Fabrice Bellard.
84,763 karma · joined February 20, 2007
Closure was the homegrown framework that's everywhere. It is a different flavor of bad than Angular. Angular was basically "You too can make your Javascript look like HTML", Closure is "You too can make your Javascript look like Java", and nobody bothered to ask Why? For that matter, React was "You too can make your Javascript look like Ocaml." JQuery was the only framework that really let Javascript be Javascript (other than writing in vanilla JS, which post-ES2015 wasn't as insane as it sounds).
Products are only viable when you do a lot of work to solve a problem, which is then shared by many potential customers. It makes sense to amortize the high costs of solving the problem across all the different customers to reap economies of scale. Businesses then pay for product design, hosting, and maintenance because those costs can also be amortized, and they are cheaper than a bespoke solution for each customer. But if the bespoke solution becomes cheaper than that, because it's generated by an LLM that doesn't need to be paid a living wage, there's no reason to have the product in the first place. Just solve your damn problem and let other people solve theirs.
This is an underrated factor in the market structure of the AI bubble going on now. Anecdotally, we're not seeing new AI-based products other than the foundational models and coding assistants gain traction. Why? Because AI makes it so easy to customize the solution to your particular needs that everybody is just solving their particular needs directly. It's the opposite of the Internet boom, where the network greatly expanded the potential market, reduced the cost of reaching them, and made it economical to spend large amounts of money building a software product that had a TAM of billions. The AI boom instead enables extremely cheap customization, which shrinks the market to a single customer who uses AI to directly solve their problem rather than building a product that's generally applicable.
Open question whether this model is actually more economical than using the cloud AI service. The whole reason the industry moved to cloud computing in the first place was because computing had very high fixed costs, and the more these could be amortized over a fully-loaded query stream, the lower the unit costs.
You also can invert the relationship and run a NAS on a physical home server. My storage solution is a JBOD ("Just a bunch of disks") attached to a Proxmox virtualization host, where my home server runs as a VM inside Proxmox and then exposes selected directories via Samba to the home network. There's no second computer, I'm just running ~5 of them on one physical server. But by doing this, I get some isolation where if I've downloaded a questionable file, I can quarantine and inspect it (and roll back to a VM checkpoint if it compromises the machine) before I unleash it on the non-technical users of my home network.
When it comes to laws, libertarianism faces an internal contradiction. A libertarian believes that they shouldn't be able to control other people; therefore, if a company wants to make an unrepairable unrootable surveillance device and the population consents to using it, the libertarian believes that it is not their right to interfere. The only thing they can do is create a better repairable rootable device, which is how we got GCC, Linux, Ubuntu, Firefox, Graphene, etc.
When people are buying an asset speculatively, they often all enter the market at once, which drives up the price spectacularly. But then once they've all bought, what happens? That demand drops off; everyone who wants to own memory chip stocks already does. You're left with the value investors who already held and bought early, who see that their asset is now wildly overpriced and want to get rid of it. So the price then drops until that supply drops off and you reach a new equilibrium, often exacerbated by the speculators who see their position evaporate and want to get out.
When was the last time you saw Craig Silverstein or Steve Wozniak or Urs Hoelzle or David Cheriton or Jan Koum or Jawed Karim in the news? You don't, because they don't seek fame. At most you might see them on the donor list of local museums or cultural attractions (which they are, abundantly). Even Marissa Mayer, who was very much seeking the spotlight until she got the Yahoo CEO job, has basically fucked off to the sunset and now just runs some extravagant haunted houses in Palo Alto and helps fund the Junior Museum.
There's plenty of tech wealth that's very quiet, basically keeps to itself, and limits itself to some quiet local philanthropy. But by definition, you don't hear about quiet wealth.
I feel like many court cases are textbook instances of Everyone Sucks Here. I needed therapy after the case because it had so shaken my faith in both the justice system and humanity.
LLMs in general are pretty terrible at spatial arrangement. You should see some of the architects' critiques of AI-generated floorplans that are on YouTube; the LLM generates a layout that is just nuts. I've found them largely useless at putting together UI mocks; Figma or Photoshop is still way easier to work with if you just want to cut up UI elements and rearrange them.
Interestingly, a lot of people are looking at LLMs or automated drones as "AI", but I think that capitalism is a better contender. It's a self-healing, self-organizing system that has its own logic and seems to have escaped the desires of its human creators. It treats individual firms as the cells within it, individual humans as its DNA molecules, and it even has a form of tissue differentiation via industries. All efforts to shut it off, even through widespread warfare and "tissue" (i.e. human) death, fail, resulting in it becoming even more entrenched.
