A very different week due to various personal schedule changes.
Morning and late afternoon outings for various chores. Most of work was about the new fiscal year and getting things in order for the next few months, so I’m back doing a lot of paperwork.
Deceptively quiet day due to the US holiday, interspersed with a few meetings with faraway timezones and more late afternoon outings that cut into my “free” time.
Cranked out a couple of
PETG prototypes on my rickety Prusa. I’ve been feeling tempted to either build or buy a fully enclosed CoreXY printer, but I’ve got such a large backlog of projects that I’m not sure how I could fit that in.
Busiest day of the week, as usual. Poked despondently at some
langchain code and decided it was too brittle to be worth the effort.
Started playing around with an Artsey keyboard, which is an intriguing little gadget that uses chording (and only 8 keys) to allow for pretty comprehensive keyboard input.
The core layout is actually pretty easy to memorize, although typing on it does require a fair bit of patience–these sentences took quite a long time to type, and it’s not because I decided to go with a left-handed layout…
But it works. I can even type accented characters on both the Mac and iOS since the US layout has key combinations for that which work well enough–although they unfold into three chord sequences.
Hardware-wise, I got a mostly pre-assembled BlueHand, and even though I had to solder a bodge wire to get one of the keys working (thank goodness for having schematics available), the only real complaint I have is that the switches are a bit noisier and have more travel than I would like.
By the second day, I had already memorized most of the alphabet and a good deal of the more common symbols, and the biggest challenge I have is still the physical keyboard itself–the added key travel makes it hard to reliably chord at speed.
Spent a fair chunk of time investigating options for writing OpenAI agents and updating my AI resource list. There is a lot of
langchain stuff out there, but I found the overall quality of the code to be abysmal and would vastly prefer to use something else entirely.
Also, most of the alternatives that exist right now do not support the
function_call feature in the current revision of the OpenAI APIs, so I’m actively procrastinating on this because I want to avoid falling into the trap of writing my own wrapper.
Every time a new model comes out the prompting strategy changes, and this one is no exception–I had to spend quite some time in ComfyUI tweaking things, but the results are encouraging: