This is a fascinating take. I have been thinking about your comment for two days.
I think you're right in some cases (when working in a field one has mastered, for example), and I think I could probably go in the direction of getting it right the first time.
But the way I see it, any time I'm doing something new or innovative, I'm doing something I don't know how to do, which takes trial and error; and troubleshooting is basically figuring things out by trial and error, in a systematic way.
Though a lot of time it is used for fixing bugs, I think troubleshooting as a skill and mindset is equally useful for creating new things, where you are solving for something.
Notice that if you mouse over "9 hours ago" on the story it shows the timestamp 2025-02-25. 9 hours ago was not 2025-02-25. If you mouse over the "7 hours ago" on credit_guy's comment, the timestamp shows 2025-02-26. One day after it was submitted, two days before it made the frontpage.
Thanks for posting an archive link. My site has survived previous HN traffic spikes on Fly.io's free tier, but 256mb of RAM wasn't quite adequate this time :)
I self-host Ghost on free-tier cloud hosting, which ticks all your boxes if you have some sysadmin skills. (PikaPods is the cheapest managed Ghost hosting I know of.)
I also really like the philosophy of Bear Blog and Matoara Blog. I don't think Bear has newsletter sending, but Matoara does.
A few years ago I doggedly tried to switch to Nyxt for everyday use. I really liked the concept, but at the time, it was too buggy, and constantly crashed on me. I'm going give it another shot.
Nyxt is very promising, and I hope it gains momentum. The obvious advantage of Nyxt is programmability and keyboard-driven workflows.
However, I use old hardware, and it's a bit slow and laggy. Chromium behaves the same way, so I imagine this is due to WebKit and Blink being significantly heavier than Gecko.
Firefox is really snappy on old hardware, at least when running Linux, and uses a modest amount of memory.
Hmm for me on BSD it's the opposite. But every release something breaks in Firefox' GPU acceleration so I've kinda stopped trying to fix it. I assume chromium handles that better. Though I don't use it much.
What's considered old nowadays? My current desktop had some ~3Ghz AMD processor from ~2020, I don't even remember what it is. Maybe 16 GiB of memory. Runs Firefox and Jetbrains good enough that I've not had to consider upgrading. I use a Debian based distro.
Not GP, but I have an i5 from around 2015. 4 cores, no hyper threading), 32GB ram. And it's still good enough for most things. Adding ram and a decent SDD gave it many extra years.
In the prosess of switching it out completely now though.
What kind of work do you do with those specs? I have a machine from around the same period with similar upgrades and I feel that it really is good enough, especially if you don't play video games.
It's actually the machine I use when working from home and it works fine.
PyCharm IDE on a moderately big project. We've recently focused on moving towards fully type hinted code base, using pyright as the checker/LSP. Which means I use a good-enough-but-not-great plugin which makes pycharm use pyright instead of the buggy heuristic based built in type "engine". And this is noticeable slower, which is part of my motivation for upgrading.
Other part is to be able to play a few games. I could probably just upgrade my GPU, but I'm not sure if it's even possible (outdated PCI port on my MB I think) without getting something second hand.
Third part is that I want to play more with the new NLP technology which is rapidly reaching a point where it should be possible to do a lot of cool stuff. And I want as much as possible to run locally.
I had a similar CPU (i5 4690k) and it would have been good for a few more years, if not the 3080 I once bought after a fight with my wife. Old CPU was just slowing it too much.
Cool, thanks. So there is the flatpak version, the Electron version, and the GTK version? Is there an easy way to install the native GTK version on recent Debian(+backports)? Or only "make && make install"?
I made a script which does exactly the same thing but locally using koboldcpp for inference. It downloads MiniCPM-V 2.6 with image projector the first time you run it. If you want to use a different model you can, but you will want to edit the instruct template to match.
MiniCPM-v 2.6 is probably the best self-hosted vision model I have used so far. Not just for OCR, but also image analysis. I have it setup, so my NVR (frigate) sends couple of images upon motion alert from a driveway security camera to Ollama with minicpm-v 2.6. I’m able to get a reasonably accurate description of the vehicle that pulled into the driveway. Including describing the person that exits the vehicle and also the license plate. All sent to my phone.
