e4e78a06

I'm just saying in the absence of any alternatives (which I'm sure there are), it's better to have 1% of trillions of dollars going towards advancing humanity and technology than 0%. I am fully aware that 99% of that money is going to be burned on WeWorks and the like.

Never said TCJA or the Bush era tax cuts were used for good purposes :) but it's probably better for the country in the long run to have excess money flowing into tech VC funds rather than being spent on memecoins and shitty electronics. That way maybe 1% of the funds will have positive returns rather than all of it being set on fire.

Eh, quick DM to Wall Street Journal or Financial Times and if it's a decently large bank the PR fallout will quickly prevent any potential retaliation.

Well I think a big contributor to the current inflation problem was the last round of stimulus passed right as the pandemic was starting to wind down. At that point unemployment numbers had almost gotten back to normal and most revenue numbers for things like restaurants, travel, etc. had recovered to at least 70% of pre-pandemic levels. And then some states like California dumped even more money into the economy through multi-round state funded stimulus checks over a year after the pandemic started. The unemployment programs also should have started winding down the moment vaccines were widely available rather than 6 months after the fact.

I personally know a lot of people that just dumped their stimulus checks on memecoins or wasted it buying spurious goods. I'm sure the used car and electronics market also was greatly affected by stimulus as well.

The Fed also should have limited its stimulus to buying Treasuries rather than MBS. It makes no sense for the government to buy mortgage backed securities (basically a freebie to homeowners/homebuyers who are already wealthy).

Right now the labor market is good for workers

It is good _nominally_. Real wages are basically flat. Meanwhile, for everyone who didn't get a raise or can't currently find a new job for whatever reason, they're losing purchasing power. On top of that, the housing market has been completely destroyed by the Fed printing money and shoving it into mortgage backed securities. In many nicer areas of Southern California you used to be able to get a starter home for $600k (manageable for someone in a working class profession like a nurse or mechanic), now nothing on the market is less than $1.2M.

Also you forget that a big reason the labor market is so tight is because a couple million Boomers retired early during the pandemic.

Gentrification is a classic example of supply and demand from microecon. As an area becomes nicer to live in it gains more and more demand with people with high willingness and ability to pay. Supply remains fixed, because of US zoning laws, or grows more slowly than people come into the neighborhood. The price goes up because of _demand_ going up without supply going up.

Why does it gentrify in the first place? Because the initial group of "gentrifiers" took the time and money to develop the area to make it more appealing, thus increasing demand.

Economics says you can't fix it no matter how much regulation you impose because at some point all the surrounding businesses, etc. will be gentrified too and force out poor people. Even if you freeze rent, ban new businesses from coming in, etc. the existing business owners will start to cater to their new clientele simply because the demand from those customers is much higher. The only thing you _can_ do is ban people from moving to the neighborhood, at which point you've turned into the worst parts of the Soviet Union.

The US and Europe have great safety regulations for workers. It turns out given the choice women would rather not live in the middle of nowhere in -40C weather getting soaked by crude oil and mud every day, even if it pays six figures. Why do men do it? Because there's a societal expectation that they provide for the family as a breadwinner.

teachers with X degree and Y years experience get Y0,000 ±5%

That would never work in tech. And the higher paying firm you go to, the less it works. High performers in FAANG can have refresher and bonus multipliers of 2.5x the base performance rating. In HFT high performers total comp can be double or triple low performers' TC.

I would think it shouldn't work for teachers either. In the US there is a marked difference in teacher compensation between good districts and bad districts and you see the quality difference in the students that come out of those schools.

Ironic they leave out Asian men in the graph at the top. I guess those numbers wouldn't fit their "America is racist" trope.

Other than that though, standardizing pay will balance the playing field towards smaller companies. Bigger companies will face pressure to standardize pay packages per level (see: Coinbase, Uber) which limits their ability to reward top performers. Smaller companies won't have that same pressure. The pay gap is largely a function of how much each gender negotiates on average and if you take away negotiating then people will just leave to places where they're paid what they're worth.

Unfortunately the party in charge right now is trumpeting the idea that inflation must be because companies are price gouging customers rather than trying to slow down domestic spending. The other party doesn't like to balance budgets either and if the Fed even touches interest rates the market (and everyone's 401k) tanks.

We have a huge trades shortage and COVID might force a lot of supply chains to come back at least in part to the US. The demand for new construction has never been higher either.

A large portion of society that doesn't want to work on hot, dusty, dirty worksites might become obsolete with nothing useful for them to do. But there are absolutely jobs that are accessible to no-skill workers with only 1-2 years investment in schooling.

This isn’t a flaw in a healthcare system, it’s a symptom of a broken transport system.

Nobody wants to sit next to piss/weed/cig smelling people and get their shit robbed on public transit (see: BART) if they can afford a car instead. Plus, US light rail runs at much slower speeds than places like China due to NIMBYism, so it also takes longer than driving to get places.

The very least you can do is pay their healthcare .

