I thought the smarts were required to do the material science, chip design, and factory automation involved in fabs. Once this is done, it’s just clean room and careful repetition. Overly bright people are potentially damaging for manufacturing as they might get bored thinking big thoughts. And how many chip designers are actually needed?
Cost of workers, cost of manufacturing, regulatory and governmental overhead…. These influence fans more than smarts.
From the outside, chip design looks a lot like writing software. So that's loosely equivalent to asking how many software engineers are actually needed.
Credible chance regulation and overhead influence software more than intellect does. I'd certainly believe that as a premise.
Some people who don't identify as "high IQ" might get offended by this article, but I think the primary argument is valid. Many of our brightest and most ambitious minds go into finance & consulting simply because it pays the most. If we want more innovation & progress in certain fields, the financial incentives need to be aligned. In the case of semiconductors, I'm guessing there needs to be some sort of government funding due to huge startup costs and marketplace dynamics.
I’ll say it. When Black Shoals won the Nobel prize and business people started seeing PDEs, the new physics and applied math grads they hired were five years away from restructuring mortgages into unsupportable securities. Very few were smart enough to predict the eventual problem. Maybe the only people who can justify the inflated compensation should be those who can prevent collapse?
Some of the products didn’t pan out, sure. But what about all the products that did? Having deep and complex capital markets opens up a lot of possibilities in terms of building huge systems.
At my alma mater (MIT), the bright and lazy people actually went into finance also. Not sure what it's like now, but 10 years ago, many hedge funds basically just gave IQ tests as their interview.
Ages ago I had a slightly sad conversation with a manager in high tech mechanical engineering about this. We/they wanted good engineers but not enough to compete with the city over. Most of the expertise was very in-house, difficult to hire experienced people from other companies who would know the right things, so training up and retaining graduates was important. Hopefully that was a carefully reasoned choice.
Employee compensation economics don't make much sense to me. There's some factor in favour of maximising headcount which argues for many people on low pay instead of few people on high pay.
It's also considered _better_ to have your long term engineers occasionally notice the gap between their pay and the open market, leave, and then be replaced by new ones who cost what the open market predicted.
The incentives are set up so it doesn't matter. They don't lose if the company loses, if the product is shit, or if the engineers go through hell.
How many people ever got fired for selling the future for a short term gain? (how many of the few who did, didn't just get another, better job because of more experience?)
I noticed this in fundamental physics research, which has been stripped bare of the best and brightest, as if a plague of locusts had gone through every university.
Pretty much everyone who does really well in physics ends up as a quant or something similar because the pay starts at 10x what you get as a researcher and goes up from there into the stratosphere.
The only people left behind are those that didn't-quite-make-it, or truly deeply love physics and simply don't want to get an extra zero on their pay-cheque.
And yet tenure track jobs are still fiercely competitive. Maybe all the folks applying to those jobs are just the dregs who couldn't cut it as a quant, but I doubt it. Point being, that despite the fact that you can make more money in industry, universities still have an oversupply of people (both professors and grad students) who are willing to work for peanuts.
Yeah but focus of repeating "high IQ" that article had heavily detracts from that point.
It's really not that different for any skill that is hard (whether it is by requiring "smart" people or just ten years of practice to perfect the craft) to learn and niche enough that people normally don't choose that career path
Semiconductors are a low margin business. There's a reason why it's not in the USA anymore, even though it started in the USA, and it's not the lack of smart people - the profits were competed away. Now the profits are in design - NVIDIA, AMD, Apple etc.
Edit: this is not to say that the USA should not develop its own semiconductor industry again - the issue there isn't profit, it's supply chain and security, prevention of backdooring or denial by China.
I don't know offhand. Google reckons 42% "net profit margin", but for an industry which leans hard on reinvesting to stay ahead of the competition 'profit' is quite flexible.
If you're bright enough to be useful where there's this shortage, you're bright enough to see how much those jobs suck.
"We need more suckers willing to do incredible hard things for little reward beyond the work of doing them. while the management and their consultants reap millions in ongoing profits for bringing a diluted, not-quite-good-enough version of that work to market"
edit to add: These jobs are often at places founded by "genius motivated by love", but now its the second generation or more on from that. the management wants that kind of success but they want it on their schedule (gotta make the quarterlies). they want it without the risk of failure and need for multiple attempts and extended effort, which made the original company work as much as the genius, but no one recalls that part now. they want meetings and reports and rigidly regulated routines more than they actually want the genius level output.
Higher ups from these companies complain about lack of "high IQ" talent yet don't want to budge on offering competitive salaries. I have a smart friend who worked for ASML as a contractor, and they would give him empty promises of bringing him on full time, as if he still needed to prove himself despite being over qualified.
Complete clownshow from these upper managers. If workers aren't entitled to a job, you're not entitled to high IQ talent, dumbasses!
Just to highlight Singapore government pays millions to high IQ talent to be government ministers. Singapore Prime Minister for example draw about 2-3mil USD range a year in salary and can code sudoku in c++. Many of the civil servants are scholars and have PhDs.
> More specifically, although not everyone in semiconductor manufacturing requires a PhD, pretty much everyone has to be of above average intelligence and many will need to be in the top echelons of IQ.
I'm not convinced high intelligence is a necessary condition to get into chip design. You just need to be willing to spend the time to finish advanced degrees in university and get trained. This is mostly taken up by foreign students because citizens can earn well becoming an SWE right after bachelor's.
He's not saying anything about chip design. His post is based on the assumption that to get the error rate low enough for chip manufacturing they require high iq line workers which are in short supply.
Even if we assume it is, the reason is as you said, lack of incentives to even get on that path. And their below average IQ manager making much more /s
IQs are normalized by country. For example, Canadians are culturally similar enough to take the US test (including some US-specific general knowledge questions and mathematical notation) but score higher on average due to a variety of social factors.
The score is then adjusted downwards so that the median Canadian IQ is set at 100.
>For example, Canadians are culturally similar enough to take the US test (including some US-specific general knowledge questions
Any IQ test with any sort of validity will not contain any country-specific questions. ("Regatta" became irrelevant as a "gotcha" for those questioning the validity of IQ tests so long ago that it's long since become an emeritus member of same, really.)
> and mathematical notation) but score higher on average due to a variety of social factors.
I am not aware of any evidence that Canadians score higher than Americans on IQ tests when normalized by audience.
Americans do quite well on PISA, for example, compared to their ethnic relatives <https://www.unz.com/isteve/the-new-2018-pisa-school-test-sco...> Asian Americans do better than Asians; whites do better than Europeans; Latinos do better than Latin Americans; and blacks do better than Africans.
Hispanics and especially blacks' scores drag the US average down. Both white and Asian Americans score higher than Canada (and white+Asian is essentially Canada's racial makeup), and higher than New Zealand, Australia, the Netherlands, Belgium, Germany, France, Norway, and the UK; Estonia is below US Asians but above white Americans. Norway is by far the wealthiest Nordic state but its average is only two points higher than the US national average, despite not having a demographic that is 13% of the population and scores 85 points below the white American average.
High iq is defined as some standard deviations above average. a definition can’t be a shortage. if you define something to be 2 in a 100, than you don’t get to say 2 in a 100 is a shortage.
There's a shortage of 2-in-a-hundreds that I can hire (because people like that don't hit the street all that often). There's even more of a shortage that I can hire at the cost I'm willing to pay, which is what I think the real (unstated) issue is.
But yeah, that's just like everything. There's a shortage of luxury vacation rentals that fit my budget, too. There's a shortage of fine steak at McDonald's prices.
If you want these people, you have to pay them the way they're accustomed to be paying (most of them are smart enough to know their value), and you have to give them something to do that they actually want to work on. Won't do one or both of those, and now you're complaining about a shortage? It's hard to be you, I guess, but nobody else cares.
(Note: "You", above, does not mean the parent. It means the would-be employer.)
"Sophisticated" workers are staying out of semiconductors because it's an industrial job. As such, all semiconductor jobs suck relative to the same job anywhere else.
Any software jobs in semiconductors pay far less than software elsewhere. Hardware jobs require you to be on site continuously often with weird hours--generally in places that you probably don't want to be living at.
I can go on and on. But anybody I knew of who could leave semiconductors did so and never looked back.
I believe OP is saying “at or below” would be more accurate - the current wording completely misses those at the median (which in a normal distribution is the largest single group within the set with a shared result).
Maybe my IQ is too low to understand this, but isn't IQ derived from a presumed normal distribution, which is symmetric about the mean, and therefore contains a set of 50% the population that is below the mean?
Or by "making claims" do you mean the claims of the premise of an IQ test? That is, it's ludicrous to assume that "actual intelligence" (whatever is being approximated by IQ tests) would fit perfectly into a normal distribution, and it's therefore ludicrous to claim IQ can measure intelligence?
For the statement to be true, there would have to be zero people with a median IQ, i.e., not a normal distribution but a sharply bimodal distribution.
