Plain text is still one of the best technologies we have

Original Article ↗ Hacker News Discussion ↗

Plain text is praised as one of the most durable and portable ways to store information, yet commenters highlight how messy it actually is in practice, from incompatible line endings and encodings (ASCII vs. Unicode/UTF-8) to ambiguities in number, date, and language conventions. Many argue that, despite these flaws, text-based formats remain vastly easier to inspect, preserve, parse, and version-control than proprietary or binary formats, which can render data inaccessible over time. Alongside this, the thread touches on tools and formats built around plaintext (for accounting, scripting, and writing), debates over Unicode’s complexity and emojis, and rising fatigue over AI-generated writing and unreliable detection tools.

AI-Generated Content Concerns

  • Multiple commenters suspect the article and even the whole blog are LLM-assisted or LLM-written.
  • Evidence cited: high posting frequency, similar post length/structure, narrow thematic focus, AI-looking author photo, and a GitHub repo with many “co-authored by Claude” commits.
  • Others push back: the piece reads human, includes minor mistakes, and lacks obvious LLM “tells.” Several note that AI detectors (e.g., Pangram) have high false-positive rates, especially for ESL writers and good prose.
  • Some express exhaustion with AI slop, while others warn that “this was written by AI” is becoming a lazy way to dismiss content without engaging with it.
  • There is mention (unclear, based on memory) that HN may be using AI tools to penalize LLM content.

Plain Text’s Strengths and Hidden Complexity

  • Strong agreement that plain text is durable, portable, and likely readable decades from now, especially thanks to ASCII and (increasingly) UTF‑8.
  • Several note that “plain” is doing heavy lifting: line endings vary (CR, LF, CRLF, even historical variants), encodings differ, and metadata like date/number formats or language is not inherent.
  • Some argue this ambiguity is acceptable or even a feature; others wish for standardized metadata (charset, line endings) and lament that filesystem extended attributes aren’t used for this.
  • Discussion of historical device and OS conventions around CR/LF and how tools have adapted.

Unicode, ASCII, and Text Semantics

  • Debate over whether ASCII is the only truly “plain” charset versus Unicode being the real modern standard.
  • Clarifications around encodings: Unicode as abstract code points, UTF‑8 vs UTF‑16 vs UTF‑32, and grapheme clusters that span multiple code points.
  • Some call Unicode a “mess,” especially around emojis, CJK unification, and sorting rules that can’t capture all cultural nuances (e.g., German umlaut vs French trema).
  • Others argue complexity is inevitable when supporting all human writing; emojis are defended as stress‑testing bad text assumptions and pushing Unicode adoption.

Plain Text Ecosystem: Tools, Formats, and Parsing

  • Examples of plain-text-centric tools: accounting systems (ledger, hledger, beancount), screenplay formats (Fountain), and engines using textual scene descriptions.
  • A new format, Brashtag, is presented as a simpler, more programmable alternative to Markdown, closer to “human‑writable tree data” than to JSON.
  • Some see Brashtag as redundant with existing formats; its creator emphasizes composability and easy tooling over built‑in rendering.
  • Questions about good tools for turning text plus grammar into parsers; suggestions include several Python and language-oriented parsing libraries.
  • Recognition that text requires parsing effort but is easier to inspect and reverse‑engineer than opaque binary blobs.

Tradeoffs: Text vs Rich/Binary and Future Durability

  • Plain text is praised as nearly universal, but binary formats are argued to be less scary now that AI agents can infer structure, decompress archives, and write custom tools.
  • In scientific archiving, text works well for tables but becomes unwieldy for large 2‑D image data, where an elegant long-term solution is still sought.
  • Some argue that because no single entity fully “owns” text, it remains resilient, despite standards bodies controlling ASCII/Unicode.
  • Broader media trend noted: many users and publishers prefer video over text despite text being far more compact and faster to consume.