As a native German speaker, I have also referred to a chatbot in English as "he", and similar to you, a native English speaker, felt jarred by it. It was definitely not out of any personification or humanization though. In German, I would say it is "der Chatbot" (from "der Roboter"), which in German is a male noun so I would refer to it as "er" (the male pronoun) - which in my head I autotranslated to "he". Most of the time, though, I think of it (and refer to it) as an LLM, which is "das Sprachmodell" (neutrum), so I automatically translate it to "it".
So that's another, maybe more harmless reason for it.
Andrew Zonenberg posted a Twitter thread a year or two ago where he fixed a missing PCB trace some layers down a PCB, with a stereo microscope, precision mill and very steady hands.
Edit: here's the thread. It's a 6 layer PCB with a short on L5 that needs to be fixed from the L1 side.
I don't know if their math is incorrect, but your intuition of the limit case is - downhill slope is the ratio of height change per length. A perpendicular angle thus has "infinite percent" slope (since the denominator, length, is zero), which intuitively matches an infinite landing distance.
I always understood this requirement as "garage will run fine on hardware with 1GB RAM total" - meaning the 1GB includes the RAM used by the OS and other processes. I think that most current consumer hardware that is a, potential garage host, even on the low end, has at least 1GB total RAM.
I keep wanting to get into Kicad, and the learning curve did become easier in the last years. But what I _want_ is to get a usable board into my hands, not just create a beautiful layout file. Historically, my pain points were were footprints, parts sourcing and SMT soldering. From a hobbyist-with-limited-time perspective, EasyEDA and the integrated JLCPCB assembly flow solved all these problems good enough that the hurdle to figure out how to do it with KiCAD was always higher. The minute I find a similar level of convenience in KiCAD + plugins, I'll gladly ditch being tied to online.
Immich manages to detect my kids faces much better than expected. I only have two years, but it is spot on with kid #1 from newborn to 2yo, and it manages to not mix up the new baby photos of #2 with the baby photos of #1.
In my 44k photos there are zero statues face detected, the only flukes are a few photos from a restaurant with a celebrity picture wall.
Not my experience hosting immich for close to two years now. There was only one "breaking change" a long time ago where you would have to manually change a docker image in the compose file, but since then things have been smooth for me.
Immich may not be the pinnacle of all software development, but with the alternative being Google photos:
- Uploading too many photos won't clog my email and vice versa
- I'm not afraid of getting locked out of my photo account for unclear reasons and being unable to reach anyone to regain access
- If I upload family photos from the beach, then my account won't get automatically flagged/disabled for whatever
- Backups are trivially easy compared to Google takeout
- The devs are reachable and responsive. Encounter a problem? You'll at least reach a human being instead of getting stranded with a useless non-support forum
I would instead say that my (and my family's) photos are too important to me to pass their hosting on to a company known for its arbitrary decisions and then being an impenetrable labyrinth if there is an issue.
So you do pay some price, but it is an illusion to think that the price of Google photos (be that in cash, your data or your effort) is much lower.
Things that did break during this time:
- my hacky remote filesystem
- network connectivity of a too cheap server
but these were on me and my stinginess.
You could keep the Hetzner VPS with storage for faster online serving of assets and connect a second immich instance only for machine learning on your home server. That way you'd get the best of both worlds: fast media serving and higher performance. That would mean that images are uploaded to the Hetzner server, but the compute-intensive image classification takes place on your home server.
So that's another, maybe more harmless reason for it.