For the hundreds of people reading this article right now - you might be amused to know that you're accessing it from a mac mini on my desk:
https://www.contraption.co/a-mini-data-center/
(The CPU load from this is pretty negligible).
https://www.contraption.co/a-mini-data-center/
(The CPU load from this is pretty negligible).
What else am I missing?
>What else am I missing?
You are missing Reddit.
https://en.wikibooks.org/wiki/Demystifying_Depression
Still holds up well two decades later.
anyone?
Of course back then we didn't use dynamic anything, a static web page worked.
Now get off my lawn!
And it wasn't static: We because peoples connections were mostly so slow, we used a CGI that shelled out to ping to estimate connection speed, and return either a static image (if you were on a dialup) or a fancy animated gif if you were on anything faster.
(the ping-test was obviously not reliable - if you were visiting from somewhere with high latency, you'd get the low bandwidth version too, no matter how high your throughput was - but that was rare enough; it worked surprisingly well)
Thousands and thousands of lines of quality conversation, interaction, humanity.
To be honest, I kind of miss those days.
I love to think that the web of the future is just going to be everyones' mac Mini or whatever.
Big Data™ has always irked me, frankly.
I started thinking about that around an implementation that could run under Google’s App Engine free tier, but never completed it.
/s
In most cases businesses justify the cost of managed databases for less risk of downtime. A HA postgres server on crunchy can cost over $500/mo for a measly 4vCPU.
I would agree that it's the least of concerns but for a different reason. Spending all your time optimizing for optimal performance (assuming sensible indexing for what you have) by continuously redesigning your DB structure when you don't even know what your company will be doing next year isn't worth the time for a few hundred a month you might save.
I read this as "in building your startup, you should be paranoid about team members never making mistakes". I really try to read otherwise, but can't.
It's a cool project. I might do it too. I have an M4 Mini sitting on my desk that I got for $550.
If you're considering k8s, take a look at Kamal (also from DHH): https://kamal-deploy.org/
I think it makes more sense for small clusters.
I don’t find it approaching unruliness, quite the opposite really.
Why did you go for Cloudfare tunnel instead of wireguard?
We also hosted web pages before the average script kiddie could run tens of Gbps DDoS on sites for the lolz. And before ISPs used CGNAT making direct inbound connections impossible.
Even having IPv6 is not a proper solution because of laggy ISPs(currently reaching ~50%) and the even the ISPs who deploy, do not deploy it properly. (dynamic prefixes or inbound blocked IPv6)
Add to the mix that lot of people does not understand IPv6, internet became more centralized and will keep doing so for the foreseeable future.
I wrote a visualizer for the traffic that I think people will appreciate [1]. I will post it next month once I add it on github. It was fun to watch an article that went #1 on HN.
Render is crazy expensive for blog sites and hobby apps.
https://gist.github.com/philipithomas/ed57890dc2f928658d2c90...
I run a windows server at my office where we connect to it using RDP from multiple locations. If I could instead buy the hardware and use cloudflare tunnels to let my team RDP to it then it would save me a lot of money. I could recoup my hardware cost in less than a year. Would this be possible?
(I wouldn't mind paying for cloudflare tunnels / zero trust. It just should be much smaller than the monthly payment I make to Microsoft)
I'm pretty sure Tunnels supports RDP and if you don't use a ton of bandwidth (probably under a 1TB/mo), Cloudflare probably won't bother you.
I do think Cloudflare has proven itself to be very developer/indie-friendly. One of the only tech unicorns that really doesn't impose its morality on customers.
My shorter-term goal is to switch my home internet to Starlink, so that all requests bounce off a satellite before landing at my desk.
I understand that the USA is big, but no fiber in SF?
No need for the hyperbole. I know for a fact that you don't get fiber in the remote countryside of France
AT&T has put fiber out in little patches, but only in deals with a guaranteed immediate ROI, so it would mean brand new buildings, where they know everyone will sign up, or deals like my old apartment, where they got their service included in the HOA fee, so 100% adoption rate guaranteed! AT&T loves not competing for business.
Sure, others have been able to painstakingly roll out fiber in some places, but it costs millions of dollars to string fiber on each street and to get it to buildings.
Caught an AT&T tech in the field one day, and he claimed that if 8 (or 10—memory's a little fuzzy) people in the neighborhood requested it, they'd bring it in.
I never did test it, but thought it interesting that they'd do it for that low a number. Of course, it may have been because it was already in the area.
Still, may be worth the ask for those who don't already have it.
I’ve served (much) greater-than-HN traffic from a machine probably weaker than that mini. A good bit of it dynamic. You just gotta let actual web servers (apache2 in that case) serve real files as much as possible, and use memory cache to keep db load under control.
