I have been mostly in “inbox zero” mode for the past few weeks, which generally meant checking stuff off my to-do list, trying to ignore the news, and using some fairly colorful language when I didn’t.
I even went out and had fun a couple of weekends in a row, which is somewhat of a novelty and tore chunks of otherwise rather unproductive mulling off my schedule.
That led up to Easter break and a few days in the countryside trying (somewhat unsuccessfully) to catch up on my reading, which I’ve been neglecting of late and I feel somewhat guilty about.
One thing I don’t feel guilty about, though, is pausing most of my AI stuff to see if I can get excited about any of it again. That has been quite the challenge too, even as work takes me down various related rabbit holes.
There’s such a wide gap between expectations around agentic AI and reality that my current survival strategy is to take it all in in small doses to minimize annoyance. Also, every time I type “agentic” I have a minor tussle with whatever form of spell checking happens to be at hand, which is a good reminder that the whole thing is still a bit of a joke.
That doesn’t mean I’ve sworn off AI completely, though. I’ve actually picked up a few more mini-projects, some of which I’m using Gemini 2.5 (via GitHub Copilot’s new Agent mode in Visual Studio Code):
- A small set of tools to convert quantized models into something I can use with
rkllm
or RoCm. - A new touch dashboard based on
pygame
, because upgrading Chromium on my current dashboard broke the whole thing and I’m tired of web technology bloat (it was either Godot orpygame
, but targeting a Raspberry Pi 3 with Godot is far from a trivial endeavor, so I went with something I could deploy and maintain easily). - Replacing a couple of my Tasmota devices with HomeKit-enabled ones (as well as doing some minor upgrades here and there).
- Writing an improved Proxmox ZFS monitoring script that sends me alerts when my ZFS pools are in trouble.
- Designing and 3D printing a few enclosures and other assorted bits (it’s quite rewarding to churn out PETG spacers to patch cracks in old window blinds in under 20 minutes).
I’ve also been mulling if I should do something to preempt rising hardware costs–for instance, I haven’t given up on getting a Ryzen AI HX machine for testing, since it seems like the only way to get any GPU with over 24GB RAM at an affordable price short of getting a new Mac–which I was budgeting for early next year.
But given the political climate, I think I might bring that forward a bit…