Analogy: The median U.S. male height is 5'9". Approximately 15% of the male population is 5'9". How can you have 50% of the population <= 5'8", 50% of the population >= 5'10", and 15% of the population == 5'9"?
(This isn't a mathematical constraint on the truth of the statement, but... its truth would also depend on there being zero correlation between being low-IQ and lack of labor force participation, which I find implausible.)
It seems dependent on measurement precision, effectively the bucketing resolution of the histogram. If you measure both 69.00 and 69.99 inches as 5'9" then the number of people in the "median height bucket" would be more than if you broke it into 100 buckets from 69.00 to 69.99. Since presumably nobody can be exactly 5'9" (or 5'9.00", or 5'9.000"...), then the number of people in the "median" bucket must represent the inaccuracy of the measurement.
So the greater precision you can measure, the less people are "inaccurately" at the exact median score - since buckets of .01 inches will contain fewer people than buckets of 1 inch.
(I've noticed a lot of problems end at this boundary of continuity and discreteness. I find it existentially disturbing. As a bewildered non-mathematician, I've gone down quite a few rabbit holes reading about the axiom of choice and intuitionism – My favorite was when I discovered the Hairy Ball Theorem, but I also enjoyed this idea from "Does time really flow?" that describes time as an increase in precision of measurement of uniquely numbered particles [0].)
The author is a professor of "monetary theory, financial economics, and welfare economics" yet posits the IQ requirements of chipmaking with logic that appears to have been pulled out of his ass.
"A single speck of dust can ruin a production run." As if that is the only thing requiring such delicate hands. Even if we accept his assertion that we need rocket scientist-level folks to build chips, he seems to forget that demand is shared by all the other things that need rocket scientists, including but not limited to actual rocket science.
To say nothing of the wonderful irony of an american professor using such faulty logic to write a post about how stupid americans are. And can we talk about his unspoken faith in IQ testing?
"Lord, will you destroy a 330MM person economy for the lack of of 60k Righteous High-IQ Workers?" I think maybe it's a parody, but if nothing else I'm not seeing any justification for a specific number of high iq individuals needed to maintain anything, much less a society or industry. An assumption that intelligent people must be necessary to run something complex, while perhaps reasonable, still doesn't imply anything about the proportions or how that happens. How would we know the organizational structure or incentives wouldn't determine things more than some fractional difference in number?
This is Intel erasure. OK, fine, they're a step behind the process leader for a sustained period for the first time in a few decades. But... "we can't make semiconductors in the US" ... ? Wat?
It is weird that software engineering (especially cloud and web) pay very highly while other highly skilled jobs like hardware programming typically pay much less. My suspicion is that its related to capital requirements.
A web dev has extremely low capital costs related to his work product. He can walk out of the job and be running again on a $500 laptop.
A hardware engineer is restricted by the fact that his work product is reliant on combination with large manufacturing orders. He can't fund a order for 1000s of IC units himself.
Because of this we see a lot more talent shift into the software world while other areas stagnate.
One major inhibitor (perhaps the most major inhibitor) of high IQ people is social repression. For example you cannot get a better software job merely by having a higher IQ. You have to find the right opportunity specifically looking for a high IQ person. Doing so greatly reduces odds of candidate selection for both the candidate and prospective employer.
> More specifically, although not everyone in semiconductor manufacturing requires a PhD, pretty much everyone has to be of above average intelligence and many will need to be in the top echelons of IQ.
The entire article is predicated on this asspull. Where's the evidence? Has someone done a study on the average IQ of TSMC workers? This assumes so many things:
1) That semiconductor manufacturing is more complex than manufacturing commonly done in the US. We make rockets! Complex pharmaceuticals! Nuclear weapons! What's the complexity factor of semiconductor manufacturing to those processes? 1.1x? 2x? 100?? 1000x???
2) That just because the manufacturing process is more complex, the skill of a line worker in that process must be higher than a line worker in a simpler industry. Maybe ASML makes very user-friendly machines? Again, it's possibly true, but it's just asserted as ground truth with no evidence.
3) That the higher skill required of workers must correlate 1:1 (or roughly so) with IQ. "A single speck of dust"--that sounds like workers must be detail-oriented and follow processes to the letter, and any veering off that path would have disastrous consequences. Does that sound like an environment where Feynman or Einstein or Curie would thrive just because they had "high IQ"?
And instead of justifying the central leap of logic the entire article is premised on, the author instead spends their time on such insights as "half the American population is below 100 IQ"...
This is a weird article. He's saying we can't do semiconductor manufacturing in US because we don't have enough high IQ people to work in the factory. Is this actually the case? Designing the chips is a high IQ task, but why would you need brainiacs in the factory? Versus someone really good at following instructions.