There's this heavily anthropocentric idea of intelligence that it will look like us, talk like us, and operate on the same scale as us, but it's highly likely that the next layer of intelligence will operate on a much higher level than us, and our intelligence will be to it as DNA replication is to ourselves.
IIRC there was a search revenue miss last quarter too. Likely because of AI cannibalization of search results. Classic search is very heavily monetized, and a monopoly ad market. AI mode is currently lightly monetized, and a much more segmented market, and one that is harder to target ads for. Plus there are easily half a dozen LLM vendors now, while Search was a clear monopoly.
This is something investors should be very concerned about. It's not at all clear that whatever comes post-Gemini will be nearly as profitable as Search has been for the last 20 years.
Would definitely not buy with my own money. I actually really like a lot of LG's other appliances (we have a washer, dryer, and dual ovens from them that are great), but for a TV, I just want a dumb screen that I can hook an HDMI cable to and run off a computer.
A good pop-culture analogue is the movie Mr. Holland's Opus. Mr. Holland was a struggling but talented young musician; he dreamed of composing his magnum opus and leaving his mark on the world. But he has take a job teaching music to support his wife and young son, who turns out to be deaf. The job requires long hours and he finds himself lacking much time to connect with his son. 30 years later, he's laid off. His former students come back to stage a performance of his symphony in the high school auditorium.
The point is that the students were the opus. Not the music, which is thoroughly mediocre. The music was just the vehicle to bring the students together. A lot of critics wonder why symphony at the end isn't more impressive, it being Mr. Holland's life's work, but don't realize that that's the point of the movie. Your actual life's work is in the decisions you make, day in and day out, in response to the constraints that your life actually presents you, not some image that you may have formed based on a platonic ideal of no constraints.
For homelessness, it's things like shelters, free meals, needle exchanges, etc. Make the life of the homeless easier without actually getting them into permanent housing. The actual solution to eliminate homelessness is to build more housing, along with financial arrangements to incentivize ownership and make it possible.
For taxes, it's accountants and tax preparation software like TurboTax or H&R block. The actual solution is to make the tax code simple enough that anyone can understand it and then withhold the appropriate amount so that nobody even needs to file. The big tax preparers have repeatedly lobbied against this.
For medicine, it's lots of prescription medication that you have to be on for life to "manage" your condition, as well as recurrent doctor's visits to diagnose and treat it. The actual solution for the vast majority of conditions that kill Americans today is diet, exercise, fiber, reduce stress, stop smoking, and wear your seatbelt. Modern medicine is also very adept at solving acute problems like broken bones, bacterial infections, appendicitis, etc. But once you've got a chronic condition of indeterminate source (many of which are actually solved by the advice above), you're in the system and can expect to pay through the nose for no solution.
For many construction projects (like say CAHSR), it's performing endless studies about how to build the project without actually building it. Some level of design is necessary for successful outcomes, but many agencies go around in circles with proposals and reports and environmental impact studies and voter referendums without actually having anyone actually pick up a shovel and start moving dirt.
Federal: 1240 + 0.12 * (50400 - 12401) + 0.22 * (105700 - 50401) + 0.24 * (201775 - 105771) + 0.32 * (256225 - 201776) + 0.35 * (640600 - 256226) + 0.37 * (1200000 - 640601) = 399938.83
State: 0.01 * 11078 + 0.02 * (26264 - 11080) + 0.04 * (41452 - 26265) + 0.06 * (57542 - 41453) + 0.08 * (72724 - 57543) + 0.093 * (371479 - 72725) + 0.0103 * (445771 - 371480) + 0.0113 * (742953 - 445772) + 0.0123 * (1200000 - 742954)
SF property tax on a $4.5M property: 1.18268325% for a total of about $53,220.75
Total: about $493K for an effective tax rate of about 41%, assuming this hypothetical SF resident is single and just purchased their $4.5M property this year.
I think OP is overselling the point, but not by nearly as much as people think.
1.) What they do to your 95th percentile latency. Users are often very sensitive to tail latency: a service that responds in 150ms 19 times and then takes 2s on the 20th is still perceived as annoyingly slow. With job queues, the reason for that slowness could be as simple as "it was the 20th request to arrive during a period of high demand, and backends couldn't keep up". The whole point of queues is so you can gracefully handle this case without overprovisioning your backends by a factor of 20x, but if the user is still going to consider this a miss anyway, you have to overprovision the backends anyway. There isn't really another way to handle this other than having spare backend capacity. Also note that in many cases the user hitting "refresh" doesn't cancel the existing queued job, it just adds another one to the queue. Which brings us to...
