linsomniac

I have tried to forego swap because I had hoped that it would avoid swap thrashing, where performance goes into the toilet; I'd rather a process was killed but I could still access the system, than be unable to login to it.

But I've found that even with no/little swap, the system can get into high memory pressure situations (even easier because it can't swap out unused pages of libraries and the like), and still go into a sort of thrashing state even though it isn't swapping.

My main workstation is a Dell XPS15 with 32GB of RAM and I had 8GB of swap on it. About every month the system load would skyrocket even though it had plenty of mem+cache+swap available. If I killed off my browser and slack, load would return to normal, and I could use the system for a few more days or a week before it happened again.

I struggled with this for a year or more, before (for reason's I don't remember), I added another 8GB of swap. Within the next day 80% of the extra swap was used right up. I decided to swapoff that new file and create a 64GB swapfile and just see what happened. It got to around 20-30GB IIRC before it leveled off. And, more importantly, over the next 100+ days I never had that thrashing happen again.

That was with Ubuntu 22.04. I just recently (May-ish) switched over to a NixOS setup with 64GB of swapfile, and it's been very stable.

... but if you buy my book with more of this, I'll send you even more to read!

I get it, sometimes it's fun to follow a story. But I also read a dozen or so paragraphs, then started scrolling, then gave up.

I wonder how much of this is due to the AI tooling being taught on sloppy code that humans have written. Over the last 4 decades I've looked at a lot of code on the Internet and there's a lot of slop out there.

>Are you sure this is a bandwidth issue?

It's technically not a bandwidth issue, AFAIK, it is a "profile" issue. When listening to audio only you are using a different bluetooth profile than when you are using it for two way mic+speakers.

Try playing some music on your computer and then start a video call. Even before I jump into the meeting, as soon as I hit the Google meet site and it opens up the meeting "camera preview" page, the music quality goes straight into the toilet.

This also happens to my "puck" speaker+mic array that is connected via USB.

I wish more cameras with GPSes supported SMTPE or something to mark the video with a time code. I get the sense that GPS time is accurate to within 1/30th of a second, if not way way more, but even if it just got me in the neighborhood so I could nudge it a frame or two one way or another.

I like recording things like my kids performances with a GoPro or two, and sometimes my wife or I get some shots on our phones. Multi-cam can be fantastic when you have people walking around in front of cameras, for example.

I used to have a majorly hard time syncing the multiple cameras shots, though the last time I did it (a couple months ago) I had much better luck with the "sync using audio" feature, in the past I've had problems. But why can't we have a timecode with the video stream, in the cases of GPS cameras (I think everyt camera I have now has a GPS)?

That deserves it's own top-level discussion, for sure! https://news.ycombinator.com/item?id=49457539

@bradfitz: Steal this idea ^^^ ;-)

I was just wondering yesterday if it'd be possible to have a tailscale client, or multiple tailscale clients in their own network namespaces, that could connect to multiple networks.

The slightly tricky part would be managing a local NAT that blends the two networks, and also does MagicDNS to answer on those NATed IPs. If routes conflicted, I guess you'd have to pick one?

Seems very doable though.

For reference, ~2 years ago I did a 8x12 shed for under $3K, DIY. It is not insulated or finished inside. Footer and slab were under a grand, I used some reclaimed 2x4s and bought the rest, sheathing, roofing, flashing, self-made windows were around another $1,200-$1,500. Paint was the real surprise at almost $300.

>live near your actual office, which your employer already pays for,

I've been in the game for almost 40 years now, and I'll say I've never had an office paid for by my employer that is as nice as that toughshed. I've never been in a dedicated office with a door, except when working from home.

Edit: ProTip: If you are doing concrete, rent a "Mud Mixer". Me an the kids were able to pour the 8x12x5" (130 bags of 80lbs) in well under a day, probably would have been even less if we had moved the bagged concrete to the pour site before the pour happened rather than as we were pouring.

Agreed, it was 2-3 months wait to get to see an ENT for me last year. My daughter's regular eye check identified a problem and it's 3 months to get her into an eye specialist to review it.

