2026-03-14 Sat

  • editing raw logseq markdown in neovim today, as an experiment.

    • it's a bit ... relaxing. barebones.
    • vim fingerspitzenguffeil / touch typing is coming back.
      • that feeling of celebration, when the fingers "just do"

        something like a major block indent.

    • for some reason, using nvim feels more creative than using logseq.

    • is there an lsp for logseq-flavored markdown?

    • has someone written vim in rust yet?

      • neovide/neovide: No Nonsense Neovim GUI Client in Rust

        • a rust GUI for neovim
        • never thought about wanting this, but setting the VISUAL environment variable to this

          could be very compelling

      • Helix: A Neovim inspired editor, written in Rust - Hacker News

        • built-around tree-sitter and lsp

      • rsvim/rsvim: The VIM editor reinvented in Rust+TypeScript.

        • ok, this actually is writing vim in rust. nice!
        • Home | RSVIM
        • what is typescript used for here?
          • typescript is used for configuration! brilliant.
          • would VASTLY prefer to write a plugin in typescript to lua.
          • typescript runs in rust's tokio, the rust async runtime concurrency framework
        • philosophical musing:

          • from one perspective, ... typescript as configuration language for rsvim.
            in another, rsvim ... [is] a highly specialized javascript-based runtime
            akin to node.js and deno, but purpose-built for text editor: stdin for
            receiving terminal inputs, stdout for rendering the interface

            • this is a creative project.
        • it seems crazy to save your work with typescript SDK for Rust
          • :js Rsvim.buf.writeSync(Rsvim.buf.current());<Enter>
        • apparently, to quit, one needs to do :js Rsvim.rt.exit();<Enter>
        • that's what they put in the tutorial! this is clearly an early project.
        • wait, this was an intentional decision on their part:
          • they aren't just trying to make rust vim, they are more ambitious,

            in a way. they think all of the modal ex commands like :w, :q
            should be implemented at the scripting layer, not at the rust layer.

            • As a supplement to the editing experience, Rsvim will provide an
              official plugin (say, "ex.rsvim") to recreate all the commands
              compatible with vim/neovim, including :w, :q, :echo, etc.

            • very cool idea. this way it enables not just "vim" but any kind of

              modal editor that anyone can even think of.
              maybe this could even be language specific. mind blown a bit.

        • this is a perfect example of the kind of creative project that will be

          likely revolutionized by agentic engineering in the long run. many
          ai detractors can't see it just yet, but there are so many ideas that just
          don't get developed because they are too hard. not all ideas are good,
          but some good ideas go extinct even though they contributed something
          novel and potentially revolutionary to the gene pool.

          • not everyone thinks humans are special, but if one does, it makes sense

            to care about the ideas that could have flourished if human time
            could have multiplied to support them.

  • about generalization in ai

    • ai is getting so good at logseq, at a pace that, for me, just doesn't make sense
      • in the early days, llms were good at things that were well-represented in

        the training data. they were much, much better at typescript and python
        than they were at, say, hack, scala, vimscript, ocaml, zig, rust

      • that's still true today, but it seems like llms are getting better at

        niche use cases and domain-specific languages / dsls that are
        internally consistent at a rate that can't be explained just by
        additions to training data tied to that specific context.

      • it just "feels" like there must be generalization in the training.
        • it seems like this could be tested. what if we benchmarked

          llms on their performance / accuracy of writing languages that
          are constructed so as to be net new, definitely not in the
          training data?

        • is this what arcagi is about?
          • what are we on now, arc3?

  • Filed

2026-03-12 Thu

  • overheard, about the weather this time of year: it's "wrong-coat season"
  • long live GNU and Open Source
    • considered using the downstream fork of Nix by the company DeterminateSystems and decided against it.
      • ten years ago, I argued for React against VueJS because I thought that software was so hard and expensive it needed a big company behind it to make it flexible, fast, and rock solid in a world with a rapidly changing deployable surface area.
      • now with the rise of AI/Agent/Agentic/Engineer/ing, I think it's risky to rely on a product from a company, as that company has incentives to exploit you and your data over time. our incentives are not aligned.
    • open source + gnu is the new way forward. if we combine our forces - and our ai agents, we can build a better world, designed by and for individuals, not companies.
  • squinting at the horizon: an army of pitchforks
    • had a thought during Meditation that perhaps it's time to stop being as public about interest in AI
  • funny homonym: pitchfork
    • the original word, mid-1500s was a fork meant to help farmers move pitch, aka tar, resin, or sticky substances.
    • but there are also tuning forks in music.
  • Filed

2026-03-11 Wed

Earlier