1,918 karma · joined January 29, 2013
The reason why it became second nature for you is this hand-rolling, manual work. I observed that with math - unless you manually practice problem solving (integrals, differential equations), you might know the mechanics but you'll have hard time solving them. And that knowledge of mechanics will also fade away at some point.
> LLMs have changed the equation. It used to take me 2-3 hours to write a parser for a moderately complex grammar. Now, if I hand an LLM a loosely written, BNF-ish grammar and ask for a recursive-descent parser, it finishes the job in five minutes. At this point, writing them by hand is hard to justify
So, what is the difference between LLMs generated "manual" parsers vs those with parser generators, beside that with LLM that will be done in five minutes + plus some $$ and with parser generators you'll get that for free and under a second?
> Which makes me wonder: what’s the appeal of PEGs or parser generators now? They used to make sense when hand-writing wasn't practical, but what compelling reasons are left to use them today?
The appeal is predictability - use i.e. Bison, feed it with grammar and you'll get always the same output. Not so much with LLMs.
Perhaps they simply don’t advertise it and investors (currently) love companies that spend heavily on frontier models. That said, OW models, especially when combined with RAG, work quite well, and given the current state of the industry, they may be the only sensible way to keep inference costs under control.
Why? It is a great PR to build a hype, especially before the IPO, showcasing how AI is "self-aware" and dangerous, essentially resurrecting Sam Altman's talk about how only a few should hold the keys to this (opening a route to regulation, which is his ultimate goal).
Also, collusion.wiki was recently registered and it looks too vibe-coded for my taste, so let's see will that domain be alive in a year or two.
If your account is that valuable, it's time to consider moving to a self-hosted option. Or at the very least, start by purchasing a domain and linking it to your Google account, while keeping offline backups of all your emails and documents.
The real problem isn't that Google might accidentally ban you; it's that there's no way to get a human in the loop to review the issue. My client is an enterprise GCP user paying a hefty sum every month and getting someone knowledgeable typically means wading through Gemini or talking to (non-technical) support who just read GCP documentation back to you; and it can take months something to get resolved. Now imagine the situation for someone paying nothing or a double-digit for their gmail account or workspace.
I'm aware people will complain that self-hosting is hard, insecure, etc. but it's like a backup: an annoying process you truly appreciate when shit hits the fan.
This is highly debatable. They look better, but whenever I inspect the walls and see what is holding the house together, I just shrug and leave my contact info, because I know I’ll be called to fix the mess.
> ...and building them increasingly feels like magic.
Engineering should not feel like magic to those with a proper education. Otherwise, we are merely practicing alchemy.
Why is that ironic? Since Mimir is open-source, $0 revenue users are expected. AFAIK, Grafana Labs relies heavily on go, typescript, and linux, without necessarily being their top financial contributor. They could have kept Mimir proprietary like Splunk, but whether that would have attracted the same level of adoption or community contribution is another matter.
Exactly! For instance, we had pull-based monitoring 20 years ago (Zabbix et al), but we abandoned it because it scaled poorly, favoring push-based agents (for InfluxDB, KairosDB etc). Now Prometheus is all the rage, yet we’re hitting the exact same scaling walls these systems had before. In a few years, we’ll rediscover push agents and call them the best thing since sliced bread.
Modern IT has become a ubiquitous commodity, much like the car. You don't need to know how an engine works to drive; while that knowledge might make you more efficient, it isn't strictly necessary to get from A to B. Besides, most twenty-two-year-olds ten years ago didn't know how to use ssh, either.
However, if you want to call yourself an engineer (and work in the field), you must understand the underlying mechanics. IMHO if you want to defeat a competitor today, you don’t need industrial espionage - you just have to cut their internet and/or AI subscriptions. Modern vibe engineers would struggle to function.
> The man page is dead for most users. The RFC is unread by most developers who depend on the protocols it describes.
Well, those who are accustomed to using man pages still use them today. I find them far more accurate than whatever an AI might spit out at any given moment. As for RFCs, they were always read by a small population - either those implementing the protocols or the few of us who like to brag about obscure technical details.
