Show HN: The load-bearing vocabulary of Claude

Who Should Read

Developers who find Claude-written PR descriptions and code comments awkward and want to tune the output style

Core Mechanics

• The source is only a title
  • There is no body text, so the site's method and numbers cannot be confirmed
  • The title is 'The load-bearing vocabulary of Claude'
• The topic is Claude's word habits
  • The title says it covers Claude's signature vocabulary
  • The specific word list is not in the source

Community Reactions

• The author says the data refreshes daily
  • The author's comment says GitHub Actions updates the data and analysis daily
  • They are adding a search bar and raising data to 1000 PRs per day
• Banning words to fix the style, per commenters
  • One commenter saw odd word choices like 'name(s)' as a verb, 'carries', 'verdict' while coding and banned certain words
  • Another added an Orwell rule to a global prompt and says Claude replied it conflicted with its system prompt
• Opinion that it is not just Claude
  • One commenter says Codex also uses load-bearing
  • Another says a different model's chat overuses 'unusually' and put it on a banned-words instruction
• Objections to the analysis scope
  • One commenter says only two clusters exist, Claudish and Spanish/French, so clustering English only would be better
  • Another asks for sentence-structure analysis, not just vocabulary

How to Apply

• Add a banned-word list to global instructions
  • Today, pick 5 Claude phrases that annoy you and add them to the system prompt as banned
  • List plain replacements next to them
• Spot-check PR text afterward
  • Search agent-written PR descriptions and comments for words like sidecar and load-bearing
  • Have Claude rewrite flagged sentences as concrete actions

Try It

# Paste into the system prompt/custom instructions of Claude or ChatGPT (simple version of a commenter's approach)

Follow these rules when writing.
- Use concrete actions and causes instead of metaphors or idioms.
- Banned words: load-bearing, crux, first-class citizen, {{extra banned words}}
- Do not omit what a noun refers to; write 'The abc function ...'.

Rewrite the text below under these rules.
{{문서}}

Next: search the rewrite yourself to check whether banned words remain.

Terminology

Related Papers

• AI is a tool and using it well takes skill

  A rebuttal to Hickey: LLMs are tools, and skill in directing and checking them decides the outcome

• Reflective Prompt Optimization of a Deep-Research Agent: A Topic-Disjoint, Multi-Seed Protocol for Measuring Transferable Gains and Prompt-Level Overfitting

  Reflective prompt optimization raised held-out citation reward from 0.702 to 0.781 on a deep-research agent.

• GPT‑6 and Intelligent UI for everyone

  GPT-6 reportedly adds generated UI to chat; with no article text, only comment reactions are available.

• Base Models Can Reason By Taking a Cue From Training Data

  Prefilling just two opening tokens lifted Olmo-3-7B's MATH-500 accuracy from 42% to 78%.

• Labels Override Definitions in Jev-Style Typed Decision Models

  Renaming options to A/B raised routing accuracy by +0.1511 over meaningful labels.

• How Much Prompt Is Enough? A Blackbox Minimization of Few-Shots in LLMs

  Delta debugging trimmed few-shot prompts by 65.3% on average with unchanged propositional output

Related Resources

• https://louisabraham.github.io/load-bearing/
• https://github.com/louisabraham/load-bearing