Notes From An Unspecified Time

After a couple of weeks fielding a few (un)usually stressful projects, I eventually managed to pull my usual stunt of falling ill during a short vacation, during which I mostly sipped tea, watched TV, and played instead of doing anything productive like, for instance, cleaning up my office, which is a complete mess.

But I did manage to get a few things done, albeit they were a bit all over the place. For instance, somehow I tried my hand at implementing K-Means inside Node-RED and other vaguely futile stunts (I wanted my RSS summarizer to use embeddings to group items).

I also played a bit with the Rockchip NPU and LLM tooling:

DeepSeek in a nutshell
Testing out a new Rockchip board.

Yes, that’s yet another DeepSeek distillation, and not a particularly smart one at that. But this time it was running on a Rockchip RK3588 (most specifically, the new ArmSom AI Module 7, which I got an early sample of for testing), and that is interesting for me since I intend to try converting across another model (I am very close to shoehorning a usable LLM into an industrial monitoring solution).

The AI Doldrums

What I’ve found during the past few days I was stuck indoors is that I am increasingly put off by the effort involved in building AI “solutions” – they’re just no fun. Even kicking the tires on new tooling ends up being a waste of time, and (most importantly) I’m not learning anything except new and unexciting ways to concatenate prompts.

Although I am impressed by Claude 3.7 in (if you’ve missed the announcement, it’s already available via GitHub Copilot) and it is nice to ask for quick and dirty code suitable for chores, I still feel that by using AI at all I’m missing the creative spark that makes programming enjoyable.

And, worse, I don’t recognize the code it generates as mine. Which for someone like me who has an uncanny ability to remember code and patterns means I can’t really navigate it as well as I’d like.

In truth, even feels like too much overhead, although I appreciate the results.

Music Gear

The only upside is that I have finally gotten to test the Picotracker I started assembling , and it works OK, even if the firmware is quirky and it seems to have trouble with my limited sample library (I just dumped all my samples onto a new SD card, and I probably need to convert them to a different, smaller format).

Picotracker
The Picotracker, in all its glory.

I also seem (knock on wood) to have “fixed” our , which (since I know the fix is temporary) led me down the rabbit hole of considering alternatives and, in turn, bingeing far too many YouTube videos about the PSR-SX920 arranger keyboard.

Despite the cheesy reputation of arrangers, it seems to be a better bang for the buck than the Arturia AstroLab (of which I already have most soft instruments anyway).

Given , I should probably start reviewing music gear again. Hey Polyend, let’s have a chat, we’re both in Europe anyway…

This page is referenced in: