The Human in the Loop [01knvtp06bs16p928wzy44r429]

The Human in the Loop 🔗 **Source:** [The Human in the Loop](https://pento.net/2026/04/07/the-human-in-the-loop/) > [!note]+ Document Notes > tags: #artificial-intelligence #software-development #writing #wordpress #llm > author: Gary Pendergast > source: pento.net > reference: https://pento.net/2026/04/07/the-human-in-the-loop/ > date created: 2024-12-19 14:30 > > # 🏷️ Title: The Human in the Loop >  ## 🧰 Action Inbox - [H1] Evaluate LLM tools by asking where creativity lives in your process — not just "can an LLM do this?" #artificial-intelligence - [H2] Build LLM-assisted writing tools that augment rather than replace human creativity #wordpress - [H3] Test Claudaborative Editing for WordPress as human-centric writing assistance #writing ## 🧭 Table of Contents - [[#📝 Summary]] - [[#📖 Main Content]] - [[#🗒️ Notes]] - [[#📚 References]] - [[#🗣️ Paraphrase]] - [[#📰 Newsletter]] - [[#✅ Action Items]] - [[#🧪 Hypotheses]] - [[#📊 Methodology]] - [[#❓ Questions]] - [[#🔮 Future Research]] - [[#💡 Implications]] - [[#🧠 Learning Styles Assessment]] - [[#📊 Accessibility Assessment]] - [[#🌍 DEIB Assessment]] - [[#✅ Logical Validity]] - [[#🛡️ DARVO Analysis]] - [[#🎯 Conclusion]] ### 🔗 References - [Claudaborative Editing](https://pento.net/claudaborative-editing/) - [Gary Pendergast's Blog](https://pento.net/) ## 📝 Summary - **What happened**: Gary Pendergast explores the "Human in the Loop" paradigm for LLM tools, arguing that creativity location determines proper usage - **Why it matters**: Distinguishes between appropriate LLM assistance (coding implementation) and inappropriate replacement (creative writing) - **Key stakeholders**: Software developers, content creators, WordPress users, AI tool builders - **Risks and tradeoffs**: LLM-generated content lacks human creativity and produces measurably different "perplexity" patterns - **Bottom line**: LLMs should augment human creativity, not replace it—the key is identifying where creativity lives in each process ## 📖 Main Content Pendergast argues that LLMs have reached a new equilibrium where experienced engineers can guide them to produce high-quality code through small course corrections. This works because in software development, creativity lives in architectural decisions and specifications, not in the specific code implementation. However, this paradigm breaks down when applied to creative writing. When text itself is the creative act, using LLMs to generate content delegates away the human's creativity rather than augmenting it. The author introduces the concept of "perplexity"—a measure of text randomness—explaining why LLM-generated text feels off even when technically competent. The piece introduces Claudaborative Editing, a WordPress plugin designed to assist rather than replace human writing. This represents the author's vision for proper LLM integration: tools that enhance human creativity while keeping the creative process fundamentally human-driven. ## 🗒️ Notes #### Article Structure Analysis — [H1] Opening: LLM Evolution in Coding - Key idea: LLMs evolved from glorified autocomplete to tools that produce high-quality code with human guidance - Evidence: "Even 12 months ago, they were little more than glorified auto-complete tools" #### Core Philosophy — [H1] Code as Poetry vs Outcomes - Key idea: When LLMs write code, creativity shifts from implementation to architectural specification - Evidence: "You can write beautiful software by writing thoughtful specifications, instead" #### The Temptation Warning — [H2] Scope Creep Beyond Code - Key idea: Success with coding LLMs tempts overuse in inappropriate domains like creative writing - Evidence: "I could ask Claude to write 10 paragraphs... but would you have even read this far if you thought this post was LLM generated?" #### Reader Respect Principle — [H1] Human Writing as Creative Act - Key idea: Creative writing requires human authorship to maintain authenticity and respect for readers - Evidence: "When the text is the creative act, there's no way for the LLM to write the text without compromising your creativity" #### Technical Evidence — [H2] Perplexity and Uncanny Valley - Key idea: LLM text is measurably different due to lower randomness/perplexity - Evidence: "LLMs are... giant statistical models that are really good at figuring out 'what's the most likely bit of text to come next'" #### Solution Framework — [H2] LLM as Assistant, Not Author - Key idea: LLMs can help with writing process without generating the actual text - Evidence: "Claudaborative Editing... allowing me to plan, write, review, and publish... every word of it was written by me alone" #### Decision Framework — [H1] Where Does Creativity Live? - Key idea: Evaluate LLM appropriateness by identifying where creativity exists in each process - Evidence: "For software, that's in the design decisions... For a blog post... the creativity lives in the act of writing" #### Core Principle — [H1] Augment vs Replace Philosophy - Key idea: Best LLM tools enhance rather than suppress human creativity - Evidence: "the best uses of LLM tools are when they augment humans, rather than try to replace them" ## 📚 References - [The Human in the Loop](https://pento.net/2026/04/07/the-human-in-the-loop/) - [Claudaborative Editing