I have no doubt that most of the power and capability of LLMs will be taken away from us because they allow people like me to do things like the below. Before, I just wouldn’t have bothered because it would’ve been a thousand hours of work.
But now, it’s 20 hours of work and four million tokens.
Do what you can while you can. Soon it’ll be very restricted, I already know.
I and the LLM made an extension that will piss off devs, designers, and everyone in that space who wants to control what I can do, but that helps me enormously. And it took a lot of work and research. The LLM did all the heavy lifting, really, but I did the background research to see what was possible.
The extension restores true back button functionality even on sites that don’t allow it and even those that actively attempt to prevent it. I burned a fuckload of tokens doing this, brothers and sisters. But it works pretty well now for something that is not supposed to even be possible. It is possible, of course. Anything is with the will and the tools.
I’d made an earlier version of this extension but it was honestly pretty crap. It broke tons of sites, was slow, and just in general was not worth using. The new one actually functions correctly. Basically, it snapshots the page and stuffs that into local storage and then when you slap that back button, you get the snapshot of the page exactly as it was before, transparently. And, people, it’s so much faster than the worthless fucking redraw that every page insists on doing now.
No, I will not release it generally. It’s just for me. And my partner, if she wants it. I simply refuse to be anyone else’s tech support outside of my actual job.
This extension rules. It is a middle finger to all those who attempt to control my computer against my wishes and intentions.
Fuck you, I won’t do what you tell me.
I ws looking at Navidrome for personal music streaming but then saw this.
That’s a dealbreaker. This software is written by younger Millennials and Gen Z who don’t know what a fucking directory is. Despite their clownish ignorance a folder structure is the best way to handle a lot of data and probably always will be. Not all music is tagged or tagged accurately. Putting it in a damn folder is the quick and dirty way to organize it easily.
You know tons of people have asked for it if they put disclaimers all over the place about why they won’t do it. Because they are fucking idiots. I logged into GitHub to complain about it to them. I know it won’t help, but it made me feel better.
Castro Podcasts — Things I got wrong: Support.
Castro is an opinionated app and I’ve thought a lot about what we’re building and what we’re going to work on next. It’s unlikely I’m going to implement the request. If I did, by catering more to persnickety power users, we run the risk of alienating newer users who don’t know how things work. But our power users probably aren’t going anywhere, at least they’re a little harder to shake, and alienating new users is the death of the product.
This is exactly backwards. But I think the piece accidentally reveals a fundamental misunderstanding that is very common to people who create software. The reality is that your power users are the ones who funnel new users to to you. Firefox never would’ve taken off without its power users; I personally got hundreds of people to use Firefox, many of whom never would’ve heard of it at all without my influence.
When they decided to kick their power users in the teeth for largely made-up post-hoc manufactured bullshit reasons is when that software began its long usage decline. The Mozilla clowns blame the power users (at least in part) for this, but in reality it’s those users who told them what would happen if they made the bad decisions they did. And they were right, as I’ve pointed out before.
Dustin’s counterproductive reaction — as it almost always is — is a power and pride thing. Power users are often completely correct about what would make your product better for everyone. But that rubs many creators the wrong way because it means more work, and someone else having at least some measure of de facto control, and that someone saw something that you did not. It means the software becomes less yours if you let someone tell you what would in fact work better for them. And many people (such as the Firefox devs) just enjoy having control over others.
I have nearly no idea what Castro is or what it does. It seems to be about podcasts, which means I’d never touch it. But Dustin’s contentions really misunderstand what power users offer in a software ecosystem, and what it means to alienate them.
For instance, I can tell that I’d never use Castro at all even if I were into podcasts because any creator who makes a point to shit on power users is one whose product I’d not even consider in the first place.
About 60% of the electrical power we use is due to all the compute, storage and networking gear we have in our home. That is not an exact tally; it’s possible it’s more than that since that number is a conservative estimate.
That’s what happens when you get two computer nerds together. We have more sophisticated infrastructure (and more of it) than many medium-sized businesses.
