Notes for August 26-September 8

And so it came to pass that I took a break from writing (and most other things) for a couple of weeks, and now I’m back. Sort of.

Like last year, I read six or seven books of various genres–some new, some old, some just to try to shake away the blues and try new things. Regretfully, as time goes by, one’s likelihood of finding something truly new and exciting diminishes, so I won’t go into detail about them here (even the technical ones were, to be honest, a bit of a letdown–the art of writing decent programming or systems books seems to be on the wane, somewhat like the average quality of software itself).

But I did finally catch up on Slow Horses a day before the new season dropped, so that almost rated as a success by the standards of one of the lead characters (if you watched it, you can guess who).

However, as far as my hobbies go, I failed miserably at trying to complete some of my projects–as usual, vacations are a grey area of time, space and willpower where I tend to go off into the weeds, and this year there was a steeper cliff between the days I could stay home and the ones we spent on the beach.

I did manage to spend some quality time sorting out non-computing things in my office. Sadly, none of my electronics projects progressed at all, even though I actually took a RP2040 board with me and spent a couple of hours trying to get it to do something useful at the hotel (yes, really).

But a few critical parts got held up in customs in my absence, so I honestly don’t know when I will pick those projects up again.

Since coming back from the beach, I’ve also been printing a lot of utilitarian things (like Gridfinity bins for electronic components) and “de-fragmenting” my drawers.

With the SK1 now enclosed, I’m tackling bigger prints than usual but I’m still not happy with the quality of the calibration. There’s never enough time to tweak things properly, but for functional prints it’s plenty good enough as is.

My travel kit has changed a bit, but not by much–the K7 Max and the Radxa Zero were the only updates, and even then I ended up connecting back home to (half–I powered down the rest) of my Proxmox cluster via Tailscale because there were things that needed more RAM than what I was carrying along.

And yes, you can go a long way with an iPad Mini if you put your mind to it (or if you don’t have anything else…). Living inside mosh and tmux sessions felt like a throwback to the good old days, but the K7 max, was a joy to use despite dwarfing the iPad, so I don’t regret taking it along.

But I spent a while pondering simpler, purer computing (as usual).

There is something to be said for the simplicity of a thin client, but also the polish of a good tty-based interface–and terminal utilities have come a long way, so for the sake of modernization, I’ve started moving my vim configuration to Neovim (I think eight years is mature enough…) and settled on wezterm as my new terminal emulator of choice (I’ve finally given up waiting for Apple to modernize Terminal.app).

So far I’m quite happy with it–well, except for the hideous macOS icon. Seriously, it’s the one thing that put me off trying it for so long.

Somehow, having a terminal that looks like this makes me feel like I’m in the future:

We’ve had this technology for decades, and we don’t use it enough. If you want an imgcat alias, here’s mine:

# if we have magick installed, then define imgcat
[ -x "$(command -v magick)" ] && alias imgcat='_i() { magick $@ sixel:- };_i'

This will work for things like imgcat image.png and imgcat *.jpg (and let you see images in full Sixel graphics glory instead of the blurry mess most imgcat utilities output), but I didn’t really have the patience to make it take standard input (yet) because I can always fall back to magick if needed.

Since I was bereft of nearly all significant computing resources, I ended up taking my old writing aids, tweaking the prompts for gpt-4o-mini and modernizing them a bit, although thanks to the new 128k token APIs I ended up with a few new prompting techniques to flag interesting news items.

This was a serendipitous prelude to the new Reeder launch, and what it effectively means is that I have a stack of prompts to go through and work into something more coherent that will blend in with my current news reading habits.

I will be poking at it and letting it brew steadily over the next few weeks…

Oh, and I finally tried FLUX.1. Like the original stable diffusion, the results were… interesting. I’m not ready to share them yet, but I’m definitely going to keep poking at at it for fun.

Most of the rest of the AI industry, as usual, is a dumpster fire. I’m not going to go into details, though.

Amidst the sweltering Summer heat, I sampled an explosive cocktail of Rust, C++ and… Hy. I’m not sure what I was thinking, especially since last year I was hopeful the latter was more stable.

But no, the Hy maintainers keep breaking the syntax (which, if you think about it, is a really hard thing to do on a LISP), so there was no joy there. I have no idea what they are thinking, but as someone whose code was (at one time) one of the main public examples of how to use the language, I’m not impressed.

I did manage to realign the ML part of my brain enough to write some embedded Rust without swearing too much, but I’m absolutely positive I don’t like it. I get the advantages, but when you’re used to focusing on data and using abstractions to transform it easily, the amount of non-incidental detail and effort you have to put into the language itself to do the simplest things is just… wrong.

After all, we built abstractions to make computers easier to program, not to get quagmired in.

I’m definitely not ready for that, but I have to, and unsurprisingly I’m not looking forward to it.

Let’s just see how it goes tomorrow and hope for the best, shall we?