I'm currently polishing one of my packages for publication, and for that do a lot of testing in a vanilla Emacs. It's shocking how many of even (to me) absolute core functionality is actually customization I did a decade or two ago, and then forgot about.
And the yet again I got told over and over again that Bambu didn't really mean to, and if they did they learned their lesson, and after all you can still keep them offline. And spending more for a prusa obviously is silly.
I'm really getting too old dealing with morons who didn't learn anything after the same patterns repeating for decades now.
I assumed he just had DNS at AWS, but after re-reading I guess he has DNS _and_ domain registrations at AWS, which would be a special kind of stupid. That's something we were advising customers against already back when cloud wasn't a thing yet to enable fast transfers when stuff goes south
(to clarify: DNS+domain at the same service can be OK, as long as you have nothing else there. As soon as you start having other stuff, keep the DNS there, but move the domain registration away. Depending on which domain make sure you have auth keys, access to the admin domain or whatever would enable moving the domain without registrar cooperation. In my hosting days I did my fair share of emergency transfers and infrastructure to help companies get their basics online again after a SNAFU - totally doable to have first mail coming in again within a working day)
I've been in a similar situation - a surprising amount of companies really just click to create instances. Last time I've encountered that at a customer I improved things a bit by creating templates, and scripting instance creation based on those templates - but ideally we'd have had the templates themselves as well as the network side generated by ansible.
But that's the problem: The complexity of doing that properly is pretty much the same as just doing your own hardware (which is what I'm working with most of the time - handling stuff on physical servers). And at that point the question should be why you're paying AWS so much money and pay your people to automate AWS workflows when you could just pay them to automate workflows on physical hardware, which would be way cheaper to run than the AWS instances.
This is always the weird things in those rants. He's complaining that after 4 days his mails are offline.
Now I'm doing a mix of physical servers in rented rackspace, and rented servers - but even there I can have billing mixups where they deactivate servers for no good reason. And to get email working again the limiting factor would be the DNS TTL - new servers would be online somewhere else within hours of it going down. (And yes, I tested that just last year - one hoster threatened cutoff due to non-payment on a paid invoice, which prompted me to move the mail server just in case while getting this resolved).
I've found a lot of issues come through somewhat naive networking setup - which is encouraged by the "just yolo it" installation instruction in the documentation. If you want to start understanding what's going on you'll end up in very weird corners very quickly. Also, if you don't want the API endpoints available to the world the documentation is not much help.
I've found things more stable if you can give a dedicated interface just for internal k3s communication. It can be a bridge interface on top of a vlan interface - but not the vlan interface itself, or some things will break in very interesting ways. Also, even when using IPv6, just stick with internal IPs and nat everything - touching internal IP ranges is no fun. Plus, if there's a chance you'd ever want to use dual stack, set it up with internal v6 addresses, and just don't use the v6 addresses for now. There's also a lot of unintuitive behaviour around dual stack networking - and lots of areas where documentation is just plain wrong.
I'm scripting our stuff with ansible - one of the more useful things was the realisation that in some areas changes which shouldn't break anything can lead to cluster communication being interrupted, which is a very interesting thing to deal with, especially when you can't pin it to that change that didn't touch anything close to that, and therefore should not be responsible. I've learned, and sprinkled checks to make sure all members can still reach each other in there now, so that at least when I break it on changes I directly know why.
Back in the early 90s I've discovered that there's not really any backchannel happening when printing to a dot matrix printer over the parallel port. And adding multiple ISA parallel port cards set to same IO address wouldn't cause any hardware issues, but just spit out the same data on all cards.
Which meant, as long as no paper was jamming, I could send data all three printers I had access to at the time could understand, and would save 1/3rd of printing time by having everything spit out in parallel.
My mother (parents bedroom adjacent to mine) did not think that'd qualify as ASMR at 2 in the morning.
