This reminds me of when I used to do IT work for small businesses in college. One printing company I worked for, had about 100 computers on their network, and was using public ipv4 addresses, that they did not own, on their internal network. I forget what range they were using now. But imagine seeing a DHCP server handing out addresses like 142.250.110.1/16 on a LAN and the public ip being something totally different.
It was funny, because when I brought it up to them, it was hard to articulate why it was a problem and I couldn't convince them it was worth the effort of trying to fix. They never ran into a specific issue due to this while I was there but it felt so gross.
It's always hard because when you contrive possible examples of how it goes wrong, every single example sounds contrived because of course they are contrived.
Sure one day your printer might start spewing random json code meant for some microservice of the rightful IP owner.
Sure the IP's might be owned by the Air Force and one day they might start getting traffic from your pos ipad that they decide looks like an attempt to attack one of their internal secret networks...
Sure one day traffic meant to go to your printer ends up flooding and dossing a windmill controller, preventing the rightful operators from turning it the right direction during bad weather and causing $25M damage...
And of course the real failures are more like, only people from the Maldives can't send email to your email server, a failure with no impact.
The problem is that they wouldn't be able to talk to any internet service that legitimately used those addresses. Whether that was likely to be a problem very much depends on whose addresses they were.
There's a good reason to do this if you can't be certain what reserved subnets are used in a given network and you absolutely need a static ip for some reason.
At least that's the conclusion I've arrived at at some point, but I don't remember what was the exact use case anymore.
However there are still several global IPv4 ranges that are not local reserved ranges but are effectively reserved and you could use them if you really want to without any issues.
I used to do consulting for companies that weren't connected to the Internet yet. This was the early-mid-1990s and "Private IP Network" assignments literally did not exist until 1996.
Therefore, people who weren't on the Internet yet, or didn't expect to be connected to any particular WAN, or even had their own intranet WAN without public access, they improvised like crazy. Like, whatever IP address ranges worked for them, they just plugged 'em in.
Since it was indeed early, and IPv4 networks were being handed out like candy, the savvy companies just applied for their own network space, hopefully of an appropriate size. The applications at the time didn't specify or require that they route the traffic or put it into BGP, so registering your intranet with IANA was the right thing to do, and future-proof. Unfortunately, in the wild-west early days before CIDR, IANA needed to find either a chunk of class C, or class B without significant nuance. And that is how Apple, HP, and Ford came to control gigantic swaths of "Class A" space.
Renumbering private networks was one of the nightmare scenarios of consulting firms like mine. And it hasn't really stopped, I suppose. If your company does an M/A their private LANs may clash, and they either renumber or they NAT.
By the way, OpenWRT isn't supposed to be using ".lan". That's a ridiculous non-compliant behavior. Someone should've stopped that, long ago.
There are ways to subdomain properly and IETF has provided dedicated TLDs for it even now, but back in the day, it was not easy, and I get it. When I needed DNS for my homelab, I discovered that the best practice among bad practices was to subdomain your ISP's. Like, make up a fictitious subdomain and hang it off their real one. Then your DNS had an anchor in reality, even if it was all made up, never delegated, unreachable, unresolvable from the outside.
When I was working with Linux based kiosks and PoS systems I had a good share of issues with the Microsoft MVP signature use of .local on their forests. Back then their training material recommended .local for Active Directory services.
Specifically, either a registered name or a subdomain of a registered name reserved for this purpose, because the A records for the apex should or must — I don't remember which, but it's definitely the default configuration — point to the domain controllers, and you probably don't want to host your public web site on the same addresses as your DCs (or for that matter, your internal and public records necessarily hosted on the same server).
Historically (20+ years ago), some Microsoft documentation suggested .local as an example of an unregistered domain that could be used for this purpose, which was problematic when .local was subsequently reserved for multicast DNS.
Meanwhile, I watched a very prominent Microsoft consultancy shop install a client's ActiveDirectory as <nameoforganization>.ad. You know, the country of Andorra? That has an actual real ccTLD?
Given that there don't appear to be any BGP announcements for 25.anything, is it possible that the MoD uses this address range internally, or not at all, and has agreed to continue to do so?
