IdeaWell

Project ideas from Hacker News discussions.

Show HN: The load-bearing vocabulary of Claude

Original Article

Hacker News Discussion

📝 Discussion Summary (Click to expand)

4 Prevalent Themes in Discussion About Claude's Language

  1. Frustration with Incomprehensible Jargon and Overuse of Specific Phrases
    Users consistently complain that Claude's output is filled with confusing, repetitive terminology ("load-bearing seam," "fibred," "capstone") that obscures meaning and feels artificially inflated.

    "The fibred side folded its capstone into the existing name, so the kinds are asymmetric." What on earth does it mean to fold a capstone into a name‽ — black_knight
    "You need a PhD to understand its explanation of a code snippet." — fny
    "It just overuses jargon that has NO meaning. Sometimes, it actually hand waves too much as well while trying to dumb down stuff for you." — Labo333

  2. Speculation About Root Causes (RLHF, Training Artifacts, Agentic Behavior)
    Many theorize that Claude's style stems from reinforcement learning (RLHF) rewarding novelty, internal agent communication, or training on AI-generated text, rather than inherent intelligence.

    "I suspect it's the same thing going on in the hidden thinking tokens." — fwip (referring to agent summarization)
    "My unprovable pet theory is that, especially for writing about technical topics, the RL process has an open-ended way to weight things for quality..." — zbentley
    "It's like it has a built in e bullshit generator." — swader999 (linking to bullshitgenerator.com)

  3. Debate Over Whether This Represents a Problem or Just Different Communication
    Some argue Claude's style is actively harmful and non-human, while others see it as neutral or even beneficial for technical precision, noting humans may adapt to it.

    "This represents a great argument against the somewhat frequent claim that 'AI writing patterns reflect human writing patterns'. Rather, AI writing is distinctly not human..." — bakugo
    "I suspect, as we continue forward, humans will slowly start to adopt the language of LLMs..." — nater5000 (noting his own adoption of "... such as , , etc., ...")
    "Is it actually better though? Can you even measure it? Unguided, Claude will just produce mediocrity..." — incrudible

  4. Practical Observations, Workarounds, and Broader Impacts
    Users share coping strategies (output styles, switching models), note real-world adoption of Claude-isms in workplaces, and discuss implications for documentation and communication.

    "They just addressed this with 'Output styles'" — josefresco (linking to Claude's documentation)
    "In the enterprise, we are already adopting them. 'Seam'... is now not only our defacto way to describe the boundary..." — threecheese
    "I've found myself rewriting my own words to avoid Claudisms (because they can be cringe and I don't want people to think I'm copy/pasting Claude output at them)." — cpeterso
    "I just subscribed to ChatGPT yesterday, as I've become impatient with Claude for a text-dense project I'm working on." — rubslopes

🚀 Project Ideas

PlainClaude: AI-to-Human Language Rewriter

Summary

  • Takes Claude's verbose, jargon‑heavy output and rewrites it into clear, concise plain English while preserving technical meaning.
  • Uses prompt chaining or a lightweight fine‑tuned model to strip Claude‑isms such as “load‑bearing”, “vacuous”, and over‑used metaphors.

Details

Key Value
Target Audience Developers, engineers, technical writers who rely on Claude for code explanations, documentation, or debugging
Core Feature Rewrite Claude output to plain language via API or browser extension
Tech Stack Python, FastAPI, HuggingFace Transformers (LoRA adapters), optional Claude API for rewriting
Difficulty Medium
Monetization Revenue-ready: subscription tier ($5/mo) for higher usage, free tier limited to 50 rewrites/day

Notes

  • Users complained that Claude’s explanations “require a PhD to understand” (fny: “You need a PhD to understand its explanation of a code snippet”).
  • Could spark discussion on reducing AI jargon and improving readability of LLM‑generated content.

ClaudeJargonSpotter: Real‑time Highlighter & Suggestion Engine

Summary

  • Browser extension that highlights Claude‑specific words (load‑bearing, seam, vacuous, provenance, etc.) in any webpage or editor.
  • Provides one‑click suggestions for simpler synonyms drawn from a community‑curated list.

Details

Key Value
Target Audience Anyone who reads or writes with Claude output (developers, PMs, researchers)
Core Feature Detect and highlight Claude‑isms, offer alternatives
Tech Stack JavaScript/TypeScript, WebExtensions, local word‑list + fuzzy matching
Difficulty Low
Monetization Hobby (open source, optional donations via GitHub Sponsors)

Notes

  • Commenters noted the overload of terms like “load‑bearing” and “vacuous” making text annoying (sroussey, halfmatthalfcat).
  • Could become a community‑maintained glossary, fostering discussion on AI language drift.

ClaudeStyleAdapter: LoRA Fine‑tuning Service for Personal Claude

Summary

  • Lets users upload their own writing samples (emails, docs, code comments) to train a lightweight LoRA adapter that steers Claude toward their preferred style and away from Claude‑isms.
  • Exposes the adapter via a simple API wrapper for seamless integration.

Details

Key Value
Target Audience Power users and teams that heavily rely on Claude for internal communications
Core Feature Upload samples → train LoRA → download or use via API
Tech Stack Python, PEFT library, HuggingFace, FastAPI, cloud storage (S3)
Difficulty High
Monetization Revenue-ready: pay‑per‑training job ($2) + monthly hosting fee

Notes

  • Users found built‑in output styles insufficient (torarnv: “It still spits out incomprehensible gibberish”; sunaookami: “hasn't changed the output for me at all”).
  • Enables teams to customize Claude's voice, reducing alien jargon and increasing adoption.

GitHub PR Claude‑ism Cleaner Action

Summary

  • GitHub Action that scans pull request titles, bodies, and review comments for Claude‑specific jargon and suggests plain‑language replacements before merging.
  • Optionally auto‑rewrites using a lightweight model.

Details

Key Value
Target Audience Engineering teams using GitHub and Claude for code reviews
Core Feature Detect Claude‑isms in PR metadata, post review comments with suggestions
Tech Stack Node.js, GitHub Actions toolkit, regex/Trie for word detection, optional call to a small LLM for rewrite
Difficulty Medium
Monetization Hobby (free open‑source action); premium hosted version with higher limits available

Notes

  • Several commenters wanted commit logs to be more human (fnj, jfultz) and disliked AI‑sloppy PR comments.
  • Would improve PR hygiene and reduce AI‑generated noise in codebase history.

Claude‑Speak Lexicon: Interactive Reference & Learning Tool

Summary

  • Web app that catalogs Claude‑specific terms (load‑bearing, seam, vacuous, provenance, etc.) with definitions, example usage, and frequency trends from the load‑bearing dataset.
  • Includes flash‑card mode to help users recognize and avoid them.

Details

Key Value
Target Audience Curious developers, linguists, anyone wanting to understand or avoid Claude‑isms
Core Feature Searchable dictionary, usage graphs, quiz mode
Tech Stack React, Vite, existing load‑bearing dataset JSON, hosted on Vercel
Difficulty Low
Monetization Hobby (open source, optional sponsorship)

Notes

  • Users asked for a search bar for terms (simlevesque, khatkhati) and a dictionary (clbrmbr).
  • Could become a go‑to resource for spotting AI slop, fostering discussion on language evolution.

Read Later