And I wouldn’t have it any other way.
TrueNAS is tons faster than the Synology — partly due to that it’s just on much newer, quicker hardware — but in many ways it’s amateur hour. And I want to punch whoever thought it should be so locked down right in the chigooblies. This isn’t consumer-class infra. Gramma is not going to send her money to Nigeria if you don’t prevent me from monitoring fans. And all because I don’t have a BMC?
Yeah, fuck you, buddy.
I don’t know that there is anything better than TrueNAS, alas. But it is free, and that’s something.
Containers, especially the Linux (as opposed to the FreeBSD) variant, are inherently less secure by nature than virtual machines.
Also, FYI, we sysadmin types can spin up VMs just as quickly as containers and have been able to do so for 20+ years. It’s just devs didn’t know anything about it and couldn’t control it, so they reinvented something we could already do — but poorly and very insecurely.
For the first time an LLM — Claude Opus 4.8 on High — solved a technical problem I would’ve never been able to solve. At least not in a reasonable amount of time. And I’m very, very good.
The problem manifested as I was installing Veeam Community Edition on an air-gapped Hyper-V server. The installer kept complaining it could not continue because it was unable find/validate certificates. So I researched and (get this) from the vendor’s own site I downloaded the missing certificates they claimed were needed. Installed them in the correct place in Windows Server. The Veeam install still failed after that.
As happens all too often, their own support page was incorrect and the certificate actually needed was not the one they had you download. Again, vendor’s own site, vendor’s own support pages. Totally incorrect.
Here is Claude’s summary of what was wrong and the fix:
“The certificate was in the box the whole time
Air-gapped a Windows box, then tried to install Veeam Community Edition. It refused with an Authenticode error: CERT_E_CHAINING, “a certificate chain could not be built to a trusted root authority.”
The cause is what air-gapping actually does. A connected Windows box doesn’t ship knowing every root it trusts; it fetches missing intermediates on demand and refreshes its trusted roots from Microsoft in the background. Cut the wire and both stop silently. The trust store just freezes at install-day state and starts declining signatures without telling you why.
I chased the roots first, including the ones Veeam lists on their own site. No luck: Veeam moved their code signing to an Entrust EV cert and never updated the page, so the root their installer needs isn’t on the list of roots they tell you to install.
The fix was to stop trusting the documentation and read the binary. An Authenticode signature embeds its own certificate chain, roots included — two here, one for the signature (Entrust) and one for the timestamp (DigiCert), each anchored to a root the air-gapped box had never seen. Pull them out of the file, drop them in the machine’s root store, done. No internet, no vendor KB.
The machine had everything it needed to verify that installer. It just wanted to phone a friend first, and I’d taken the phone away.”
More fake bullshit-ass security causing endless problems that makes you 0% more secure. Just clownish dipshittery. Veeam used to be good.
Why, Microsoft. Why must you be like this. Who hurt you.
The thing is, though Graph is terrible, I am sure its replacement will be far worse.
Graph is very developer-y in that is extremely, absurdly verbose and also ridiculously complex. I’ve had similar experiences to the above. What used to be single-line commands are now 30-50 lines.
I was forced by MSOL deprecation to rewrite a Powershell script that I used rather commonly. It was around 100 lines long. It’s now 2,000 lines of Graph slop. And it’s not even as functional as the old one as actual capabilities were removed.
I do not understand why insane complexity is so appealing to dev types and I guess I never will.
Correct. AI is the most substantive tech since at least GUIs became the primary way of interacting with a computer. Or perhaps since computers were widely adopted by the masses in the 1990s.
Anyone who claims anything else is just wrong in every way it is possible to be wrong.
I am extremely lucky in that using AI effectively requires one to be intelligent in exactly the ways that I am most perspicacious and effective (understanding large-scale data quickly, holding a lot in mind at once), while AI does not require or even penalizes the ways that I am not (math, complex syntax).
Got lucky, I guess. Other people in related fields are not gonna have that great a time, though. That is already quite clear.
