Weeknotes 2025 W23: Moonlander
Quick bits:
- Tomorrow is a public holiday, and I took Tuesday off, too. I deserve it.
The bad luck that I’ve had with cycling, like the recent cycling accident, has given me nightmares where I wake up in the middle of the night in a panic, replaying accidents — even long-ago ones — in my head. That’s psychological trauma right there.
I have received a letter from the police with a witness form that I need to fill out. Odd that I wasn’t the witness but the victim in this traffic accident. I am not looking forward to filling out that form because describing the accident in detail in German is just not something I feel like I have the energy for.
Semi-related: My phone has started auto-correcting “cycling” to “fucking.” I had no idea that it could autocomplete bad words.
Some progress on Deniki. For starters, I gave it a very basic icon. It’s not great, but it’ll do:1
I added basic localization support, too, and I’ve even begun an attempt at translating Deniki to Spanish! It is not going well, because I do not really speak Spanish. A for affort effort, though, right?
I have also added keyboard shortcuts for common Markdown syntax. Now you can use ⌘I for italics (or emphasis, technically) and ⌘B for bold (or strong emphasis, technically).
Lastly, I switched the storage format to be a package rather than a single JSON file. A package with Markdown files is much more usable; Git diffs have become meaningful again. But, as you can imagine, I broke the existing files when I switched to this new storage format. I coded the ability to read the old format and write the new format, but I removed that migration option before I migrated all my existing files. Sometimes, I really am not the smartest.
My ZSA Moonlander arrived on Monday and I’ve been using it exclusively, even for work.
It is not an easy keyboard to get used to. I have been customizing it extensively to fit my needs, which has made the switch considerably easier. Take a look at my keyboard layout as it stands right now. I am certainly not fully satisfied with it, and further changes are coming.
It nonetheless is a struggle. In particular, I find the columnar layout to be challenging. The Z, X, C, and V keys are the ones I mistype most.
The worst part: now that I’ve gotten used at it, I struggle to type on a regular keyboard.
In setting up my new keyboard, I discovered two fancy standard macOS keyboard shortcuts: ⌃A for moving to the beginning of the current paragraph, and ⌃E to move to the end. How useful! There are plenty more shortcuts, too.
In my continued adventures to figure out what the heck people are raving about when it comes to AI, I’ve been playing around with “agentic” AI. Not with much success.
I am distinctly in the AI skeptics camp. But at the very least you can’t blame me for not trying it out and thus not having the experience to back up my skepticism.
On a particular task that I decided to use agentic AI for, the first problem I ran into is that the AI had trouble with the space in the path of the repository I was working on. I know that there is the camp of people who loudly yell that you should never have spaces in directory and file names. I am not in that camp; I believe that software that can’t handle spaces in paths is broken.
The second problem I ran into is that the tests that I tasked the AI to write missed the point entirely. I had to repeatedly revert the changes it made, and it kept suggesting the same irrelevant changes over and over.
Finally, the last problem is that it “fixed” a failing test by… commenting out the entire test, then claiming it was done, and then asking me whether I needed anything else.
I could not have come up with a worse solution if I tried. The whole process felt like interacting with an extraordinarily incompetent intern with a mountain of misplaced confidence. I ended up not using anything the AI created for me. Using it was a waste of time, and I could have done a much better job, and faster too, by myself.
I know that all of this is just anecdotal evidence, and not something to draw general conclusions from. But for me, 95% of my own experience has been disappointing and particularly frustrating. I know that there’s the argument that it is still “early days” but frankly, if it really is, then I wish AI wasn’t continuously rammed down my throat.
And rammed down my throat it is. My employer has made the use of AI mandatory in daily work. I don’t know what the implications of that are. Will I receive a warning if I don’t use AI?
There is only one corporate approved LLM, and presumably my employer gets to follow all the work I do with AI in detail. Privacy is out of the window; everything I do with AI at work can be monitored.
If AI were genuinely useful for employees, why would there be a need for such a heavy-handed top-down decision mandating the use of AI?
Having spent some time with LLMs for coding, some other concerns became clear to me:
-
It makes us lose our craft. The promise of having a tool at your fingertips for manifesting thoughts and ideas is attractive — I won’t lie. But if we stop using our craft, we will lose our independent ability to create.
-
We are tying ourselves to megacorporations. There are no indie LLMs. Sooner or later, when we are locked in, we will start paying the price. Willingly giving away the means of production is a genuinely dangerous idea.
-
We are relying on IP theft. In my time experimenting with AI, on multiple occasions the AI produced something that I recognized from elsewhere, reproducing it without permission.
-
It makes us accept disinformation. So much of what these AIs produce is wrong. Rarely obvious, often subtle. I find this incredibly dangerous.
-
We are throwing security out of the window. The agentic AI approach I described above only works when you use the MCP to let the AI execute arbitrary code. It is wild to me that this is considered acceptable.
That is not to speak of the climate impact: new data centers and power plants are being built just to handle AI infrastructure.
I am tired of AI being at the center of attention and I think I will ignore it, along with the discourse around it, for a while. Even when I am not using it, it is sucking up my energy.
Entertainment:
- Swimming Pool2 is still great.
Links:
-
Clair Obscur is an Endless Canvas of Grief (Jacob Geller)
-
Server Mono: Yum!
-
Ducks often stand still in rain. Likely to save energy and stay warm, but we have our own theory: Praise the duck gods!
-
Operant Conditioning by Software Bugs – Embedded in Academia (John Regehr): Old, but still relevant.
-
I made this 3D game with ONLY code (Carter Semrad): Impressive!
-
Talking Heads - Psycho Killer (Official Video): Well, I did not expect a music video to come out about 50 years late.