate the playing field even for the people playing fairly, just because there are some bad actors? Would destroying all SaaS actually cause the cheaters to sell used cars & life insurance?
Until AI isn’t trained on all open source code ever written, regardless of license, which I doubt will ever happen, isn’t SaaS-writing AI in some sense building a larger scale & more concentrated version of what you’re hoping to destroy?
Personally, I hope and want everyone selling used cars and life insurance to be honest and upstanding. some of them are.
Are more “controls” what is necessary here? The problem wasn’t plastic contamination, it was the presence of stearates. Distinguishing between stearates and microplastics sounds like a classification problem, not a control problem.
There is practically universal recognition among microplastics researchers that contamination is possible and that strong quality controls are needed, and to be transparent and reproducible, they have a habit of documenting their methodology. Many papers and discussions suggest avoiding all plastics as part of the methodology, e.g. “Do’s and don’ts of microplastic research: a comprehensive guide” https://www.oaepublish.com/articles/wecn.2023.61
Another thing to consider is that papers generally compare against baseline/control samples, and overestimating microplastics in baseline samples may lead to a lower ratio of reported microplastics in the test samples, not higher.
Many papers in this field are missing obvious controls, but you’re correct that controls alone are insufficient to solve this problem.
When you are taking measurements at the detection limit of any molecule that is widespread in the environment, you are going to have a difficult time of distinguishing signal from background. This requires sampling and replication and rigorous application of statistical inference.
> Another thing to consider is that papers generally compare against baseline/control samples,
Right, that’s what a control is.
> and overestimating microplastics in baseline samples may lead to a lower ratio of reported microplastics in the test samples, not higher.
There’s no such thing as “overestimating in baseline samples”, unless you’re just doing a different measurement entirely.
What you’re trying to say is that if there’s a chemical everywhere, the prevalence makes it harder to claim that small measurement differences in the “treatment” arm are significant. This is a feature, not a bug.
You’re still bringing up different issues than this article we are commenting on.
> There’s no such thing as “overestimating in baseline samples”
What do you mean? Contamination and mis-measurement of control samples is a thing that actually happens all the time, and invalidates experiments when discovered.
> What you’re trying to say is that if there’s a chemical everywhere, the prevalence makes it harder to claim that small measurement differences in the “treatment” arm are significant.
No. What I was trying to say is that if the control is either mis-measured, for example by accidentally counting stearates as microplastics, or contaminated, then the summary outcome may underestimate or understate the prevalence of microplastics in the test sample, even though the measurement over-estimated it.
> What do you mean? Contamination and mis-measurement of control samples is a thing that actually happens all the time, and invalidates experiments when discovered.
The entire point of a control is to test for that sort of contamination (or more generally, for malfunctions in the experimental workflow). In the case of a negative control, specifically, you're looking for an "positive" where one should not exist. If an experiment is set up such that you can obtain differential contamination in the controls but not the experimental arms, as you've described, then the entire experiment is invalid.
> What I was trying to say is that if the control is either mis-measured, for example by accidentally counting stearates as microplastics, or contaminated, then the summary outcome may underestimate or understate the prevalence of microplastics in the test sample, even though the measurement over-estimated it.
The control cannot be "mis-measured", any more or less than the other arms can be "mis-measured". You treat them identically, otherwise the control is not a control. Neither example you've given are exceptions: if the assay mistakes chemical B for chemical A, then it will also do so for the non-controls. If the experimental process contaminates the controls, it will also contaminate the non-controls.
What you're missing is that there's no absolute "correct" measurement -- yes, the control may itself be contaminated with something you don't even know about, thus "understating" the absolute measurement of whatever thing you're looking for, but the absolute measurement was never the goal. You're looking for between-group differences, nothing more.
Just to make it clearer, if I were going to run an extremely naïve experiment of this sort (i.e. detection of trace chemical contamination C via super-sensitive assay A) with any hope of validity, I'd want to do multiple replications of a dilution series, each with independent negative and positive controls. I'd then use something like ANOVA to look for significant deviations across the group means. This is like the "science 101" version of the experimental design. Any failure of any control means the experiment goes in the trash. Any "significant" result that doesn't follow the expected dilution series patterns, again, goes in the trash.
(This is, of course, after doing everything you can to mitigate for baseline levels of the contaminant in the lab environment, which is a process that itself probably requires multiple failed iterations of the experiment I just described.)
Most of the plastic contamination papers I have read are far, far from even that naïve baseline.
> The entire point of a control is to test for that sort of contamination
No, the point of a control is to give you a reference point that shares all the systemic biases and unknown unknowns, not to detect those biases. If you follow the same procedure on a known null and on your experiment and observe an effect, assuming you really did exactly the same thing except the studied intervention, you can subtract out the bias.