The very least they could do is come into the country legally. And by the way, a lot of the really crappy jobs are done by legal residents, like garbagemen, sewer maintenance, lineman, etc. Guess what? My local taxes pay for those workers to have good salaries and I'm happy to pay. The US also has a visa for farm workers to come to the US as well.

national borders create an arbitrary and discriminatory barrier to free movement people and enforce QoL disparities across the world.

Spoken like someone who is privileged and wealthy enough to be unaffected by open borders. Globalization has destroyed the American factory job, along with its high wages, and you still claim open borders are the way forward? Pro tip, don't claim to be morally superior when you work as a software engineer making $$$ that has a huge demand supply imbalance. If you really believe what you're saying then go work in India in a bodyshop making $10k and working 80h weeks, but pay your US cost of living. That's what millions of blue collar workers are facing when you open borders.

Please exit your bubble and talk to some real working class people for once. People like you are like the engineers I meet who openly talk about how self driving cars are going to disrupt the industry as they get into an Uber. Zero self-awareness.

everyone in the administrative region gets covered for everything

That leads to rampant abuses. For example in certain East Asian city states with "universal" healthcare people would use ambulances as taxis because they were free. You need copays to prevent this kind of abuse.

And you also forget that the US has a big illegal immigration problem. By and large illegal immigrants make minimum wage or lower, generally under the table (i.e. not paying taxes on it). By covering healthcare for them you are automatically subsidizing illegal immigrants at the cost of citizens and permanent residents. Is that fair?

If you think these aren't legitimate outcomes of allowing everyone to have free healthcare then you're naive.

Climate change is perhaps one of the many reasons both the Roman and Byzantine empires fell or lost ability to project power. The Byzantine conquest of Italy was finishing up when these plagues hit and they never recovered their original strength again.

One has to wonder whether modern agriculture is no longer as affected by these kinds of changes or whether the war in Ukraine is just a sign of things to come.

Not quite true, because on-die data transfer is a lot cheaper than off-die. If you look at full system power consumption between even chiplet based and monolithic Ryzen products there is a large difference.

Not sure why you were downvoted but this is true - desktop Ryzen chips have terrible idle power usage due to the IO die. The APU variants have lower IPC and performance due to lower cache, but have good idle power usage. There's no way to get both good idle power and good IPC.

As long as public transportation in the US continues to be unsafe, dirty, and slow people will continue to drive cars. It doesn't matter how much public transport you build out, if I have to sit next to a guy smelling like piss I will never get on the subway when I have a car.

That cache is not uniform time access. It costs over 100ns to cross the IO die to access another die's L3, almost as much as going to main memory. In practice you have to treat it as 8 separate 32 MB L3 caches.

Also, not everything fits into cache.

Many GB5 (and real world) tasks are memory bandwidth bottlenecked, which greatly favors M1 Max because it has over double a Threadripper's memory bandwidth.

Cinebench R23 is a worst case scenario for M1 because it doesn't have high core utilization and sits in L2 cache. If you look at a broader set of benchmarks (SPEC) then M1 Max in laptop form is competitive anywhere from a 5800x to a 5950x.

Strongly disagree if we are talking a complex problem like coding a distributed system. It is very possible for an implementation to be 90% correct but it won't compile or it's wrong for very subtle reasons. And maybe if there was another 30 minutes it would be debugged correctly but it didn't work out in the 45 minute interview period. There is signal contained in the 90% right attempt that can show a lot of knowledge and skill which you'd be missing out using autorejects.

We are not talking some CRUD app where it is trivial to make each component work or a Leetcode problem that you can spit out in 20 minutes.

Curious how this will work with more complex cases? E.g. distributed systems, concurrency safety, etc. Also, how do you deal with solutions that don't compile but are 95% of the way there conceptually? I find that's a common occurrence in non-FAANG settings that generally would still mean a pass to the next round (sometimes even with strong perf ratings).

And how do you prevent someone from cheating if there's no engineer there on the interview?

If you have any serious medical issues, you carry your own insurance

I've never seen this at any tech company in the SFBA. They cover 100% of the premium for yourself and it's only <$200/mo to cover a whole family of dependents. Their plans are always among the best. Copays are the lowest by far of anything I've ever seen, including ACA marketplace plans, my parents' plans, school offered plans, and non-"tech" company plans.

$200/mo is way, way less than I pay in taxes to CA state alone.

since I prefer Kaiser

Obviously if you want Kaiser and they don't offer it you have to buy it yourself. Personally I don't feel like the difference is too huge that it justifies buying insurance myself. I used to have Kaiser and you get slightly smaller delays (3 days instead of say a week) to see specialists and that's it.

I'd rather know what is "normal" i.e. common.

I think using comparative benchmarks instead of absolute benchmarks is where we differ. The absolute bar - being able to grok large codebases, work on complex systems, etc. with some help from a senior engineer - is not high. It's the equivalent of a proper undergrad education in CS. So when you set the bar at a, in my opinion, very achievable level, the benchmark is $70/hr for interns. We're not talking Jeff Dean or Linus levels of skill here. You can't just look at the percentiles and say it's unachievable. You have to look at it objectively. We're in a talent shortage precisely because there aren't enough people that meet the low bar.