Plugin](https://pento.net/claudaborative-editing/) - [Research on LLM vs Human Text Perplexity](https://pento.net/2026/04/07/the-human-in-the-loop/) ## 🗣️ Paraphrase - LLMs have evolved to effectively assist experienced developers in producing quality code by responding well to human guidance and course corrections - The success with coding applications creates temptation to apply LLMs inappropriately to creative writing, where the human authorship is the essential creative element - Proper LLM integration involves building tools that support human creative processes rather than replacing human creativity entirely ## 📰 Newsletter WordPress developers and content creators now have a clearer framework for evaluating AI tools. Gary Pendergast's "Human in the Loop" principle helps distinguish between appropriate LLM assistance (like code generation where creativity lives in architecture) and inappropriate replacement (like content writing where creativity lives in the text itself). Key insight: Ask "where does creativity live?" before implementing LLM tools. For WordPress development, LLMs can handle implementation while humans focus on design decisions. For content creation, LLMs should assist the writing process without generating the actual content. Pendergast's Claudaborative Editing plugin demonstrates this philosophy by integrating AI assistance directly into the WordPress editor for planning, reviewing, and improving human-written content. Resources: [Learn WordPress](https://learn.wordpress.org/), [Developer Docs](https://developer.wordpress.org/), [WP-CLI](https://wp-cli.org/) ## ✅ Action Items #### WordPress - Core AI: Evaluate Claudaborative Editing approach for core AI features, ensuring human creativity remains central - Training: Develop lesson plans teaching developers how to identify appropriate LLM use cases using the "where does creativity live?" framework - Documentation: Create guidelines for WordPress plugin developers on human-centric AI tool design - Community: Foster discussion around ethical AI integration that augments rather than replaces human creativity #### DevRel - Developer advocacy: Present talks on the "Human in the Loop" paradigm at WordCamps and developer conferences - Education: Create workshops teaching developers to guide LLMs effectively for code generation while maintaining architectural creativity - Developer experience: Evaluate WordPress development tools to ensure they follow augment-not-replace principles - Community management: Moderate discussions about AI tool integration with focus on preserving human creativity #### Community - Contributors: Establish guidelines for AI-assisted contributions that maintain human authorship and creativity - Sponsors: Support projects like Claudaborative Editing that demonstrate ethical AI integration - Ecosystem: Coordinate cross-project standards for human-centric AI tool development ## 🧪 Hypotheses - LLM-assisted development tools produce higher quality outcomes when humans focus on architectural decisions rather than implementation details - Content created with AI assistance (but human authorship) will have higher reader engagement than fully AI-generated content - WordPress plugins following the "augment not replace" principle will see better adoption than those that automate away human creativity - Developer productivity increases more with LLM tools that preserve human decision-making authority over creative aspects ### 📊 Methodology - A/B test WordPress sites with human-authored vs LLM-generated content measuring engagement metrics and reader retention - Survey developers using Claudaborative Editing vs traditional AI writing tools on creativity preservation and output quality - Measure perplexity scores across WordPress content to identify LLM-generated vs human-authored posts ### ❓ Questions - How can WordPress core AI features implement the "creativity location" framework to ensure appropriate human involvement? - What measurable criteria beyond perplexity can identify the boundary between human and LLM creativity? - How might the WordPress.org plugin directory evaluate and categorize AI-assisted vs AI-generated tools? ### 🔮 Future Research - Longitudinal study of developer skills evolution when working with human-in-loop LLM tools - Investigation into reader perception and trust metrics for AI-assisted vs traditional content - Development of WordPress-specific guidelines for ethical AI tool integration ### 💡 Implications - **Practice**: WordPress developers should evaluate AI tools by identifying where creativity lives in their specific workflow - **Policy**: WordPress.org may need guidelines distinguishing between human-augmenting and human-replacing AI tools - **Education**: Developer training should emphasize architectural thinking and LLM guidance skills over memorizing syntax - **Funding**: Prioritize AI tools that enhance rather than replace human creativity in WordPress ecosystem ### 🧠 Learning Styles Assessment Reading level estimate: 10th grade (Flesch-Kincaid) #### 👀 Visual learners - Present: Clear section headings, philosophical framework structure - Improve: Diagrams showing creativity location in different processes, before/after examples #### 🔊 Aural learners - Present: Conversational tone, quotable principles - Improve: Audio version of post, podcast discussion of