Miniflux

I’ve been trying to increase my consumption of blogs and “indie web” content while winding-down my reliance on aggregators like Hacker News and social platforms like Twitter to find articles to read. As such, I’ve been using an RSS reader more and more. Until this week, I’d been using Feedly. I was never completely satisfied with Feedly – it seems to lean into its algorithmic “discovery” features pretty heavily, and it’s iOS app has some annoying bugs that never seem to get fixed.

This week, I setup an instance of Miniflux on a Raspberry Pi, and have been happy with it so far. Miniflux has a minimalist design (it’s web UI looks similar to Hacker News or Lobsters), it’s open source, written in Go, and was really easy to deploy with docker-compose.

I was a bit worried that my Raspberry Pi wouldn’t be powerful enough, but so far I haven’t encountered any issues. I currently subscribe to ~300 feeds, and while the refresh time is a bit longer than Feedly (i.e. it takes longer between an article’s publication to RSS and when it shows up in the feed reader), it’s not that noticeable in practice. Update: It appears that Miniflux’s polling time is configurable, so this is actually a nonissue!

The other cool thing about Miniflux is that is has an API that’s quite easy to use. There’s also a small ecosystem of projects that add functionality, like filtering, to Miniflux. I setup miniflux-sidekick alongside my Miniflux instance, which allows me to filter out articles (like sponsored articles, podcast announcements, etc.) from my feeds. Feedly has a feature like this, but only as part of their paid subscription service.

I’m happy to be slowly building up a collection of self hosted services. πŸ˜„ Once you get things setup, the maintenance burden is surprisingly low…

Email

I’d been meaning for a while to setup an email address on a custom domain. I don’t really like giving out my personal email address(es), and I think having an email address on a custom domain looks more professional/cool than a {g,hot}mail address.

I finally got around to that this week! I considered going with another self-hosted solution to this, but I read enough opinions discouraging this that I relented and chose a managed subscription service.

You can now reach me at “ben at congdon dot dev”, if you feel so inclined. βœ‰οΈ

  • GPT3: An Even Bigger Language Model - OpenAI recently released a new language model, GPT-3, which is a ~100x larger version of it’s previous model, GPT-2. I find these GPT models to be fascinating. It’s spooky how good GPT-3 is at generating text that looks like it was written by a human. Computerphile made a pretty good explainer comparing GPT-2 and GPT-3.

    • Also, take a look at this much longer exploration into GPT-3 generating Creative Fiction – it contains a variety of samples of GPT-3 generated text.

    • Emoji storytelling – Using GPT-3 to generate emoji summaries of movies. Some of GPT-3’s summaries are quite good:

      The Godfather: πŸ‘¨πŸ‘©πŸ‘§πŸ•΅πŸ»πŸ‘²πŸ’₯

      Star Trek: πŸ––πŸš€

      Game of Thrones: πŸ‘‘πŸΊπŸ°

      Jurassic Park: πŸ‘¦πŸ¦–πŸ¦•πŸ¦–

      The Hunger Games: 🏹πŸ₯ŠπŸŒ½πŸ†

  • In search of the best custom keyboard layout