The question should be, if I work my ass off and get good at what I do, how much could I potentially make as a random new grad / 2-3 yoe SWE? And the answer to that is if you meet the already very low bar, you can make a lot. I never compare myself to the average and you shouldn't either.

I know when my friends at startups look to hire they don't target percentiles, they target an absolute level of skill or intelligence.

0-1 years of experience there at Vercel pays about $70k-$100k

That would be getting underpaid if you had the skills I mentioned above.

Now I have about 3 years of SWE experience

My whole point is that years of experience don't necessarily make you a better engineer. It correlates, but in general having more skill is a step function change compared to another year of experience. A senior engineer at mediocre level will be less productive than a new grad engineer at high skill that needs their hand held on design, because the high skill engineer is capable of doing things the mediocre one can't even complete with infinite time.

I know many startups who are now paying _interns_ more than other companies are paying senior devs. It is conceivable to get paid $70+/h at Series A startups as an intern and $140-150k + 20-40k in options as a new grad.

Speaking frankly as someone who has worked at bog standard F500 companies that pay "not so much" to HFT, experience is not nearly as important as raw skill. A 20+ year experience engineer who has mediocre skills will never be able to write a compiler, work on complex distributed systems, or bring new verticals to market (think AR, VR, ML applications) very quickly. A skilled new grad might be able to contribute meaningfully if their hand is held on architecture and design.

For example, it's a reasonable expectation for those skilled new grads to be able to grok and start doing easy tasks in a complex codebase (e.g. on PyTorch or Spark itself or business logic in a distributed system) within a week or less, and for them to require at most 1-2 hours of help from an existing engineer to get started. I posit that a 20+ year mediocre engineer is never going to get the skills to contribute to something like PyTorch if they haven't developed those skills in the 20 years they've worked.

I think you would be surprised how many 100k engineers at Fortune 500 companies can't do things FAANG engineers would consider very basic. It's not that there's a 7x and a 1x engineer but that there's a 0.1x and a 1x engineer.

For example I did a stint at a F500 company where 1) a data engineer did not know how to use multiprocessing to parallelize a multi-terabyte S3 upload 2) an internal tool that fired off hundreds of network requests did not use concurrency and hence had 30+ second latency 3) a certain other team was dumping hundreds of terabytes of logs that nobody even knew the schema of into S3 automatically, and none of it was ever touched. Also seen: "data scientists" who did not know Python and executives that spent more time talking about Scrum than any concrete execution.

It feels very different when you have to pay for your own snacks. I definitely substantially moderated my consumption during remote work (and it's not like the amount they were spending on snacks gets paid to you either). It definitely improved my eating habits but I also miss free Diet Coke and protein shakes.

If I was getting paid the snack amount in cash then I'd probably be ok but as is free breakfast+lunch+snacks saves me on the order of $500/mo. And as someone who is pretty mediocre at cooking the quality of food is better as well.

There are other costs that either make very wide OoO more difficult or more costly. x86 has a lot more flag-based instructions compared to Arm. That adds more dependencies that the reorder engine has to sort through. x86 variable length decoding takes log(n) in decode circuit depth, which either forces a longer pipeline or limits clocks. And obviously AVX512 units are just huge because a decision was made to make them the same latency as normal MUL/ADD/FMA. And x86 designs have to scale in clockspeed from server / tablet (~2.5GHz) to desktop and high perf laptop (5GHz+). That forces suboptimal designs like the 5 cycle L1 in Golden Cove. Meanwhile Apple has a 3 cycle 192kB L1.

You don't solve the problem in your head, you actually take the time to solve it either on Leetcode or on a piece of paper. I've found that in the long run you don't end up memorizing the problem itself but the technique to solve the problem.

For example, I was learning Manacher's algorithm and I directly put the link to the corresponding Leetcode problem into Anki. Set your minimum interval to 1 day and just follow whatever Anki tells you to study. You don't need many cards to memorize the problem solving techniques, my entire probability deck has under 150 cards and half of that is concepts. It turns out the human brain is pretty good at generalizing from a few problems.

Make sure you don't put every problem you encounter because the review workload will get very heavy very fast if you do that. Only put the challenging ones or the ones you get wrong.

You say this now, but I've had to work on 5000 line (literally, 5k lines) classes in legacy code where there are a huge number of dependencies injected manually because the class is doing initialization for a huge number of components. You could argue that this was bad design but I'm sure it didn't start out like that, but just accumulated cruft over time and was too dangerous to refactor. (if prod went down it was at least a few hundred $k down the drain). DI makes it much simpler to wire things like this up without making mistakes like accidentally initializing the wrong class implementation, etc. In polymorphism heavy situations it reduces cognitive complexity a lot.