We may have to wait for another three years.

Which is also a clue that he isn’t short selling.

There are two ways of making money when a stock goes down. One is to sell the stock short. The other is to buy a put option.

A short sale is extremely risky. Say the shares are at $50 and you think they’re going to go down, so you sell 1000 shares you don’t own (short selling) and agree to buy them back by some date in the future. If you’re right and the stock tanks to $20, you can buy the shares and pocket $30,000. But, if the stock doesn’t sink, you might have to buy the shares for $60 each, so you lose $10 per stock, or $10,000. If there are tons of people shorting the stock, you can get a short squeeze, where everybody needs to buy shares to close out their short position, and because everybody needs to buy, the stock price rockets up, so you get people having to buy a stock that used to be $50 for $200, leading to $150,000 in losses for a 1000 share short where the maximum possible gain was only $50,000.

An option is much safer. There you’re buying the option to sell the shares at a certain price at some time in the future. Say you think a stock is going to crash. It’s currently trading at $50/share. You can buy 1000 put options at a strike price of $40 with a date 1 month in the future. It will cost you something to buy those options, say $1 per share, so $1000. If the stock goes up or stays at $50, your bet didn’t work out. You don’t have to sell the shares, you just tear up the options contract. You’re out whatever you paid for the option, say $1000 here. But, say the stock tanks and it’s now at $20/share. Now your bet did pay off. You can buy 1000 shares at $20 each for $20,000, then immediately exercise your option and sell them for $40,000, netting you $20,000. With put options the upside is significantly smaller, but the potential downside is tiny. It’s just the cost of the options.

Someone predicting a crash within 3 years isn’t going to short sell the shares. Between now and then the shares could continue to rise for a while, and they’d be on the hook for a huge payout in that case. If they buy options the down side is much smaller. They may have to re-buy new options a bunch of times. But in the worst case they just have to let the options expire unused and eat whatever cost they paid for them.

For the coming AI crash, I don’t think it will be very soon. I think there will be a crash. But, I think the government will try to keep the bubble from bursting. Too much of the US economy is now invested in AI. So, even under Biden, or Harris, or Obama they’d try to prevent a catastrophic crash by using taxpayer money to prevent the most damaging bubble burst. With Trump, there’s going to be even more government interference in the market. His backers are crypto bros. They’re the ones making him billions on his meme coins. They bankrolled JD Vance’s political career. If they demand that he rescues their failing companies, he’ll do it. And, since the GOP does whatever Trump wants, they’ll just fork over literal trillions in taxpayer dollars to keep things from crashing. But, eventually there will have to be a crash, because there’s just not a sustainable business model in any of this, at least not at anything like the current scale.

Also, the way short positions work is that the people who are most successful at shorting a stock are the ones who have a megaphone to announce they’ve shorted the stock. They go on as many podcasts, news shows, interviews, etc. as possible to say things are going to crash. Because, the more people who hear about it, the more hesitation there will be to invest, which means the more chances of their prediction coming through.

So, he’s not just some guy who is betting on the bubble bursting, he’s a guy who is now heavily incentivized to cause the bubble to burst so he can make his investors a lot of money.

Which bubble was that?

AFAIK none of the previous AI bubbles really had much investment. There was hype, but it was hype within the computer science, or cognitive science fields. But, maybe there’s a financial bubble bit that I’m missing.

Uber used accounting tricks to hide their true losses for years. They’ve only recently managed to become profitable by squeezing both drivers and passengers at the same time. Is that sustainable? Almost certainly not, but, for the moment, they’re getting away with it.

That’s the only reason I don’t think it will pop in the next 6 months or so. Even Biden or Obama would have stepped in to try to prevent the economy from crashing. But, there’s the Trump factor. First of all, some of his biggest backers are from the AI “industry”. His VP is tied to Peter Thiel, his biggest donors are Crypto and AI bros. The vast majority of his own personal money is tied up in the current Crypto bubble. In addition, he’s obviously so easily bribed. Even if he he wasn’t interested in intervening otherwise, he could easily be bribed to intervene.

Because of Trump, and the fact that the house, senate and judiciary are all Trump lackeys, I think the bubble will survive until at least the 2026 midterms. If the Democrats take back control of the House and Senate they could take control over spending from Trump, which might mean the bubble is allowed to pop. But, I wouldn’t be surprised to see Trump hand over literal trillions in taxpayer dollars to keep the bubble inflated.