2.) They can turn simple failures into cascading failures. There were several postmortems that went something like "Service X became overloaded because of an unexpected flood of requests, leading to several individual replicas shutting down. This led to more requests being routed to the remaining replicas, which overloaded them too and led to all of Service X going down. When SRE attempted restart Service X, requests queued in the job queue were all retried en masse, which led to an overload of the partially-restarted service and a subsequent failure. SRE had to limit requests upstream and manually drain all job queues and bring Service X back cluster by cluster to restore service health."
The root principle here is that any distributed system needs a concept of backpressure. When critical downstream dependencies are overloaded, they need to pass this information back up the stack to the entry point, which needs to start denying requests from the user or do a simpler fallback that doesn't put load on the overloaded service. Naive queueing does not work, because the requests are still sitting there in the queue waiting to overload the downstream service once it becomes available again.
You can bolt backpressure onto a job queue system (by eg. rate-limiting requests to a service that has just come back up, or rate-limiting based on response time, and/or falling back to simpler algorithms), but at that point, it's a backpressure system, not a job queue. The semantics are very different from a system that guarantees eventual delivery, just not sure when. You need to be able to handle partial failures and adapt with different algorithms at multiple points within the system.
Higher interest rates could absolutely deflate AI capex. Indeed, other than CIOs deciding en masse that AI doesn't actually save them money, it's possibly the only thing that could deflate the AI bubble. Higher rates decreases the availability of capital; the bulk of the marginal capital losses are in "risk-on" assets like (currently) AI investments; there goes the funding source that's been used to build all these data centers. It's very much like how the 1.75% increase in rates from 1999-2000 caused the dot-com bust, or the 4.25% increase in rates from 2004-2006 caused the 2009 crash, or the 5% increase in rates from 2022-2023 caused the last tech crash.
After the initial set of hostilities ended, the market largely shrugged off the Straight of Hormuz crisis. But now that it's 3 months later, the Straight is still not open, and if anything hostilities are intensifying, the market is finally starting to price in higher inflation and higher rates. That's why 10Y Treasuries have started creeping up almost to 5%, and GOOG has fallen 12% since its $400 high a couple months ago, and why MSFT is down more than 20% since its 2025 high. The memory chip makers are more exposed, with more of a bubble, and so they could potentially crash even harder if rates start to rise.
I've hired ex-Amazon engineers at Google who believe that your job is to fuck over as many other teams as possible in the name of delivering your feature as quickly as possible so you can climb the ladder, and who believe your manager's job is to back you up while you fuck over those other teams and steal credit from other engineers on them. They tend to struggle in Google's more collaborative culture, and I'd imagine they'd also struggle in other companies that are a little less sociopathic. The problem is particularly acute with people who have never worked anywhere but Amazon; older employees coming out of Amazon are more like "Well, that's just how Amazon is, I'll adapt myself to whatever culture my current job has", but junior engineers are often shaped and molded by it because they've never known anything else.
I'm often a big fan of going to industry early because a.) you make more money early which can then compound and b.) you actually do learn a lot of tacit knowledge working in industry, very often more relevant than what you'd learn as a student. But if the choice is between Amazon and Cambridge I would probably do Cambridge.
There's a hidden cost to this, and that's trust in the mapping app. I've stopped using Google Navigation for routine trips where I know how to get to the destination. It used to be worth it for the traffic information, but now that there's a good chance that it'll take you on a slower route for reasons that may be beneficial to Google but detrimental to you, it's not worth it. My brother-in-law switched to Apple Maps for similar reasons: he's finding it simply gives better results nowadays.
https://en.wikipedia.org/wiki/Battle_of_Kyiv_(2022)
It was the failure of Russian logistics and the triumph of Ukrainian logistics that beat them back.
A lot of HN is folks in their late 30s, 40s, and early 50s (and sometimes even older!), so many folks here would've overlapped with the radio era. A lot of folks here were involved in making YouTube/Instagram/TikTok, not listening to it.
https://www.youtube.com/watch?v=HgmUgFEFzco&list=RDHgmUgFEFz...
The mechanism of harm for asbestos is known to be that the fibers enter the lungs and can't be expelled, eventually leading to cancer. Its interaction with T-cells is quite irrelevant there.