Write Drunk, Edit Sober: Creating Generative Content Responsibly
summary
Shows that 15% of incoming emails used the M‑dash by Dec 2025 and that 30% of tech abstracts spiked with the word delve after GPT’s rise, then demonstrates concrete fixes: a GitHub-driven session abstract pipeline called Mandrean Muse that runs prompts at temperature 0.8, posts PRs, performs an intent check, and a validator slop slurp that runs 33 trope checks from tropes.fyi plus spaCy lemmatization to flag offending words for deterministic sober edits.
AI-written summary · about
chapters (10)
- 0:00Introduction
- 1:48AI slop in professional communication
- 9:11Why we embraced LLMs too quickly
- 15:51Building responsible abstract generators
- 23:12Editing sober: removing AI slop
- 29:12Automation bias with AI content
- 38:19Why mitigation strategies fail
- 44:43Authenticity and review in content
- 50:19Tools to detect AI slop
- 55:35Call to action and Q&A
related talks
12mNavigating the World of AI Development Without Going Insane
35mBuilding AI-powered user interfaces
37mHow ChatGPT Got Me Drunk
43mComedy Writing With Small Generative Models
01h 08mIn Prompts We Trust
37mBuilding AI-powered user interfaces
55mAI Showdown: The Good, The Glitchy, and The Risky
39mLower the Barrier of Entry: Empowering Non-writer Contributions To Docs