Somewhat related: a couple years ago my spouse had cancer, and the doctor could do surgery either 5 days after our insurance was changing companies, or 2 months after that, those were the next two openings. The insurance said that if we proceeded with the surgery, without the new insurance company giving us approval, but they couldn't start approving it until the insurance started, and the pre-approval from the other company didn't give us anything. We finally decided to go ahead with it and take on the risk of ~$70K if it was denied. Which we decided to do rather than my spouse spending a few months worrying. Turned out that was a really good call, because when they did the surgery it turned out that instead of the expected slow growth, it was actually growing extremely fast. That was probably the difference between it being fatal and it being treatable.

The US system is far from smooth.

Some things that people dislike about Flock:

Flock has had a a number of security issues where people have demonstrated that the cameras could be accessed easily by anyone because of (IIRC) the use of default passwords.

Flock have said that they weren't doing this or that (I don't recall the exact details, I think initially they may have said it was just recording license plates), but then people found out they were doing a lot more than that.

Flock and police departments have been found to have been doing way more than using the cameras to investigate crime, including stalking friends and lovers and exes, and in one case one of the guys was found to have been regularly watching a young girls gymnastics gym a lot.

female is also on that list...

Claude Code and Codex are, in my experience, quite good at packaging software for Linux. I haven't gone through the full list of Linux deployment mechanisms, but I've had good success with little more than "Please make a set of github actions that publishes packages for releases as deb and rpm files for x86-64 and ARM, and makes a deb repo in github pages, and also for nixos."

>You are basically describing predatory pricing

I'm describing business. Period. It's incredibly rare to encounter products above a fairly small revenue level (niche products), that haven't come from a company that spent significant time reinvesting profits and therefore being unprofitable. Same for using outside investment.

Yes, there are companies who have burned outside investment on things that didn't pan out.

Love them or hate them, the AI companies are doing a huge amount of research. And research companies have always cost more than they've generated in revenue.

>doesn't turn a profit.

This isn't the condemnation of AI that you think it is.

When you are building an industry, you don't want to be in a "cash cow" phase. Because the person who is willing to reinvest profits (and further: seek and sink investments) into growth and innovation is going to eat your lunch.

~8 years ago my BIL condemned Tesla with "they aren't profitable" and I explained this to him. Now they're a trillion dollar company.

I thought Fable 5 was pretty good, but with all the shenanigans (specifically the we're pulling it on X, j/k, you get it another week; too bad you burned all your Fable tokens)...

I decided to try Codex+gpt-5.6-sol-xhigh, and it's really good. It reminds me a lot of Fable 5, but it also is a LOT harder to use all your tokens with. Plus they give you a free reset a week right now. I'd even go so far as to say I think it's better than Fable 5. With an asterisk...

That asterisk: For anything very big, I'm having it do review loops using Claude Code + Opus 5.

For example, I gave it a few paragraphs describing a bi-directional rsync without the Unison database. It worked solo for 4-5 hours and in my testing it works great. I gave it a few small feedbacks (give me progress in human units not bytes, there's a bug syncing a specific file rather than directory, etc).

I'm going to take it that you were being sarcastic.

However, in the case of the apt cacher, the DDoS of Ubuntu infrastructure showed that the existing apt cachers left a *LOT* of room for improvement, not to mention that apt-cacher-ng has regularly (over the last 13 years of use) gotten wedged and needed intervention.

This cacher understands Deb package repo structure, and when it finds new meta-data it grabs and checks the meta-data, downloads new version of any package you regularly download (configurable), and then makes a snapshot of that, which it then promotes for cache users to see.

I can shut down my Internet connection and then install a fresh Ubuntu from ISO, then "apt update; apt dist-upgrade" successfully via the cache.

If you are saying that AI isn't solving novel problems, you are waaaaaay off.

But you're also missing the point: This branch of the conversation is about AI tools writing wildly inflated code. Which I proposed a few data-points refuting.