I am saying this to be a little snooty. Your take is very dismissive to a large group of people whose experiences you don't know. I agree that split/custom keyboards and vim are awesome; I have used them for 25 years. But the idea that you can reduce RSI treatment and everyone's situation to something so simple just isn't true. It is fine to recommend solutions to people who are suffering, but to dismiss their problems as being "solved" comes across as basically calling them stupid for not helping themselves.
I know you didn't say that, and you softened your language, but many people with disabilities or chronic conditions have to deal with everyone they share their problems with telling them their "cure". Its exhausting.
Thanks for saying this. I already use Vim (in my browser too). I've used it on and off in an editor as well. I've tried reading books (like I mentioned in the post) which is something you only do when you're truly desperate. For some of us, our body will only allow so much.
I am sure the OP was just trying to help. I, too, when I started having such problems chalked it up to "surely some adjustments will help". And they did! But it never quite ceased.
First off, I'm not going for a "gotcha" reply here, I hear all of what you said and I don't see your reply as adversarial.
> but to dismiss their problems as being "solved" comes across as basically calling them stupid for not helping themselves.
Yep, I did kind of say that now that I re-read it. It does come from a point of aggravation (again, in The Before Times) of people expressing, with pride, how little they know of their tools (editor and VCS mainly). And for sure, they probably made/are making a lot more money than I am, so I guess they win?
I must take exception, though, with you bringing up disabilities. This article never mentioned them, so I didn't either. This is a whole other can of worms, but I will say that after working with a couple of people who have pretty extreme disabilities, there are some extremely good (non-"AI"... although of course we know they have some sort of "pre-AI"-AI in them) tools out there are extremely good, and they rocked the f^ck out of them (yes, I intentionally used "extreme" three times there).
So ya, I guess I was being a bit snooty as I was always annoyed how few people learned their tools. Not even just their tools, but their programming language of choice as well! It's all a spectrum, of course. But, as I must admit, it turns out they were right so... ¯\_(ツ)_/¯
Also, as far as exhausting goes, I've never been more exhausted than trying to follow what these awful f^cking companies are doing.
I had a raspberry pi 3a+ and the old raspberry pi touchscreen laying around. So I'm writing a custom dashboard for it using Rust/Slint[1]. It directly uses opengl without any display manager/window system. It mostly talks to home assistant. Right now, when I start a 3d print, it automatically switches to a view with live camera of the printer and some stats. I can monitor the status of my washer/dryer so I know when they are done (using cheap TP-Link Tapo smart plugs[2]). My favorite thing it does, is if any motion is detected on my Ring cameras, it just automatically pulls up that live feed. Took a little work to get decent playback performance but the pi 3a+ has hardware h264 decoding.
Its nice, overall, to have a little dedicated touchscreen on my desk that I can easily tweak to display whatever I want. Its silent and low power.
They also did this for the original steam controller, which I used a lot. When the back panel broke (its a trigger and battery cover) I was able to 3d print a replacement that has held up great.
I really want to know how he solved this problem, which I also face:
>last year i bought a Rodecaster Duo to solve some audio woes to allow myself and my girlfriend to have microphones to our respective computers when gaming together and talking on discord in the same room without any echo
the rodecaster can connect to two computers, and we are both generally in the same discord call. so we have both microphones routed into one input for a computer, and the other person joins with their mic muted and the audio just comes from one client. since the mixing is local there's no echo. email me if you have more questions :)
So both mics will pick up both people (at least somewhat, in the same room) - but because there is no, I assume 20-100ms latency going through the system, to discord, and back - it avoids a slight difference in timing of the two mics picking up the same sound slightly differently. Is that right?
also the audio output of each computer is routed thru the box as well, so i can mix my girlfriend’s computer into her headphones as well as my microphone, so she can hear me with noise canceling headphones, or turn off my microphone if i’m working so she can do stuff without my mic in her ears.