I didn't read much of the page - just was scrolling a bit to see what the fuck that thing is doing, and that was more than enough to know that I'll never touch whatever those people are doing.
> On the one hand, moving from assembly language to C made programmers less effective in some ways and more effective in others. On the other hand, the transition from writing code by hand to using AI is arguably a bigger shift,
I think that's a bad comparison, mainly for two reasons:
We still have people dealing with assembly for building the compilers. We don't quite have that for AI - which became very clear with the claude code leak: We just have people trying to get the correct behaviour, but without understanding why.
The second reason has more practical implications: I generally rely on my compiler to be deterministic, and not introduce security issues (which they sometimes do, by optimising away safeguards placed by the programmer - but that is relatively rare, and we can trace and fix that). But generally as a developer I can rely on my compiler to produce machine or byte code I don't have to think about.
The same is not true for AI - the regularly produce insecure code. This can partially be mitigated by having another AI review it - but it's not a proper solution. Until we get proper AI which can actually understand things we need somebody in the loop who can understand the code.
I'd recommend for every developer to get one or more colourblind friends. I have some, and regularly send them screenshots of what I'm working on to get feedback what they can see and what they can read/can't read.
They've been absolutely invaluable for making sure their kind of people can't use my apps properly.
For a lot of tasks smaller models work fine, though. Nowadays the problem is less model quality/speed, but more that it's a bit annoying to mix it in one workflow, with easy switching.
I'm currently making an effort to switch to local for stuff that can be local - initially stand alone tasks, longer term a nice harness for mixing. One example would be OCR/image description - I have hooks from dired to throw an image to local translategemma 27b which extracts the text, translates it to english, as necessary, adds a picture description, and - if it feels like - extra context. Works perfectly fine on my macbook.
Another example would be generating documentation - local qwen3 coder with a 256k context window does a great job at going through a codebase to check what is and isn't documented, and prepare a draft. I still replace pretty much all of the text - but it's good at collecting the technical details.
> Even if everyone eats a perfect portion for themselves, if you overestimate the total then you'll have food waste if the food can't be preserved.
That'd be just poor planning on part of the hotel/restaurant. It'd be a valid excuse when starting new, but after a few weeks that should be under control.
If you only do breakfast buffets it's a bit harder - but you monitor the situation, and as breakfast time approaches the end you reduce things you can't store or re-use otherwise. Pretty much any hotel I've been to in the last few years had that kind of items run out without restocking them when we had a late breakfast.
If you also do lunch/dinner buffets you have some more options, and can have some dishes reusing the leftovers. I've also seen that regularly - they had the planned dishes, and a few smaller pots with something they came up with to reuse whatever was left over.
How is that surprising? We've been taking that into account for any LLM related tooling for over a year now that we either can drop it, or have it designed in a way that we can switch to a selfhosted model when throwing money at hardware would pay for itself quickly.
It's just another instance of cloud dependency, and people should've learned something from that over the last two decades.
> Or, maybe, don’t: when people do, they take much more than they eat. Compared with ordering from the menu, all-you-can-eat breakfasts waste more food—up to twice as much, according to one study.
Is that a cultural thing? We have pretty much zero food waste on any buffet as you can easily only take what you actually want to eat. It's just basic good education to be considerate with resources, especially food resources - and I rarely see people taking more than they actually eat, so it's not just an "our family" thing. If you do throw away a lot of foot on a buffet you're just an inconsiderate asshole - and if a restaurant location has significant food waste from that they should just start charging for leftovers.
Yubikey (and nitrokey and other HSMs) are technically smart cards, which perform crypto operations on the card. This can be an issue when doing lots of operations, as the interface is quite slow.
Anything pkcs#11 you can proxy. I'm using that on some systems - I have an old notebook with a nitrokey hsm at home. It binds pkcs11-proxy to a local wireguard interface, so I'm registering systems I want to be able to use those keys to that notebooks wireguard. They still need a pin for unlocking a session as well.