# Write Drunk, Edit Sober: Creating Generative Content Responsibly - Matthijs van der Veer

## Executive summary

The talk addresses the proliferation of 'AI Slop'—low-quality, inauthentic content generated by LLMs—and provides a framework for developers to build automated systems that enforce quality. The core philosophy is shifting focus from merely generating content ('Write Drunk') to defining clear intent and implementing rigorous review processes ('Edit Sober'). Practical solutions involve using structured data inputs (like GitHub Issues) to capture user intent, applying Natural Language Processing (NLP) techniques like lemmatization for stylistic checks, and building automated quality gates that flag common LLM patterns (e.g., excessive emojis, M dashes, or generic corporate tropes).

## Key takeaways

- Identify 'AI Slop' Patterns: Common signs of low-quality AI content include overuse of emojis, the M dash (`—`), bold text, overly positive/generic language, and vague phrasing like 'In an era...' (0:023 - 0:045).
- Focus on Intent over Content: The most critical step in generating quality AI content is defining the core intent: What do you want the audience to learn? Who should be here? And what can they accomplish? This structured approach guides the LLM (0:158 - 0:236).
- Automated Quality Gates are Essential: Developers can build tools that enforce quality by implementing deterministic checks. Techniques include using NLP libraries like `spacy` for lemmatization to identify overused or non-standard vocabulary, and running pattern matching against known 'tropes' (0:475 - 1:239).
- Mitigate Automation Bias with Metrics: To combat the tendency to overly trust machine output (automation bias), systems should provide explicit metrics, such as a confidence score or probability of error, rather than simply offering recommendations. Offering raw information is safer than recommending an action (1:390 - 2:150).

## Technical details

- NLP and Lemmatization: Using NLP libraries (e.g., `spacy`) to perform lemmatization—reducing words to their base form (e.g., 'deployments' to 'deployment'). This is a deterministic method for identifying overused or non-standard vocabulary patterns in generated text (0:523 - 1:046).
- Structured Workflow Design: Building content generation tools around structured inputs, such as GitHub Issues. By forcing the user to answer specific questions about audience and intent before generating an abstract, the process captures necessary context that LLMs often miss (1:098 - 1:245).
- Bias Mitigation in AI Systems: To reduce automation bias, systems should prioritize displaying raw information (e.g., search results) rather than providing synthesized recommendations based on those results. This forces the user to perform critical thinking and review (1:370 - 2:050).
- Deterministic vs. Generative Logic: The speaker advocates for using deterministic code logic (like NLP or string comparison) for quality checks, arguing that these methods are more reliable and predictable than relying solely on LLM output for validation (2:285 - 2:410).

## Practical implications

- Implement mandatory structured inputs (e.g., required fields for audience and intent) before allowing content generation.
- Integrate NLP checks into CI/CD pipelines to automatically flag stylistic issues like excessive M dashes or overused corporate jargon.
- Design user interfaces that present raw, verifiable information rather than synthesized recommendations to minimize automation bias.
- Treat LLM output as a first draft; the 'Edit Sober' phase must involve human-in-the-loop review and manual rewriting.

## Topics

Generative AI, Natural Language Processing (NLP), Content Quality Assurance, Software Development Workflow, Automation Bias, GitHub Issues/PRs, spacy (Python library), tropes.fyi, Slop Slurp

Source: https://www.youtube.com/watch?v=e4Ig3w1dG-8