Comparing the coming crash to the dot com crash is like comparing a rough landing to the various crashes on Sept 11th, 2001.

The dot com crash was mostly isolated in high tech. Because it was lead by the Japanese economy starting to fail, and followed by the Sept 11th attacks, the various combined crashes resulted in the S&P 500 falling by about 50% from its peak to the bottom, but it was already back up to the peak value in 2007, then the global financial crisis hit.

This bubble is much bigger. Some analysts say the AI bubble is 17x the size of the Dot Com bubble, and 4x the size of the 2007/08 real estate bubble. AI stocks were 40% of all US GDP growth in 2025, and 80% of all growth in US stocks.

Nvidia’s stock price has gone up 1700% in just 2 years. OpenAI is planning to go public on a valuation of $1 trillion despite losing vast amounts of money. Just 7 US tech companies make up 36% of the entire US stock market, and they’re all heavily betting on AI.

At least when the dot com bubble popped, it left some useful things behind, like huge amounts of dark fibre. But, the AI processors are so specialized they can’t be used for much of anything else. They also wear out, sometimes within months. The datacenter buildings themselves can maybe be repurposed to being general purpose datacenters, but, a lot of the contents will have to be thrown out.

Thanks for the details.

Yeah, ISP routers suck. You wouldn’t believe how bad the one I use is. If you turn off DHCP on the router you lose the ability to set the router’s IP address and netmask. (And the netmask is locked to a /32). The only way to set the router’s IP address is to turn on DHCP, while DHCP is on set the router’s address, and then turn off DHCP. Needless to say, the router’s DHCP is completely off.

Then I told my phone and laptop to send any ips from my local subnet though the wireguard tunnel

Wait, so when you had your wireguard VPN up, you told it to route most traffic through the VPN, but IPs which were the same as your home network (I’m guessing maybe a 192.168.0.1/32 or something) you told it to send those through a different tunnel to your home network?

The end result is that if you went to say tuffminecraft.local and you were on your laptop in a hotel or something, it would use wireguard to send the packets to your home minecraft server as if you were at home?

What setup did this require? A wireguard server at home accepting connections from outside, a wireguard client on your laptop and phone… I guess the wireguard client would have to know to forward any “.local” DNS query over the tunnel to the wireguard server which would then contact technitium?

Also the dhcp server on technitium can be set to automatically generate and propogate a domain name for any device that connects

I think this is pretty standard with dhcp/dns. I have that with my pihole, but some devices don’t handle DHCP the way others do, so they don’t get nice names assigned via DNS. I think that’s a limitation of DHCP and everyone’s different implementation of it, rather than a limitation of pihole, dnsmasq, etc. But, maybe technitium handles weird DHCP clients better?

I haven’t heard of technitium, what is it that you like better than pi-hole?

This is easy to do these days with Pi-hole.

“WTF is this bullshit?!”

I think this is literally what I said the other day.

Someone sent me a link for something to buy at a local branch of a national store. The advantage of the link was that it said what aisle of the store the product was in, so I didn’t have to search everywhere once I got there. I opened the link and I literally couldn’t see the page. First of all there was a pop-over ad of some kind obscuring the entire page. Once I closed that, I was confronted with the cookie banner. Once I dealt with that, there was the regular store page filled with its own ads / “promotions”. I had to hunt everywhere to find the one useful bit of data, which was the aisle number. It took me at least 20 seconds from opening the link to being able to find this aisle number and it left me incredibly frustrated.

Ah, I was just looking at parliamentary systems. You’re right that FPTP really screws a lot of stuff up.

Whereas in a PR parliamentary system, it is extremely rare for any one party to have a majority

I don’t think that’s true. In Canadian elections majority governments are more common than non-majority governments. In Australia it looks like every government has been a majority government since the non-labour parties merged in 1910.

The modern system of prime Ministers where the executive comes from parliament seems to play out better in modern politics.

Yes and no. The problem is that in parliamentary systems like that, if the government has a majority then they’re unstoppable. In a system with a president who has some actual authority, or a king who isn’t merely a figurehead, the Prime Minister can’t just do everything he wants. There still needs to be some negotiation.