1) "High" IQ is necessary to make semi-conductors? Why, because of low tolerances? In that case it just needs people who are really careful.
2) IQ is universal. This is not the case - a 100 IQ is set to be the middle of the curve - so a 100 IQ today would be a 120 IQ 30 years ago. (Flynn effect)
3) IQ is static. This is also not the case. Anyone can improve their IQ. If you really want a high IQ, practice taking IQ tests. An Air Force officer managed to get his IQ up to 185 because it was his job to administer IQ tests. Education has also been shown to increase IQ.
4) IQ is a useful metric. There is some evidence to suggest that it relates to improved academic performance, but a lot of the early science was based on bad data.
64 comments
[ 3.0 ms ] story [ 134 ms ] threadCost of workers, cost of manufacturing, regulatory and governmental overhead…. These influence fans more than smarts.
Credible chance regulation and overhead influence software more than intellect does. I'd certainly believe that as a premise.
Just the drop in the stock market equated to the earnings of 75,000 quants averaging $7,800 per day for 45 years.
Employee compensation economics don't make much sense to me. There's some factor in favour of maximising headcount which argues for many people on low pay instead of few people on high pay.
It's also considered _better_ to have your long term engineers occasionally notice the gap between their pay and the open market, leave, and then be replaced by new ones who cost what the open market predicted.
How many people ever got fired for selling the future for a short term gain? (how many of the few who did, didn't just get another, better job because of more experience?)
Pretty much everyone who does really well in physics ends up as a quant or something similar because the pay starts at 10x what you get as a researcher and goes up from there into the stratosphere.
The only people left behind are those that didn't-quite-make-it, or truly deeply love physics and simply don't want to get an extra zero on their pay-cheque.
It's really not that different for any skill that is hard (whether it is by requiring "smart" people or just ten years of practice to perfect the craft) to learn and niche enough that people normally don't choose that career path
Edit: this is not to say that the USA should not develop its own semiconductor industry again - the issue there isn't profit, it's supply chain and security, prevention of backdooring or denial by China.
"We need more suckers willing to do incredible hard things for little reward beyond the work of doing them. while the management and their consultants reap millions in ongoing profits for bringing a diluted, not-quite-good-enough version of that work to market"
edit to add: These jobs are often at places founded by "genius motivated by love", but now its the second generation or more on from that. the management wants that kind of success but they want it on their schedule (gotta make the quarterlies). they want it without the risk of failure and need for multiple attempts and extended effort, which made the original company work as much as the genius, but no one recalls that part now. they want meetings and reports and rigidly regulated routines more than they actually want the genius level output.
Complete clownshow from these upper managers. If workers aren't entitled to a job, you're not entitled to high IQ talent, dumbasses!
I'm not convinced high intelligence is a necessary condition to get into chip design. You just need to be willing to spend the time to finish advanced degrees in university and get trained. This is mostly taken up by foreign students because citizens can earn well becoming an SWE right after bachelor's.
The score is then adjusted downwards so that the median Canadian IQ is set at 100.
Any IQ test with any sort of validity will not contain any country-specific questions. ("Regatta" became irrelevant as a "gotcha" for those questioning the validity of IQ tests so long ago that it's long since become an emeritus member of same, really.)
> and mathematical notation) but score higher on average due to a variety of social factors.
I am not aware of any evidence that Canadians score higher than Americans on IQ tests when normalized by audience.
Americans do quite well on PISA, for example, compared to their ethnic relatives <https://www.unz.com/isteve/the-new-2018-pisa-school-test-sco...> Asian Americans do better than Asians; whites do better than Europeans; Latinos do better than Latin Americans; and blacks do better than Africans.
Hispanics and especially blacks' scores drag the US average down. Both white and Asian Americans score higher than Canada (and white+Asian is essentially Canada's racial makeup), and higher than New Zealand, Australia, the Netherlands, Belgium, Germany, France, Norway, and the UK; Estonia is below US Asians but above white Americans. Norway is by far the wealthiest Nordic state but its average is only two points higher than the US national average, despite not having a demographic that is 13% of the population and scores 85 points below the white American average.
This was satire right? Am I missing the joke? He's just using the definition of IQ and a normal distribution to come up with some number...
But yeah, that's just like everything. There's a shortage of luxury vacation rentals that fit my budget, too. There's a shortage of fine steak at McDonald's prices.
If you want these people, you have to pay them the way they're accustomed to be paying (most of them are smart enough to know their value), and you have to give them something to do that they actually want to work on. Won't do one or both of those, and now you're complaining about a shortage? It's hard to be you, I guess, but nobody else cares.
