mandevil

So traditional Venture firms are not part of this (1). VC buys equity in their start-ups, not debt, so interest rates aren't a direct concern of theirs. And VC firms generally do not borrow, they raise investment funds of investors money and because it is high risk, generally speaking no one lends money to these funds.

Private Equity firms on the other hand do generally take on a lot of debt- they have their own money but they borrow even more (2) and use that amount of money to buy out the current owners of a company. Basically they borrow to turn a steady stream of revenue into the future (an existing software company that is profitable) into a big pile of cash today for the current owners. And then they need to pay off that borrowing over time, from the steady stream of revenue that the software company generates. And if they are wrong about the size of the steady stream of revenue, they basically declare bankruptcy and the people who loaned the money (banks but also pension firms etc.) now run the company.

1: The way that VC changes is that the rising interest rates mean that the returns to safe investments are higher, so in order to attract new capital VC firms have to promise even higher returns as well.

2: They borrow money to get higher returns- if you have 5 dollars in a stock and it goes up 20% you have 6 dollars. If you have 5 dollars and borrow 5 dollars and invest it in a stock that goes up 20% you have 7 dollars after you pay back the loan. The catch is, if the stock instead goes down by 20%, after you pay back the loan you only have 3 dollars. (And if the stock goes down 50% you are wiped out.) But PE firms are confident that their investments will always be winners, so they take out loans like that.

For a while (2018-2022) it was hot to buy up software companies that dominated a single niche and jack up prices. The bet was that since software is such a 0-1 business that they would have great moats and have strong pricing power. My father-in-law is blind and used to use JAWS until it got bought up by a private equity firm (Vispero) which tried to dominate the software accessibility market and jacked up the prices to annual subscriptions. He's a electrical engineer so he was comfortable switching to free software NVDA, but a lot of people paid up.

Now with LLM's that business model is in shambles and the finance side of it is less attractive with interest rates so high. I suspect that this is actually driving the software engineer hiring slow-down, because these companies were actually doing most of the employing of software engineers, much more than the big tech companies.

I don't think this is true. Twitter is able to support roughly as many users as before on only 20% of the employees, but those users aren't Twitter's customers. Their customers were the advertisers who paid them money to advertise to those users. And the majority of Twitter's staff before the acquisition were working for those customers- either keeping conversations to be the sorts of things advertisers wanted to be associated with or categorizing and data-mining the messages that had been sent to figure out the ads to show users or building a advertising management platform for those customers to use or maintaining relationships with all of the advertisers.

After the acquisition, advertising revenue dramatically dropped (to a low of about 1/3 in 2025 though it has apparently recovered up to only down about 50%!), because all of those functions were gutted. This is why Twitter tried to switch to "buying a blue-check" and other premium features (e.g. access to Grok), because they had fired all of the people necessary to run a real advertising business and they needed revenue.

And I know that you don't need that many employees to support Twitter's customer facing business because Bluesky does basically the same thing (message storage/search backend+web app+two mobile apps) and they worked for over a year on just 20 million in total raises (and no advertising or any other revenue)- they did a Series B in 2025 that raised over 100 million, but they were operational for over a year on just 20 million in raise. So yeah, it doesn't take that many people to just present the messages of users to other users, but what Twitter did was to cut back on what they did for their customers and accept the big drop in revenue that came with it, and that's not really proof of anything.

20 years ago some friends and I were walking by 33 Thomas St and discussed how that must be the building that they imprison Magneto in. Weird looking building.

MIT uses a weird class numbering system. Instead of CS101, they call it 6.0002.

The first digit is the school that offers it, then after the decimal point they have their number of classes. (EECS is unusual in that it has four digits after the decimal point, most every other course has just three.) In general, classes that are required start with 001 and increment up, such that 001 is the intro course, 002 follows from that, etc. Then electives start around 015 or so.

Course 1 is Civil Engineering, Course 2 is Mechanical engineering, Course 3 is materials science, etc. 6 is Electrical and Computer engineering.

So 2.009 is a mechanical engineering course, in this case one for students towards the end of their time. It's called Product Engineering Processes, where you create a new product from beginning to production.

The decline and fall was within a lifetime. Not just a random person's lifetime, but that of a single famous individual. By any of the usual endpoints (Indian Independence, Suez Crisis, Winds of Change/South Africa's withdrawal from the Commonwealth) the Empire ended while Winston Churchill was alive- he was born in the 1874, at the height of the Empire, graduated Sandhurst and joined the British Army in 1895, took Queen Victoria's shilling in several colonial wars, then was Secretary of State for the Colonies in the 1920s, and died after Britain had withdrawn from most of her African holdings.