What job interviews taught me about Kubernetes.
This article is wrong. I’ll explain why below.
First, Kubernetes really sucks. It’s just terrible, terrible technology. The networking stack is atrocious. The Kubeclowns took something that should be a routing table and an iptables rule someone could actually read and buried it under three layers of abstraction that all lie to you in different, incompatible ways. And the networking stuff in there is just so slow. Of course it is — it was created by programmers who understand networking about as well as a llama grasps astrophysics.
I could easily write a 5,000-word rant about how shitty Kubernetes is, and how poorly-designed, about how it incompetently relocates abstractions to inappropriate places that make anything 1,000 times harder to troubleshoot. YAML is also a horrid config language that is ridiculously ill-suited for this purpose.
There are so very many insane and terrible decisions in the architecture of Kubernetes. It is one of the worst technologies I’ve ever used. It’s designed by idiots and a suitable choice only for doofuses.
Back to the main theme, though. Kubernetes became stylish to use and that explains part of its adoption.
But the main reason the article is wrong and that it doesn’t understand what is happening with all the Kubernetes BS is that it ignores the power shift that occurred over the last decade or so in the IT space. The dominance shifted from sysadmin types to developers. And Kubernetes is very friendly to developers. It allows them to package and ship broken nonsense with very little real testing. Of course, devs love that because who wants to test, amirite?
Also, devs are inherently very attracted to complexity. Kubernetes is exceedingly complex so this causes them to love it. It has endless (and pointless) knobs to turn, sliders to move around, switches to clickity-clack, endless dials to twiddle and tweak. All worthless and counterproductive, but devs absolutely go gaga over crap like that.
As I said, Kubernetes won because developers are now dominant. It allowed them to ignore security, good networking practices, compliance, and push absolute garbage to production “reproducibly1.” It’s trash technology2 that achieved preeminence because it allowed devs to run amok in areas they did not understand and hated to deal with.
And yes, I use Kubernetes every day in production. Not my choice, but I am not just guessing about how much of a crapfest it is. I live it.
I’ve been running TrueNAS Scale on the new NAS, the one that is home-built (not from a vendor). And I don’t like TrueNAS itself much. It’s yet another power user tool that has removed or gimped its most important and useful features for no apparent reason.
For instance, they’ve removed the recycle bin on SMB so that “Previous Versions” in Windows no longer works. I don’t even run Windows at home on my daily use endpoints, but that’s a real loss. And moronic. Their “solution” is to use ZFS snapshots, but this is a totally different tech that is not as easy and is far more dangerous.
TrueNAS also doesn’t handle certain files generated on the Mac correctly — specifically ones with Alternate Data Stream stuff going on — while commercial NASes seem to deal with these files fine. And no, the supposed common fix did not correct the issue.
The GUI is also pretty bad. It manages to be both too simple and too confusing at the same time somehow, and is missing many features and capabilities (many of which did exist previously and were removed for ideological reasons).
Of the major NASes and their OSes/GUIs I’ve used recently, I’d give QNAP’s a 7, Synology’s an 8.5, and TrueNAS a 4. It feels amateurish and like something I used in the early 2000s.
I will keep using it for now but it’s pretty meh.
Everyone just seems so dogmatically dumb about too many things. It’s why I can’t help but be so insulting.
People claim that there is no future in which AI models improve and that we get better at building the underlying infrastructure supporting them. However, right now, AI runs mostly on chips that were designed to play Half Life and Final Fantasy in data centers built to provide a place for an SAP or email server to operate. And the frontier models are about as optimized as a 1979 Chevy Pinto.
To think all of that won’t get better…fucking ludicrous. Are you even thinking? Are you even trying to think?
The difference between acting in every way like you understand something and actually understanding something is far less than people think or want to believe.
As AI approaches “pretending” to understand something with near-perfect fidelity to actually comprehending it, the collapse of certitude around what “understanding” means will cause people to hate AI even more. Fear will overtake all, and dread.