> I'm based in Russia...
This would be an immediate disqualification for any tech job I can think of.
405 karma · joined April 20, 2013
This would be an immediate disqualification for any tech job I can think of.
When I moved into my current home 22 years ago, I bought five Hunter ceiling fans from Home Depot and installed them in the bedrooms and living room. Each one cost maybe $100 or so. They've been used more or less daily since then (the one in the master bedroom runs at medium speed all night, every night), and they're still silent and wobble-free. Other than wiping dust off of the blades from time to time, I've done precisely nothing to them.
They don't have mobile apps (there was no such thing as a "mobile app" when I bought them) and they don't have remote controls, but they're pretty great.
The vast majority of people are not good prose writers. Nor are LLMs.
Human test passed. AIs make giant flashy mistakes. They don't fuck up basic grammar.
Because it tends to require a lot of thought and effort along totally separate engineering pathways than most of the actual functional coding. Good UI design is not easy, and the skills to do it don't directly overlap with good coding skills. It's hard.
> I suspect part of the reason is that most gui frameworks are not really designed for making dense economical interfaces.
Don't blame the tooling for a broad failing of many developers. It's because constructing a genuinely dense and economical interface for most apps requires considerable time and effort, and frankly, most developers are crap at understanding how users other than themselves use applications. Those that aren't crap at it tend to have worked in large teams with dedicated UX people to help drag them away from their preconceptions.
If that doesn't bother you, then good for you, and we have different feelings about privacy.
No need—the current ones are uploaded directly into my mailbox regularly by Randy Weber, my district rep.
In other words, they are using gas turbines because the politicians are stupid.
Somebody hasn't read the docs and doesn't understand how nginx's "if" works, because it's fantastically useful.
The cert itself only has CN=cloudflare.pay. It lacks an org, an address, or any other identifying info. It's not OV/EV, so no details there, either.
The domain's whois is also devoid of identifying details:
https://rdap.nominet.uk/pay/domain/cloudflare.pay
Registered through 101domain, with nothing except a registrar abuse contact.
I mean, great that this is legit, but CF could have done a better job with making it actually _look_ legit. This looks sketchy as fuck.
edit - gawd, nevermind. they don't even have anything useful for cloudflare.com. Same GTS cert, redacted whois info. lol. how did we even get here.
I imagine it's somewhat the same boiling rage I feel when I watch youtube videos with obviously AI-written scripts, or read obviously AI-created written content in any area. I can't code, but do write and edit professionally, and the lackadaisical approach many people take with the form and presentation of their written words can be pretty damn offensive. (The responses I hear when I raise these objections are more or less the same ones I raise re: coding, though — "Well, I can't write, and the LLM says what I want to say way better than I could ever say it!" and so on.)
So it goes, I guess, heh.
FWIW, Conde has no input on my work. Conde is the machine that provides HR and legal and payroll. Ars is a more or less self-contained world and since I got kicked into management a few years back, my output is my own. I wrote this up because this is how the project unfolded; no one assigned it to me or said I needed to write a July story to make a quota. It was fun. I did the write-up because it felt like an Ars-y thing to write up.
Definitely heard re: wanting to stumble through things and see the warts, but I have to disagree on the "learning enough python to make this work would still have been a weekend project." That's simply not true—if it were, I would have done it.
But it's not like the project didn't require effort. I put in considerable time on the parts that weren't python coding and CAD. Planning out and setting up the systemd service and timers, packaging up the service to ensure the thing was running under a properly isolated service account context, writing the installer, setting up the deployment pipeline—I didn't want to do the usual lazy thing I see so often and just take a barely functional mess of conflicting dependencies and shove them into a docker container and say "Here, this works on my computer, your problem now."
But like I said elsewhere in the thread, at 48 I'm old enough and ossified enough that I've come to accept that coding is simply not a skill I've got in me, and cranking out even something relatively simple like this project is going to be impossible for me without help. I've tried, too—I've bootstrapped myself through code academies, tried DIY courses, thrown myself into code-adjacent weekend projects, and otherwise tried to get to some level of mastery past "hello world," and I just kinda....suck. I bounced off of coding 30 years ago in college, I've bounced off it professionally for anything more than simple bash scripting, and I've bounced off of it in hobbies. I don't know if it's that my english major brain just doesn't do well in chopping problems down into small enough chunks or what—but I know after a lifetime that it's something I'm never going to be able to do.
That and bowling. Never scored over 100 in a game in my life, and likely at this point never will. Some things are just beyond my reach.
This was my thinking as well, and why I rejected the idea. I'm not installing some rando company's information-harvesting data-selling garbage app on my phone to do something that the clock should do itself.
I don't sound like LLMs. LLMs sound like _me,_ lol.
acme.sh --install-cert -d grafana.tuxnet.dev --key-file /etc/ssl/private/grafana.tuxnet.dev.key --fullchain-file /etc/ssl/certs/grafana.tuxnet.dev.crt --reloadcmd "systemctl reload nginx"
Then you can ditch your custom cron and let acme handle everything on its own, as intended.
edit - after actually reading the tweets (had to use xcancel) and visiting the source git repo, switching to MTP for speculative decode makes things a hell of a lot faster, and the abliterated model plus dflash makes it even faster! I'm now seeing 70-90 tok/sec for most stuff. I like!
But it's also really easy to trip up. I fed it some of my Ars pieces and asked it to analyze themes and composition, and it got into a looping argument with me over how it was unable to analyze "my" writing because "the user cannot be the article author, the user is the user, the user did not write the article, the article author wrote the article." I was utterly unable to convince it that I was in fact me.
Qwen3.6-35B-A3B hums along at about 50GB of RAM used with --gpu-memory-utilization=0.42. I haven't tried Qwen3.6-27B (I'd likely grab Qwen3.6-27B-FP8, I think), but I'm curious to see if it makes much of a difference.
As someone who's been playing since alpha 1.2.0 starting back in october 2010, there has never been a moment in minecraft's entire history where the game hasn't been utterly fucked with insane bugs. This is status quo.