(Note: "You", above, does not mean the parent. It means the would-be employer.)
Any software jobs in semiconductors pay far less than software elsewhere. Hardware jobs require you to be on site continuously often with weird hours--generally in places that you probably don't want to be living at.
I can go on and on. But anybody I knew of who could leave semiconductors did so and never looked back.
I wish people would stop making claims that mathematically cannot be true in normal distributions.
Or by "making claims" do you mean the claims of the premise of an IQ test? That is, it's ludicrous to assume that "actual intelligence" (whatever is being approximated by IQ tests) would fit perfectly into a normal distribution, and it's therefore ludicrous to claim IQ can measure intelligence?
Analogy: The median U.S. male height is 5'9". Approximately 15% of the male population is 5'9". How can you have 50% of the population <= 5'8", 50% of the population >= 5'10", and 15% of the population == 5'9"?
(This isn't a mathematical constraint on the truth of the statement, but... its truth would also depend on there being zero correlation between being low-IQ and lack of labor force participation, which I find implausible.)
It seems dependent on measurement precision, effectively the bucketing resolution of the histogram. If you measure both 69.00 and 69.99 inches as 5'9" then the number of people in the "median height bucket" would be more than if you broke it into 100 buckets from 69.00 to 69.99. Since presumably nobody can be exactly 5'9" (or 5'9.00", or 5'9.000"...), then the number of people in the "median" bucket must represent the inaccuracy of the measurement.
So the greater precision you can measure, the less people are "inaccurately" at the exact median score - since buckets of .01 inches will contain fewer people than buckets of 1 inch.
(I've noticed a lot of problems end at this boundary of continuity and discreteness. I find it existentially disturbing. As a bewildered non-mathematician, I've gone down quite a few rabbit holes reading about the axiom of choice and intuitionism – My favorite was when I discovered the Hairy Ball Theorem, but I also enjoyed this idea from "Does time really flow?" that describes time as an increase in precision of measurement of uniquely numbered particles [0].)
[0] https://www.quantamagazine.org/does-time-really-flow-new-clu...
"A single speck of dust can ruin a production run." As if that is the only thing requiring such delicate hands. Even if we accept his assertion that we need rocket scientist-level folks to build chips, he seems to forget that demand is shared by all the other things that need rocket scientists, including but not limited to actual rocket science.
To say nothing of the wonderful irony of an american professor using such faulty logic to write a post about how stupid americans are. And can we talk about his unspoken faith in IQ testing?
A web dev has extremely low capital costs related to his work product. He can walk out of the job and be running again on a $500 laptop.
A hardware engineer is restricted by the fact that his work product is reliant on combination with large manufacturing orders. He can't fund a order for 1000s of IC units himself.
Because of this we see a lot more talent shift into the software world while other areas stagnate.
The entire article is predicated on this asspull. Where's the evidence? Has someone done a study on the average IQ of TSMC workers? This assumes so many things:
1) That semiconductor manufacturing is more complex than manufacturing commonly done in the US. We make rockets! Complex pharmaceuticals! Nuclear weapons! What's the complexity factor of semiconductor manufacturing to those processes? 1.1x? 2x? 100?? 1000x???
2) That just because the manufacturing process is more complex, the skill of a line worker in that process must be higher than a line worker in a simpler industry. Maybe ASML makes very user-friendly machines? Again, it's possibly true, but it's just asserted as ground truth with no evidence.
3) That the higher skill required of workers must correlate 1:1 (or roughly so) with IQ. "A single speck of dust"--that sounds like workers must be detail-oriented and follow processes to the letter, and any veering off that path would have disastrous consequences. Does that sound like an environment where Feynman or Einstein or Curie would thrive just because they had "high IQ"?
And instead of justifying the central leap of logic the entire article is premised on, the author instead spends their time on such insights as "half the American population is below 100 IQ"...
1) "High" IQ is necessary to make semi-conductors? Why, because of low tolerances? In that case it just needs people who are really careful.
2) IQ is universal. This is not the case - a 100 IQ is set to be the middle of the curve - so a 100 IQ today would be a 120 IQ 30 years ago. (Flynn effect)
3) IQ is static. This is also not the case. Anyone can improve their IQ. If you really want a high IQ, practice taking IQ tests. An Air Force officer managed to get his IQ up to 185 because it was his job to administer IQ tests. Education has also been shown to increase IQ.
4) IQ is a useful metric. There is some evidence to suggest that it relates to improved academic performance, but a lot of the early science was based on bad data.