This one example of technical jargon diverging from colloquial or intuitive use, and it is the type of thing people who haven't had statistics or scientific process education often struggle with because they keep applying their colloquial intuitions.
You talk like you understand this on the rest of the comment so I'm confused by this framing, and the person you are replying to points out (in my reading ) that contamination of the control 1) does happen in practice (in the sense that there was an accidental intervention) and 2) if the gloves contaminated both the measurements and control the same way then the control is exactly serving it's purposes
You’re repeating several of my points in your own words, supporting them and not arguing with them, even though your language and emphasis suggests you think you are arguing.
> then the entire experiment is invalid
Isn’t that what I said? You even quoted me saying it. But I didn’t say anything about only control being contaminated or mis-measured, I think you’re assuming something I didn’t say. Validity is, of course, compromised if the control is compromised, regardless of what happens to the test samples.
> The control cannot be “mis-measured” […] yes, the control may itself be contaminated […]
So which is it? Isn’t the article we’re commenting on talking about the possibility of mis-measuring? Are you suggesting this article cannot possibly be an issue when measuring control samples? Why not?
Controls absolutely can be mis-measured or contaminated or both. It has been known to happen. It’s bad when this happens because it means the experiment has to be re-done.
> If the experimental process contaminates the controls, it will also contaminate the non-controls
Yes! This is exactly what I was implying, and is exactly how you might end up underestimating the relative presence of whatever you’re looking for in the test, if your classification procedure overestimates it.
> You’re looking for between-group differences
Yes! and this is why if, for example, you didn’t notice your control had stearates and you counted them as microplastics accidentally, and then reported that your test sample had 2x more microplastics than your control, you might have missed the fact that your test actually had 10x more microplastics, or that your control actually had none when you thought incorrectly that it had some.
This, of course, is not the only possible outcome, not the only way that the results might be distorted. But this is one possible outcome that the Michigan paper at hand is warning against, no?
> Most of the papers I have read are far, far from even that naïve baseline.
Short of it, or exceeding it? Based on earlier comments, I assume you mean they’re not meeting your standards. I don’t know what you’ve read, and my brief googling did not seem to support your claims here so far. Can you provide some references? It would be especially helpful if you showed recent/modern SOTA papers, work that is considered accurate, and is highly referenced.
I agree completely. My point is that documenting methodology is standard practice, as is strict quality control, in the microplastics literature. I don’t know what controls are missing according to GP, and we don’t yet have references here to back up that claim. By and large I think researchers are aware of the difficulties measuring this stuff, and doing everything they can to ensure valid science.
$200/mo is a lot, sure, but the shocking part of that comparison is your rent. I didn’t know $400/mo apartments still existed. For most people in the US and EU, $200 would be closer to 15%-20% of rent I think? My cell phone bill for my family is almost $200/mo.
Last year, at first, $200 seemed crazy. Now that I’m getting addicted to coding agents, not so much. Some companies are paying API rates for AI for employees, and it’s a lot more than $200/mo. It seems like funny money, and I’m not sure it’ll last.
As you've probably guessed, I don't live in the US, so the price are drastically different. I live in the EU. And for my case, I love in really small flat for some years, so the rent couldn't go up a lot.
> most people in the US and EU, $200 would be closer to 15%-20% of rent I think?
> the average rent is north of $1000/mo.
I really don't know where you get your number from, $1000/mo average is really wild to me. With this amount, you can rent a flat for a whole family in the heart of the city. Nobody of my more well-of friends have a rent this high.
Or maybe you have some capital city in mind like Paris or London?
> I really don’t know where you get your number from
I googled it. According to Google, London’s average rent is around €2,700, around 3x higher than the average. I assume the number of people living there and paying that much balances against the number of people like you living in smaller towns and rural areas who are paying lower rents.
But yes, rents have become very high everywhere. I live in a medium sized city in the US not anywhere near a coast, and most kids attending the local university are paying over $1000/mo for a 1-bedroom place. The primary way to get cheaper rent is to have flat-mates, try to get 3 or 4 people into a place that rents for, say, $2500/mo.
I was paying $2k/mo in San Francisco 25 years ago for a place that was maybe 90m^2, and since then rents have gone way up. Google says the average now is just under $4k/mo. In some nicer neighborhoods, some people pay $8k/mo for a single bedroom. This big-city rent in SF, LA, NY, Chicago, Miami, etc. balances against the small towns in the US where you can find a room for $500/mo, which is why the average is above $1k.
It is my belief that rent price scales with the leftover income people have after they've paid for other necessities. Ie if you're from a poorer country/area then things like milk and gasoline will cost a similar amount (maybe 2x difference), but rent will cost a lot less. As people in a country get richer they start paying a larger and larger share of their income as rent of various forms.
Even the US has places with cheap rent/housing. The downside is that there's no (well-paying) work nearby.