You do not have a right to drive a car in the United States. It is a privilege for which you must be licensed, which necessarily means you can lose your license and not be able to drive. There are medical conditions which absolutely prohibit you from driving. At least in the state where I did Drivers Ed, if you had certain kinds of seizures you could not drive again until you'd been seizure free for 18 months. It is your responsibility to arrange your life to match that.

No matter what other thought experiment you create, you do not have a right to drive a car in the US, it can be revoked quite easily.

Fortunately for those companies, they understand this and are taking steps to address the shortage. It's just that the physical reality of our universe prevents any production increase for the next two or so years. Let's say you start a fab today. By the time it gets online, and is able to produce commercial quantities of RAM- let's say you get lucky and it's 2030- you will be competing against at least two brand new fabs from Micron, at least three more brand new ones from SK Hynix, and I'm not even sure how many from Samsung. By the time your new fab is online there will be much more production from all of the existing players who have already mastered the process of producing RAM and so will have an advantage over you, who is learning. The supply crunch is entirely short term, for which there are no fixes in a universe where the laws of causality apply. The only solution to the immediate supply crunch is to have started building a new memory fab back in 2021 or 2022.

But the reason that semiconductors are so vulnerable to boom/bust cycles (going back decades) is that takes a very long time to build the fab- you have nothing for years, and then the fab completes and enormous numbers of chips get made. So if you get demand wrong either direction you have problems. Either you flood the market or prices go way up.

I don't think it is possible to beat the fab construction delays. This isn't about things being slow in the US: TSMC took four years to go from site selection to some production at their Fab 22 plant in Taiwan (it is currently only 60% built out, is my understanding). And TSMC is best in the world at getting frontier (logic) semiconductor fabs built, it took them slightly longer (five years) for Fab 21 in Arizona. Logic fabs are not the same as memory fabs, but they are close enough to be a good comparison for timeline.

Basically, unless you started fab construction (specifically to make memory chips) in 2023 you cannot change production for 2027 and 2028. The only thing that will affect prices between now and 2028 is demand, production can't go up because the fabs to make those chips don't exist. If you started now you might get memory chips in 2031, and can you bet 40 billion dollars that the demand will still be there? (A current generation logic fab costs about 20 billion. By 2031 a new frontier logic fab will cost about 40 billion). I don't have prices for a current memory fab, but as I understand it they are comparable.)

The de-skilling of retail clerk I think predates even the GUI. I think it goes back to the laser/upc/database systems of the 1970's and 1980's.

In 1970 retail clerk is a medium-skill medium prestige job. There were lots of stories of people who started as a retail clerk and end up as CEO, and even people who aren't that ambitious could still carve out a comfortable middle class lifestyle as a retail clerk. But that's because there was a reward for experience. Before the laser/UPC/DB system rolled out, you needed at a glance to be able to know the price of an orange, or the difference between the expensive and the cheap mushroom. Because there would be a paper tag on every cheap mushroom that said "10 cents", but what if someone swapped that paper tag onto the expensive mushrooms that were supposed to cost 30 cents? You needed to know the difference, and have the real prices memorized to be able to be fast enough to keep the customers happy.

This also meant that you couldn't have a large retail store (something at the scale of the modern Walmart or Target), because you couldn't get enough clerks who could remember all the necessary prices and so inevitably you would lose money to tag switching. Department stores solved this by forcing you to buy each good separately in each department- you would pay for your housewares in the housewares section and your clothes in the clothes section, and they had another whole army of clerks in the basement managing store credit to keep it low-friction enough. But that's very high-touch, demanding a lot of workers, and those people still have the rewards of experience so they are all comfortable careers.

But then the laser/upc/database system came out. And suddenly you would get a natural language description of the item (or even a picture in the modern systems) from scanning the UPC. And you could scale the store up as much as you wanted, just needed to buy a bigger hard disk for the database, so the stores tilted more towards capital and less towards labor. And that eviscerated the career path of most retail stores, because store owners no longer believed that keeping retail clerks around benefited them.

This is machines hallowing out a job and alienating workers from their labor for the benefit of capital- exactly what Marx had described a century earlier for factory workers in Das Kapital. But for reasons of class and gender we mostly ignore retail clerks so it doesn't seem to have attracted near as much attention.

The modern Terminal is based on a private fork of Chromium to give it the look and feel of a VT100 terminal, and integrate their private networking and security technologies. The BBT predates HTTP, and backwards compatibility is hugely important to the company- they have a museum where a second-generation Terminal from ~1985 shows the current news, because they are so dedicated to backwards compatibility that they still can support the 1985 hardware.