Still Good: Ubiquiti Networking Hardware
In a world of the enshittification of nearly everything, somehow my Ubiquiti gear is still fantastic. It has that Apple-of-2004 feel: everything works, and it works flawlessly together.
I’m grateful.
In a world of the enshittification of nearly everything, somehow my Ubiquiti gear is still fantastic. It has that Apple-of-2004 feel: everything works, and it works flawlessly together.
I’m grateful.
I’ve been using Tinfoil + DeepSeek 4.1 Flash for inference for a few weeks now. I’m using it for my personal agent harness, some OpenCode tasks, etc.
It works really well! Totally private, inexpensive, and capable!
If you’ve got the wherewithal to build your own tools like this, check it out.
My current mental model:
Local LLMs -> Best
Private LLM -> Next best
Added some 88x31 gifs to the footer.
I played around with OpenRouter over the last week. Getting credits on my account was unusually hard (multiple “is-this-fraud” warnings from my CC provider(s), which doesn’t instill confidence), but after that it made it easier to explore more models outside the Big Ones.
Today I gave OpenCode and “DeepSeek V4 Flash 0731” (great name) a task: Build a Docker container for a little toy Ruby server I built. It spent about 3 minutes on it, building a Dockerfile, running into a couple of errors, fixing those, until it was ready to go. Total cost: $0.01. Best I can tell that’s about 1/10th the price of OpenAI/Anthropic.
Still not sure I trust some of these models and inference providers, even with ZDR enabled.
But it’s very interesting what you can get at this price.
[Update: Tinfoil has the same model; I should try that. I trust their privacy guarantees, at least. https://docs.tinfoil.sh/models/chat#deepseek-v4-flash]
I spent some time this weekend getting Grafana setup to monitor my Raspberry Pi fleet. I’ve been making some network changes and was worried that it was going to take a long time to figure out if something was broken. What I wanted was some kind of central error log reporting. I ended up with some great dashboards, too!
Right off the bat, once it was setup, I was able to discover and fix a number of problems, like old services that were in attempted restart loops.
Codex did most of the heavy lifting. I hate that I have less familiarity than I would have if I did it all by hand. At the same time, I wouldn’t have gotten this far in a weekend, and for me that was worth the tradeoff for this project.
Managed to get all the photos served up via imgproxy instead of through ActiveStorage/Rails. This should make photo loading much faster and more reliable. I tried to get some CloudFlare config tweaks in place as well; hopefully many/most images end up cached heavily.
I also uploaded a ton of photos to the Japan 2024 photo set.
How magnificent
the funeral song our children or contemporaries
will write for us, a song they will make by
taking our obituary and Facebook posts,
plus random quotes from our algorithm,
and feeding them into Chat
or Gemini
or Claude.
The tears that will fall in the face of such
sanitary sweetness!
Read the whole thing by Shawn Smucker
Day 1 at Open Sauce was amazing. They addressed every single small pain point from last year, and somehow the convention hall has even more incredible projects to show off!
I spent some time the last few days implementing the Photos functionality that I’ve wanted on this site for quite a while.
I was able to get Photo Sets working, some CDN caching in place, and a few bulk uploading tools built for myself.
I still need to work through a big backlog of old photos on my computer and get them all uploaded, but I think that’s going to require getting back to my faster internet at home.
It’s been a long time coming; I’m happy that this site is starting to feel like what I’ve been wanting for a long time now!
I am deeply uncomfortable with the continued control that CloudFlare is exercising over the internet in the name of security.
Now: Precursor. Spyware by any other name that you can force upon your website visitors, so that (in theory) CF can train their systems to recognize human client devices.
I’m sympathetic to the challenges of sites being overrun by botnets, but the solution is to find and disable those botnets, not to inflict further privacy-and-battery-eroding javascript crimes on the innocent. Nor is the solution to give a single entity the ability to block anyone from 20% of the internet.
I need to get my sites off CloudFlare, too.
I extracted a gem from a personal project: Detergent, an HTML-scrubbing library. Same idea as Readability; take some HTML and rip out the ads, the navigation, the distractions, etc.
If you stepped into the dwarven capital of Ironforge on September 13, 2005, you would find only bones. Lots of bones. The city, along with every other major population center in World of Warcraft, had been ravaged by a plague that slaughtered players by the thousands, their bleached bones covering every street.
Fantastic story about an unintentional WoW plague back in the day. Somehow, I hadn’t heard about this before.