On the other hand, the world has a lot of authoritarians in it who wore (and in some cases still are) supported by popular votes. People seem really bad at picking leaders who want to serve out a 4-10 year term, then retire to a cushy life afterwards.

I think the US chose to have a president act as a sort of a king with a term limit. Other countries saw that and adapted it when they moved away from their monarchies, either giving the president king-like powers or giving them just a ceremonial role as head of state.

What’s funny is that in the UK and in many former British colonies, there’s still a king, but it’s mostly a ceremonial role these days. So, things have basically reversed. A modern king who’s a head of state is basically a figurehead. A president who is the head of a country may have monarch-type powers.

If you understand how LLMs work, that’s not surprising.

LLMs generate a sequence of words that makes sense in that context. It’s trained on trillions(?) of words from books, Wikipedia, etc. In most of the training material, when someone asks “what’s the name of the person who did X?” there’s an answer, and that answer isn’t “I have no fucking clue”.

Now, if it were trained on a whole new corpus of data that had “I have no fucking clue” a lot more often, it would see that as a reasonable thing to print sometimes so you’d get that answer a lot more often. However, it doesn’t actually understand anything. It just generates sequences of believable words. So, it wouldn’t generate “I have no fucking clue” when it doesn’t know, it would just generate it occasionally when it seemed like it was an appropriate time. So, you’d ask “Who was the first president of the USA?” and it would sometimes say “I have no fucking clue” because that’s sometimes what the training data says a response might look like when someone asks a question of that form.

Canada does have a sort-of similar system. It’s just that the “president” in Canada is “the crown”, which is the Governor General representing the current British monarch. It’s much more of a ceremonial role in Canada, but technically the Governor General does appoint the Prime Minister.

Australia has essentially the same system as Canada. In 1975 the Australian Governor General dismissed the Prime Minister and picked the leader of the opposition as Prime Minister so that he could call an election. Described like that it seems like a blatant abuse of power. But, the background was a really dysfunctional government. One party had narrow control over the house, the other had narrow control over the senate, and the senate was blocking everything the house tried to do. I don’t know the full details of what happened in that affair, but it seems like it could be a good thing if a Governor General would step in in a crisis resolve a deadlock.

Canada also has the “confidence votes” part of the crisis in France. AFAIK in Canada losing a confidence vote immediately triggers an election, unlike in France where it can just lead to a scramble to see who can become the new PM among the existing representatives. Because triggering an election is a big deal, it doesn’t tend to happen too often. But it has happened. In 2011 Stephen Harper’s government lost a confidence vote, and there was an immediate election, but he won that election. In 2007 Paul Martin’s government also fell to a confidence vote.

For Americans who don’t have a similar system, a “government collapse” isn’t as big a deal as it sounds. It sounds like there’s a complete breakdown in law and order and nobody’s in charge. Really what happens is that the arrangement that so-and-so will be prime minister and his cabinet will be X, Y and Z is off.

Sometimes it means there are new elections. But, sometimes (as in the French system) it just means that the various representatives all negotiate among themselves to choose a new prime minister. The President then appoints that person. It can vary from the president rubber stamping the decision, to the President getting involved in the negotiations and playing a key role in choosing the next PM. Once the President makes it official, that person becomes PM and then chooses a new cabinet. Before a new PM is chosen there’s a bit of chaos. The government can still vote on things, but the normal process is disrupted because there’s no “first among equals” to lead. In the case of France, normally the President doesn’t (or shouldn’t) deal with the day-to-day running of the government. But, during the previous government’s collapse Macron stepped in to do many things the Prime Minister would normally do.

One minor twist here. In theory, a French President is supposed to handle foreign policy and defence. The Prime Minister is supposed to run domestic things, including the day-to-day government functions. One reason why this government lasted 14 hours (or 27 days if you count his full time as PM) is that Macron was seen as having too heavy a hand in picking not just the PM Lecornu (picking the PM is technically his job as President), but also in picking the PM’s cabinet (which is supposed to be something the PM does himself). As soon as Lecornu announced his cabinet, the rest of the elected reps saw that it was essentially the same as the one they just voted down a month ago. They said they weren’t going to work with Lecornu’s government, so Lecornu quit immediately.

That only works if people use it.

For some reason, people have stuck with Twitter. But, the users of TikTok are not the same as the users of Twitter, so who knows, maybe they won’t stick with it.