You say that, but I asked Codex to build a Asteroids (but FPS) as a web app, and it wrote it in under 4K LOC, and a Tempest implementation in 6.5K LOC. Those seem like really reasonable line counts to me.

I also built a caching proxy server for apt packages, that's a bigger project, and that is 30.5K lines of golang, which doesn't seem bad compared to apt-cacher-ng which is 19.3K of C++ (my project does significantly more).

I'm not familiar with Stellar 7, but from a gameplay video it does look somewhat similar. Funny you should mention tanks, because that same day I had Claude build a Battlezone-inspired tank game: https://linsomniac.github.io/tankzone/

Similar, last week I had AI imagine Asteroids, but as a first person shooter, play RoidBelt in the browser at: https://linsomniac.github.io/roidbelt/

Last month I said to my wife that we should start paying homeless people to hold their signs while standing in front of the radar speed cameras the city has been putting up around our town.

>It doesn't affect knee room at all.

There are definitely airline seats where the reclining seats impact the knee room. I'm 6'2" but I seem to have more of it in the legs than typical, which really shows up in non-extended legroom airlines. I buy the upgrade whenever possible, but sometimes it's just not available.

In a regular (not extended) seat, I have two positions that work: my ankles crossed and tucked up close to my seat with knees spread, and nothing in the underseat in front of me and then putting my feet there, with my legs straightened (wedging my thighs into the edge of my seat). Even then, my knees are high enough that usually the person in front of me reclining will recline the seat back into my knees (best case) or those metal side rails into my knees. Especially enjoyable when the suddenly recline the seat.

Not everyones body geometry is the same. My son is around the same height as me, but when we sit in the car he's like half a head taller than me, he carries a lot of his height in his trunk where I carry it in my legs.

This isn't news, the real news is that the AI tooling allows YOU (specifically) to have a custom prompt tool built specifically for you, in the same or less effort it would take to set up one of the existing prompt tools.

For example, I liked the idea of this but had no interest in the shell history integration; I use atuin.

So I had codex build me one, including the 250ms wait for git, and if git takes longer than that it backgrounds it, uses the last git status until git can return an updated status. https://github.com/linsomniac/jafoprompt

CloudFlare has their plusses and minuses, but they do offer domain registration at cost, for example $10.46/year for .com (every year, not one of those deals for the first year then more expensive down the line).

Looks like that system uses the "AP Systems EZ1" microinverter I mentioned above. Their "latest news" page lists UL 3700 with a date of about a week ago, so it looks like they've gotten it.

Unclear to me: does this also do shell history? I'm using atuin and not interested in changing that. I did just switch over to oh-my-posh a few weeks ago and it's good, but I would be open to trying something else to compare, particularly since I have some directories with monorepos where oh-my-posh goes out to lunch for a couple seconds. Also oh-my-posh sometimes loses its mind and switches back to the default prompt rather than my customized one.

I'm on the fence about my current "2-line prompt, collapsing to a single line" when I press enter. I like that it gives me a lot of room, but there's some information I find I'm wishing was persisted in my shell history (git branch, previous command execution time for long running commands).

It is likely not going "back to the grid" unless you have the rest of your house shut down; where the "balcony solar" output is greater than your house usage. These systems are "back-feeding" into your circuits, and if you have it on a circuit that is less heavily loaded than the panels are providing, the other branch circuits are going to sop up that usage before it literally feeds back into the grid.

As I understand it, the US is mostly waiting for UL listing of equipment, which hasn't been a priority because no US locals have allowed them. With Utah and others starting to allow them, UL listing will likely come at some point (hopefully shortly) thereafter.

There's this device: https://usa.apsystems.com/ez1-installers/

And also this: https://connectder.com/

I've had a similar experience with NixOS configured by Claude Code, and it's been really nice. I have the same config on 3 machines, all managed in one git repo, including home manager and sops, and one of the machines I skin as a "thin client" removing most of my development tools and just using it to SSH/wezterm in to my main dev box, on a small and light ex-chromebook.