I’m not even that good. Sites fall over largely because nobody even tried to make them efficient.
I’ve had a number of database-driven sites hosted on $5/month VPS that have been on the front page here with minimal cpu or memory load.
https://developers.cloudflare.com/ssl/reference/certificate-...
The CPU is more powerful than I need. I didn't pass 8% utilization in the last 24 hours as HN slammed the box.
I had 2m35s of downtime due to power outages this week.
Not only does is have a built in UPS, but also comes with a screen, keyboard and trackpad for you need to do admin tasks physically att the console!
Although you could use a backup USB cellular modem, plugged directly into the Mac!
At some I imagine I'll inherit or acquire a cheap older M1 Macbook Air, which will be perfect!
I also have whole home battery backup and solar, so I technically have a UPS for everything!
Where I live the power tends to go out from time to time!
I monkeyed around with cheap ex-lease Dell Micro PCs with Intel 8th Gen to 11th Gen CPUs. But they not as performant as I'd like, and once you've experienced modern CPUs like Apple's M series, you dont really want to go back!
Now, however, what is the point? To learn server config? I am running my blog with GitHub pages. A couple of posts made it to the top of HN, and I never had to worry.
Always bewilders me when some sites here go down under load. I mean, where are they hosting it that a static page in 2020s has performance issues?
In fact, most of them are just implemented by feeding an LLM the text, and asking "is it AI generated?". You cannot trust that answer any more than any other LLM hallucination. LLMs don't have a magic ability to recognise their own output.
Even if your "detection tool" was using exactly the same model, at the same exact version... unless the generation was done with 0 temperature, you just wouldn't be able to confirm that the tool would actually generate the same text that you suspect of being LLM-generated. And even then, you'd need to know exactly the input tokens (including the prompt) used.
Currently, the only solution is watermarking, like what Deepmind created:
https://deepmind.google/discover/blog/watermarking-ai-genera...
but even that, it requires cooperation from all the LLM vendors. There's always going to be one (maybe self-hosted) LLM out there which won't play ball.
If you're going to accuse someone of pushing LLM-generated content, don't hide behind "computer said so", not without clearly qualifying what kind of detection technique and which "detection tool" you used.
Yes, compared to an all-or-nothing approach, it's better to be upfront about the uncertainty, especially if the tool surfaces the probability by-sentence.
But how are those probabilities computed? You mention gptzero, but https://gptzero.me/technology doesn't clarify at all how it works. They link papers using GPTZero (i.e. from other researchers), e.g. https://arxiv.org/pdf/2310.13606
And these very same papers highlight how everything is still unknown
> Despite their wide use and support of non-English languages, the extent of their zero- shot multilingual and cross-lingual proficiency in detecting MGT remains unknown. The training methodologies, weight parameters, and the spe- cific data used for these detectors remain undis- closed.
GPTZero seems to be better than some of the alternatives, but other discussions here on HN when it was launched highlight all of the false positives and false negatives it yielded:
https://news.ycombinator.com/item?id=34556681
https://news.ycombinator.com/item?id=34859348
But all of that is pretty old, there have been a couple of posts in the last year about it, but both are about the business, rather than the quality of the tool itself.
https://hn.algolia.com/?dateRange=pastYear&page=0&prefix=fal...
So, to check if now it's any better I tried it myself: I got it to yield a false negative (50% human and 50% AI rating, for a text which was wholly AI-generated), and I haven't got it to yield a false positive.
But all of this is just anecdotal evidence, I haven't run a rigorous study.
For sure, if some competent people believe that the tool won't generate false positives, I'll be mindful of it and (In the rare cases in which I write a long posts/blog articles, etc.) I'll check that it doesn't erroneously flag what I write.
It's bittersweet: if a tool that can be relied upon really exist, that would be good news. But if that tool is closed source (just like ChatGPT, Gemini, etc.) that doesn't inspire confidence. What if the closed source detection tool will suddenly start erroneously flagging a subset of human texts which it didn't before?
At least, even with the closed source LLMs, we have a bunch of papers that explain their mechanism. I hope that GPTZero will be more forthcoming about the way it works.
>Today I'm working on Chroma Cloud, designed for exploring and managing large datasets, and Next.js powers its advanced interactions and data loading requirements.
which is unlikely to have been written by an LLM.
And it mostly makes too much sense apart from "most drivers don't know how many gears their car has" which has me thinking huh? It's usually written on the shifter.
You don't have to extrapolate too many iterations before you start arguing that you can't trust you're living in base reality.
My take on both issues is that even if you're right, the choice is whether to spend your life in a state of cynical paranoia, or not.
If you figure out the consciousness equivalent of Ctrl-Shift-Esc, please report back.