AI;DR (AI; Didn't Read)
TL;DR Highlight
A 'don't read unedited AI text' policy called AI;DR hit #1 on Hacker News
Who Should Read
Developers and team leads whose Slack, PRs and docs are flooded with unedited AI output
Core Mechanics
- Unedited text goes unread
- The author won't read AI output that its sender didn't review or edit.
- He says he physically flinches when a respected person sends raw AI output.
- Using AI isn't the problem
- The author calls himself about as pro-AI as you can be.
- He accepts AI for ideas, outlines and prose refining as normal in Q3 2026.
- Customer support is an exception
- He names customer support as a place where 100% AI copy is expected.
- A wall of Claude output in a Slack discussion sends a different message than intended.
- Your name means your responsibility
- Newsletters and social posts carry your name, so you own the AI-isms too.
- Otherwise the reader could just ask Claude directly.
- AI;DR is the TL;DR for slop
- TL;DR was the answer for social media, AI;DR is the answer to AI slop.
Community Reactions
- PRs flooded with AI docs
- One commenter says coworkers dump hundreds of lines of AI docs in every PR and add AI comments on nearly every line.
- Features ship and metrics improve, but the codebase is described as post-readability.
- Send the prompt instead
- Several commenters argue people should share the prompt instead of the AI output.
- One says they ask coworkers for the prompt and model so they can own the session.
- Quality is the bar, a counterview
- Some commenters say they don't care if AI wrote it as long as it's good.
- One says only about 10% of AI writing is worth reading and 90% wastes time.
- Questions exposed the gap
- One commenter asked an article's author a question and got no answer because the examples weren't from real experience.
- Another cites an AI-written PCIe over TCP/IP post that skipped DMA and interrupts but made big claims.
How to Apply
- Attach the prompt and model to AI output
- Add a line to your team's Slack or PR template today: share the prompt and model when using AI.
- Put a 3-line summary you wrote yourself above any long AI output.
- Write PR descriptions yourself
- Explain why the change was made in your own words.
- Move long AI docs to a separate file or delete them before review.
- Check outward-facing text before publishing
- Edit AI-isms out of newsletters, release notes and blogs yourself.
- Cut any claim you couldn't answer questions about.
Try It
# Paste into Claude or ChatGPT (a simple review prompt applying the article's policy)
The {{draft}} below is an AI-written Slack/PR message. Edit it to save the reader's time.
{{draft}}
Instructions:
1. Summarize the key message in 3 lines or fewer.
2. Mark exaggerations, filler and sentences that read like AI-isms.
3. List claims I must verify myself and claims with no evidence.
Next: read the result, rewrite it in your own voice, then send it.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