It’s true that average rent prices are regional and poorer areas have lower rents, but that doesn’t tend to make much difference in urban areas and large cities where the majority of people live now. Why do you feel that rent scales with disposable income? Economists generally say the opposite based on housing being a core necesessity; that people pay rent in proportion to their income, and only what’s left over the the disposable amount. That’s why we have the 30% rule, for example.
You’re technically correct, btw, rental housing is a market and is subject to market forces, meaning what people are willing to pay. I’m just not so sure about framing rent as being lower priority than other necessities. And rent prices have been increasing faster than other necessities, and faster than income, so that might be a confounding factor in your argument.
Still, my initial reaction above is due to the fact that in the US and in Europe in most large cities, the average rent is north of $1000/mo.
>Why do you feel that rent scales with disposable income?
Because I'm from a country where 30 years ago average income was $220/month. Today it's $2475/month.
A large portion of people live in the same houses and apartments now as people did back then. The housing hasn't changed, but today renting a 70 sqm apartment costs you $800/month - the same apartment that people in 1996 lived in with their $220 monthly wage.
The reason why I think that housing is "lower priority" when setting a price is because the sale/rent price for housing is more divorced from the "manufacturing cost" compared to other goods. This happens for a number of reasons:
1. Housing scales with money. Most people live well beyond the "minimum required" for housing. You could survive living in a tiny room with a shared bathroom, but most of us want more than that. Compare this to food - rich and poor people will drink a similar amount of milk. You can't really spend 100x on milk and actually get appreciable benefits from doing so. You can with housing. (Same goes for most other goods. A $3 million car is not 100x better than a $30k car, it's not even 10x better.)
2. Housing is non-fungible. You can't have two houses in the same location. Food, furniture, and electronics are fungible.
3. Housing doesn't depreciate with use compared to other goods. You drink milk and it's gone. You drive your car and it degrades. Your house degrades simply by existing - your use of it will degrade it a little, but living in it also means you do maintenance, like cleaning, that will help keep its value.
4. Because of the above, housing is an asset that people invest in. This is a bit circular, but it also means a lot of people don't want to see housing become cheaper.
>And rent prices have been increasing faster than other necessities, and faster than income, so that might be a confounding factor in your argument.
Because the cost to produce other necessities hasn't increased as quickly as incomes have increased. We have better technology and better economies-of-scale that has made the cost of other goods cheaper. Now people have more money left over to pay for rent, so they do.
Yes rent (like everything) does scale with inflation. All the absolute numbers you’re using make reasoning about this more difficult than necessary. It’s better to use percent of income, in order to get a sense for whether today’s rent is more or less expensive for the renter. That said, you gave an example from 1996 of a $220/mo avg. income and $220/mo rent, which doesn’t add up. Google tells me that as a percent of income, rent has increased over the last 30 years. You might be right; that might be in part enabled by the cost of some goods going down. But the price of food hasn’t gone down. Higher rents also might be in part changes in spending habits, so you’d need to show those haven’t changed.
I used the wrong word, btw. The word I meant to use was ‘discretionary’ income. I think that’s what you’re suggesting, that rent is discretionary? ‘Disposable’ income is what’s left after taxes, but ‘discretionary’ income is defined as being what’s left after paying for things like rent, transportation, food, and utilities (https://www.investopedia.com/terms/d/discretionaryincome.asp)
It’s true the amount one spends on rent is a bit elastic, and that is also true for most consumer goods. The problem with your argument is that rent is not optional, like most consumer goods are. There are very few things that one cannot go without paying for at all, and rent is one of those.
If I understand correctly, what you’re suggesting is the reverse of how most people including economists think of rent. This discussion does depend heavily on what “other goods” you’re actually talking about. Can you provide more concrete examples, and show that they really are getting less expensive over time? Is your hypothesis supported by Eurostat’s HICP or the World Bank’s data on inflation and consumer prices?
What do you mean about vendor lock-in? I haven’t yet seen any meaningful barriers to switching between different companies’ coding agents. Are you talking about AI market lock-in and not vendor-specific lock-in?
> these loss making AI companies will eventually need to recoup
This is true, and while AI spend continues to rise, I’m starting to think once the dust settles and the true costs emerge and stable profits are achieved, that it may be expensive enough that it’s a limiting force.
The book is fantastic, I’d recommend reading it one way or another. ;) Speaking personally, I lose some motivation to read a book after seeing the movie. But book-based movies of course rarely if ever live up to the book. I read first, so I can’t speak to the other way around, but I think I was looking forward to the movie a lot more than I would have if I hadn’t read the book. I also suspect I was more forgiving of the movie than if I’d seen it cold.
I’m curious how to even do it. I have no idea how to choose which model to use in advance of a given task, regardless of the mental overhead.
And unless you can predict perfectly what you need, there’s going to be some overuse due to choosing the wrong model and having to redo some work with a better model, I assume?
reply