Plain-text files are at risk
10 points by awal
10 points by awal
Just 20 years ago, it was normal to use such editors regularly, even if you were a layperson.
This does not reflect my experience; the last time I saw a layperson using plain text was in the 1990s.
Ha, maybe I'm getting old, or maybe the timeline was different in different regions of the world.
It's not far-fetched to assume that maintenance of plain-text editors and the ecosystem surrounding them will stop.
That's extremely far-fetched, sorry. Is the thesis here that literally all programmers will switch to only interfacing with files indirectly, using LLMs? Including hobbyists, who might be working on low-level stuff? That sysadmins will never need to manually edit a config file on a machine with broken networking? (or, I guess, that they'll just hook up a serial cable and their LLM will edit all files using bare cat)
The barrier to creating an usable text editor isn't exactly high. I know plenty of people who wrote one just for fun. I haven't done that myself yet, but I hear that's a fun project. We've just had a 2000-line vi(1) implementation posted to the site recently. On that note, I believe all (?) major BSDs maintain a vi(1) as part of the base system, and I don't think that requires much maintenance either.
I'm posing a slippery slope argument so I understand the push-back. And if you don't wish to entertain that at all, that's understandable as well and you can ignore the rest of this comment.
Sans "literally", yes that is pretty much the thesis I am presenting here. So indeed I don't consider it extremely far-fetched.
I suppose part of this is also that I find the lack of discourse on this topic a bit worrying and I wanted to sound some alarm bells, sorry about this.
I think you are underestimating the power of access to niche options. Today, plain-text editors are accessible to everyone, even if they are ignored by everyone except SWEs. If corporations tighten the screws too far on subscriptions/enshittification, the average person could conceivably abandon "apps" and switch back to plain text for everyday things, or at least they could be stirred into doing so by their techie friends. Presumably corporations understand this lever and are conscious of it.
But if most of us stop using them as well, there would be even fewer voices in dissent when corporations get rid of them/remove them from "app stores". And limitations on proprietary platforms also have a way of affecting free software.
??? If anything, I feel like LLMs have made plain text more important, at least for programmers
Many people switched from using IDEs -- where you could imagine the UI eventually substituting for plain text -- to agents, which are adept at manipulating plain text
I mean I guess it's true that the agents may hide the bash commands they're running from you, and some people have no idea what the agents are doing
But I like that they are using plain text. LLMs even output diagrams in ASCII art !!! Rather than a format that's more non-standard or opaque to humans
And yeah I agree, I have not seen non-programmers using plain text for decades. Most people don't even know the difference between say plain text e-mail and HTML e-mail. The e-mails I receive make that obvious.
Not just for programmers, either. Non-programmer LLM power-users are having to learn what files are, and why plain text and Markdown are useful formats.
Yes. And as far as I can see, nobody has seriously suggested "we don't need text files because we have LLMs doing work" -- it's the exact opposite
We use .py .js .go files for code, .md .html for docs, .csv .json for data, etc. LLMs basically guarantee those formats will live forever !!! They are never going away. If anyone suggested we don't need them anymore, then I would question if they understand how LLMs work ...
I mean probably the split is between people who look at the code directly and people who don't. But yeah the non-programmers who have the most success with LLMs will eventually start looking at the code.
I would also add: Why did "computer use" come later to AI than breakthroughs in programming and math? e.g. getting an LLM to edit video or audio with a GUI.
I think it's clear that a big part of the reason is that math and programming are expressible symbolically, i.e. with simple plain text encodings. And those formats are very efficient for LLMs to model, reason about, and generate.
I don't get the argument. Even if only software engineer edit plaintext file, then the one using them are the one with the ability to maintain software to edit them.
It doesn't take a team of 2000 engineer to create a good text file editor. It take only a few dedicated enthusiast. If the world has enough such enthusiast to run a modern browser engine on Mac System 9, I'm pretty sure there will be enough reluctant old curmudgeon to maintain VIM until the heat death of the universe. The fact vi derivative are still amongst us after 50 years is proof enough. Heck, with AI, anybody with a bit of patience can vibe code a basic text editor.
Yes, there exists, right now, kids that edit their plaintext files exclusively with LLMs. But to stop having text editor, you need to exhaustively remove almost all the old curmudgeon from the world that will want to edit file by hand and that could maintain the text editors. "Some swan are white" is extremely far to the proposition "all swan are white". An argument that require the later form cannot stand when all you have is evidence of the former.
I think files in general are at risk. As an Android user the idea of downloading a file and loading it into an app is so clunky! I was recently trying to load my OPML subscriptions into a feed reader app and found mobile web browsers make downloads tricky. I think the platforms want everything API based to keep you locked into walled gardens.
Like other commenters I believe we will continue to be able to read and edit text files. However, the bulk of my online communication used to be plain text (on IRC) but now all of those communities have disappeared and their replacements are on Discord or another unpleasant web app.
I consider the novel, anthropologized terminology for these tools superfluous since the word "REPL" already sufficiently covers them.
I don't like the anthropomorphized terminology, either. But please don't call them REPLs. REPLs have been deterministic, and these things are most certainly not that.
I like your adaptation of the old saw about markets:
Society can remain irrational longer than you can remain a relevant nonconformist.
By plain-text files, I mean notes.txt/md/ini/json/etc.
Only one of those is plain text. Plain text means unstructured text, not just non-binary. At least, that's been my impression across the globe the last >25 years in software.
Personally, I find that the text/binary dichotomy is a problem. It's all bytes. Sure, if you open a compiled executable file with nano, you'll get gibberish, but you'll get a fair amount of gibberish if you open a "text file" in vi that's gone through "Word" and had every single- and double-quote turned into "smart quotes". Same with "a text file is a file with a name that has a '.txt' suffix". The only thing that exists is variation in the ability of programs to display a file's contents to a human legibly. Image-format files have the same property. Yesterday, I found a file with a ".jpg" suffix. feh rendered it as if it had an alpha-component, and so did Firefox. Turns out the contents of the file were in HEIC format, not JPEG format. There's a bellcurve/midwit-is-wrong meme in this.
Really? Markdown seems pretty popular, and org-mode has it's following. I'm not sure I buy the gloom