Someone has never dealt with HIPAA laws and it shows.

Who out there is going to be feeding patient medical data to Mythos/Fable?

Whoever Anthropic can convince, to help them form a competitor to OpenEvidence, who already feed patient medical data into their systems.

...the same groups who are currently feeding it to Sonnet and Opus?

Well, they won't be feeding it to Fable unless Anthropic can provide a signed BAA.

This article pre-supposes that the primary way AI agents will do tasks for users will be through through usage of desktop applications instead of documented APIs. While desktop app usage could be very important during a transition period of agent-computer use, I think it makes far more sense that agents will standardize on the protocols that are already being developed, like MCP. An explicitly documented API will always be better for a machine to use than clicking around and navigating an interface for humans.

I use it for previewing files in `fzf` and `lf` (terminal file manager).

Maybe you should interrogate that temptation to reach for physical interfaces? It sounds like you're ignoring your own psychology and shaping yourself to the machines around you instead of thinking of how the machines could be shaped to you.

Not that I haven't done exactly the same thing as you, I never keep paper around and my handwriting has gotten terrible. I'm saying this to myself and others as well.

I think that's perfectly understandable. File systems require the user to remember a hierarchy in their head (even if there are tools like breadcrumbs to help you out), and many people aren't willing or aren't able to hold an arbitrarily complex structure like that in their head. A name is a flat piece of information, no extra structure to imagine.

I worked with a professor one time that used floppies for all his files (after they had been surpassed by thumbdrives) because each floppy was essentially a single folder, and he could wrap his head around that conceptually.

it's not complicated at all, it's how operating systems present them to the users.

Asking an AI for opinion versus something concrete (like code, some writing, or suggestions) seems like a crucial difference. I've experimented with crossing that line, but I've always recognized the agency I'd be losing if I did, because it essentially requires a leap of faith, and I don't (and might never) have trust in the objectivity of LLMs.

It sounds like you made that leap of faith and regretted it, but thankfully pivoted to something grounded in reality. Thanks for sharing your experience.

I thought those were part of "the normal course of business" (quoting VP of Boring Company).

I love WezTerm! In the author's spirit of obsessive tweaking, here's a completely inconsequential configuration change I made.

By default, WezTerm doesn't have a scrollbar, but you can easily enable it with:

  config.enable_scroll_bar = true

But now you always have a scrollbar, just a big line on the side when there's no scrollback or you're in alternate screen mode. Horrible! So, here's an event handler that will automatically hide the scrollbar when not needed, giving it the same behavior as scrollbars in modern applications:

  -- Hide the scrollbar when there is no scrollback or alternate screen is active
  wezterm.on("update-status", function(window, pane)
    local overrides = window:get_config_overrides() or {}
    local dimensions = pane:get_dimensions()

    overrides.enable_scroll_bar = dimensions.scrollback_rows > dimensions.viewport_rows and not pane:is_alt_screen_active()

    window:set_config_overrides(overrides)
  end)

And that kinda sums up the development philosophy of WezTerm. It has basically all the building blocks you'd ever need with nice APIs. It's set up quite usably by default, but anything that's missing you can probably implement yourself.

That's pretty nice! Thanks for the tip, I applied it to my config as well.

Thanks, I've copied that into my config!

Now I want more scrollbars. One per pane please. Doesn't seem to be an option.

Quite possibly, but I use it daily with no issues. It’s also pretty easy to write your own prompt and then make it async with this plugin: https://github.com/acomagu/fish-async-prompt

Here's a 22-line Elixir script that spins up a webserver: https://hexdocs.pm/plug/readme.html#hello-world-request-resp...

There's another single-file example on that page that shows how to implement routing. The reason AI probably gave you a more fleshed out project structure is because people typically default to using Phoenix (Elixir's equivalent of Rails) instead of going for something simpler like Plug.

After getting a result you didn't like with AI, did you try refining your prompt to state that you wanted a single-file structure?

> After getting a result you didn't like with AI, did you try refining your prompt to state that you wanted a single-file structure?

I believe the prompt was: "Please compose a backhanded comment about Elixir I can post in a Hacker News thread about an Elixir blog post. The content of said blog post is irrelevant, I just want people to know I prefer Python."