> You can now write complete programs without understanding what a single line of them does... until something goes wrong in production at two in the morning and you are completely without tools to respond.
I’m not worried about this. When things go south, there will still be experts who will know how to fix them. But since those experts will be fewer and farther between, they will likely charge $1k/hr, and rightfully so. If you are in that field, more power to you! :D
[1] https://orgmode.org/worg/worg-about.html - Worg is a collaborative knowledge database about Org (Emacs org-mode).
"I’m not a mechanical engineer, but I watched a five-minute YouTube video on how a diesel engine works, so I can tell you that mechanical engineering is a solved problem."
And it’s worth reiterating that most (all) of these LLM/AI providers are currently operating at significant losses. If they aim to become even modestly profitable, prices will have to increase substantially.
Dollar dominance gives the US disproportionate leverage over global finance and allows it to shape the rules of the system. Absent this asymmetry, it is difficult to imagine US tariffs or financial pressure (or any kind of pressure) would carry comparable global impact.
Also, this is a shared server, not a truly dedicated one like you’d get with bare-metal providers. So, calling it "Metal" might be misleading marketing trick, but if you want someone to always blame and don’t mind overpaying for that comfort, then the managed option might be the right thing.
It depends on the kind of investors you want to attract for your startup. Some investors are interested in building a slow-growth, long-term, stable company, while others prefer a higher-risk approach and will expect fast results. IMHO choosing the right kind of capital is just as important as focusing on the product.
Now try opening it in Emacs with vlf [1] ;) Great work overall — looking forward to seeing further development!
From the ping output, I can see HN is using m5hosting.com. This is why HN was up yesterday, even though everything on CF was down.
> Writing high-throughput web applications is easier than ever. Hosting them on the open web is harder than ever.
Writing proper high-throughput applications was never easy and will never be. It is a little bit easier because we have highly optimized tools like nginx or nodejs so we can offset critical parts. And hosting is "harder than ever" if you complicate the matter, which is a quite common pattern these days. I saw people running monstrosities to serve some html & js in the name of redundancy. You'd be surprised how much a single bare-metal (hell, even a proper VM from DigitalOcean or Vultr) can handle.
gcc? It’s hard to imagine any of the projects mentioned without a good compiler.
IMHO it adds, but only if you are big enough. Netflix level. At that level, you go and dine with Bezos and negotiate a massive discount. For anyone else, I’d genuinely love to see the numbers that prove otherwise.
> There's zero personal responsibility
Unfortunately, this seems to be the unspoken mantra of modern IT management. Nobody wants to be directly accountable for anything, yet everyone wants to have their fingerprints on everything. A paradox of collaboration without ownership.
From the article: If you live in the United States today, and you accidentally knock a hole in your wall, it’s probably cheaper to buy a flatscreen TV and stick it in front of the hole, compared to hiring a handyman to fix your drywall.
Probably because the US has been focused on services for years rather than physical goods production. Everything else in US is focused on importing cheap(er) goods or materials.
> On the other hand, I think he wants to push the narrative that AI is seeing enormous productivity gains.
That is my impression as well. I would be thrilled to see this mythical 10x productivity. Even with 2x productivity, I would be highly pleased. This should mean developers (and everyone else) are producing 2x more quality, software (and general services) are 2x better? I see none of that, except 2x more junk. Did AWS, GCP, or anything else become 2x cheaper and 2x more stable? Maybe I'm living under a rock.
You'd be surprised that the most stable setups today are run this way. The problem is that this way it's hard to attract investors; they'll assume you are running on old or outdated tech. Everything should be serverless, agentic and, at least on paper, hyperscalable, because that sells further.
> Today at AWS, it is easily possible for people to spend a multiple of the cost of that hardware setup every month for far less compute power and storage.
That is actually the goal of hyperscalers: they are charging you premium for way inferior results. Also, the article stated a very cold truth: "every engineer wants a fashionable CV that will help her get the next job" and you won't definitely get a job if you said: "I moved everything from AWS and put it behind haproxy on one bare-metal box for $100/mo infra bill".