Or if she’s watching a movie or something I can also add her computer audio to my headphones. There’s even a separate audio output for host 1 where you can put ‘chat’ on, like discord on a dedicated interface, so that your application audio is clear and isolated. It’s hella expensive but it really is a great device
Not in the same league or form factor, but I have an old Jabra 65 headset, and the noise canceling is amazing. I can be playing my cello while unmuted on a call, and nobody can hear it.
I know headsets aren't everyone's cup of tea, but a mic close to the source (your mouth) with good noise canceling is a solid solution.
It saves on rewiring stuff. Maybe there's only one person talking today. Maybe they're using PC A, or perhaps they're using PC B instead.
Or maybe there's two people in the room, each on different channels altogether. In this case the other person is just uncorrelated background noise instead of a persistent echo.
Or, in-context: There's two people in the same room, both talking on the same Discord channel.
Anyway, audio routing is useful. Being able to route audio with two different PCs is a pretty neat feature of the rodecaster.
When I read this, I thought that the author would be calculating the derivative of the potential energy with respect to nuclear charge for some reason. Here, the derivative being evaluated appears to be the atomic forces, using jax magic, as opposed to the standard Hellmann-Feynman theorem approach.
Calculating the derivative of the energy with respect to nuclear charge would be fun, as it would let you perform some type of "alchemy" smoothly changing from one element to another. I'm not sure that has any practical use.
Since the gradients are computed with jax, the library can be used to differentiate with respect to all inputs including nuclear charge and basis set parameters (exponents and contraction coefficients). I agree that computing gradients for the nuclear charges could be interesting in the context of molecular design.
But for the colab demo I thought that sticking to nuclear positions (i.e atomic forces) would be easier to visualize.
I think that the audio we are listening too is from some ground recording station that isn't necessarily near the airport. We aren't listening to a recording of what the pilots heard or what air traffic control hears.
Yup. Audio is sourced from volunteers with home radios & antennas. Quality will be dependant on how far the nearest one was from LAX and their personal setup. Not necessarily representative of quality that the controller/pilot was hearing.
Technically correct...and yet let's not kid ourselves that even with a nice radio and a pair of Bose A30's, it can be hard to tease out exactly what's being said when in a congested airspace.
I can't believe how affordable Hetzner is. I just rented a bare metal 48 core AMD EPYC 9454P with 256 GB of ram and two 2 TB NVME ssds for $200/month (or $0.37 per hour). Its hard to directly compare with AWS, but I think its about 10x cheaper.
I play fortnite and marvel rivals with my family. We have lots of fun. I think this genre of game is fantastic if you play with people you know on voice comms. "Solo queuing" in these types of games is not fun for me at all, so I get what you are saying, but they are popular for a reason!
Do you think that you can use those machines for confidential workflows for enterprise use? I'm currently struggling to balance running inference workloads on expensive AWS instances where I can trust that data remains private vs using more inexpensive platforms.
Of course you cannot use these machines "for confidential workflows for enterprise use", at least with AWS you know whose computer you're working with, but also keep in mind that it's really hard to steal your data as long as your data stays in memory, and you use something like mTLS to actually get it in and out of memory via E2EE. You can figure out the rest of your security model along the way, but anything sensitive (i.e. confidential) would surely fall way out of this model.
Just currently exploring how custom AI workflows (e.g. text to sql, custom report generation using private data) can help given the current SOTA. Looking to develop tooling over the next 3-6 months. I'd like to see what we can come up with before dropping $50-100k on hardware.
I threw together a toy project to see if it would help me understand the basic concepts and my takeaway was that, if you can shape your input into something a dedicated classification model (e.g. YOLO for document layout analysis) can work with, you can farm each class out to the most appropriate model.
It turns out that I can run most of the appropriate models on my ancient laptop if I don't mind waiting for the complicated ones to finish. If I do mind, I can just send that part to OpenAI or similar. If your workflow can scale horizontally like my OCR pipeline crap, every box in your shop with RAM >= 16GB might be useful.
Apologies if this is all stuff you're familiar with.
It was funny, because when I brought it up to them, it was hard to articulate why it was a problem and I couldn't convince them it was worth the effort of trying to fix. They never ran into a specific issue due to this while I was there but it felt so gross.