concepts #### 🗣️ Verbal learners - Present: Clear definitions of perplexity, analogies like "uncanny valley" - Improve: More concrete examples of good vs poor LLM integration #### 🧩 Kinesthetic learners - Present: Claudaborative Editing plugin as hands-on example - Improve: Step-by-step guide for implementing "creativity location" evaluation #### 🧠 Logical learners - Present: Clear decision framework, evidence-based arguments about perplexity - Improve: Flowchart for evaluating LLM tool appropriateness #### 🧑‍🤝‍🧑 Social learners - Present: Comment section, community discussion prompts - Improve: Forums for discussing AI ethics in WordPress development #### 🧘‍♀️ Solitary learners - Present: Reflective questions about creativity location - Improve: Self-assessment worksheet for evaluating personal AI tool usage #### 🌿 Naturalistic learners - Present: Ecosystem thinking about AI tool impacts - Improve: Analogies to natural symbiosis vs parasitism ### 📊 Accessibility Assessment #### Screen reader semantics - Present: Proper heading hierarchy, clear article structure - Improve: Alt text for any future diagrams, ARIA labels for plugin interfaces #### Keyboard navigation - Present: Standard blog post navigation - Improve: Ensure Claudaborative Editing plugin is fully keyboard accessible #### Contrast and readability - Present: Clean typography, good contrast ratios - Improve: Test at various zoom levels, support high contrast mode #### Text resizing and reflow - Present: Responsive design handles text scaling - Improve: Verify no content clipping at 200% zoom #### Media and captions - Present: Text-only post, no media accessibility issues - Improve: Add captions to any future video demonstrations #### Standards reference - Follows WCAG 2.2 AA guidelines for text content ### 🌍 DEIB Assessment #### Cultural breadth - Present: Universal principles applicable across cultures - Improve: Examples from diverse development communities, non-English WordPress usage #### Inclusive language - Present: Gender-neutral terms, accessible technical language - Improve: Define jargon like "perplexity" more thoroughly #### Neurodiversity support - Present: Clear headings, logical structure - Improve: Visual breaks between concepts, consistent formatting #### Representation - Present: Individual perspective from experienced developer - Improve: Acknowledge different developer experience levels, diverse use cases #### Psychological safety - Present: Respectful tone about AI usage choices - Improve: Acknowledge legitimate concerns about AI impact on employment ### ✅ Logical Validity #### Strong reasoning - Deductive strength: Strong - "creativity location determines appropriate LLM usage" applied consistently - Inductive strength: Moderate - generalizes from coding experience to writing, needs more evidence - Causal reasoning: Good - explains why LLM text feels different through perplexity concept - Analogical strength: Effective - "Code is Poetry" analogy clarifies creativity concepts #### Fallacies - False dichotomy: Slightly presents augment vs replace as binary when there may be spectrum - Appeal to consequences: "Would you have read this if LLM generated?" - valid but could acknowledge other perspectives #### Tone of voice - Overall tone: Thoughtful, experienced, slightly prescriptive but respectful of reader intelligence ### 🛡️ DARVO Analysis - Deny: No defensive denials present - Attack: No attacks on opposing viewpoints - Reverse victim/offender: Not applicable - Emotional cues and rhetoric: Professional and measured, uses logical persuasion rather than emotional manipulation ## 🎯 Conclusion Pendergast provides a practical framework for WordPress developers and content creators to evaluate AI tools appropriately. The core insight—asking "where does creativity live?"—offers clear guidance for when LLM assistance enhances versus undermines human creative work. His Claudaborative Editing plugin demonstrates how to build AI tools that preserve human authorship while providing valuable assistance. This approach could guide WordPress.org policies and community standards for ethical AI integration. ## 🧱 Refactor Appendix ### 📚 Literature Split Pad ### [H1] LLMs work best when creativity lives in specification not implementation > [!quote] [The Human in the Loop](https://pento.net/2026/04/07/the-human-in-the-loop/) by [[Gary Pendergast]] via [[pento.net]]. Published 2026-04-07. Accessed 2024-12-19. > "You can write beautiful software by writing thoughtful specifications, instead." Explains why LLM coding tools succeed - creativity shifts to architectural level. [view highlight](https://pento.net/2026/04/07/the-human-in-the-loop/#H1) #xref #articleseed ID:rw_creativity_specification ### [H2] LLM text has measurably lower perplexity than human writing > [!quote] [The Human in the Loop](https://pento.net/2026/04/07/the-human-in-the-loop/) > "LLMs are... giant statistical models that are really good at figuring out 'what's the most likely bit of text to come next'" Provides technical explanation for why LLM writing feels "off" to readers. [view highlight](https://pento.net/2026/04/07/the-human-in-the-loop/#H2) #xref ID:rw_perplexity_detection --- ### 🔖 Atomic Split Pad (Zettel Candidates) #### [Z1] Evaluate AI tools by identifying creativity location - Framework applies to any human-AI collaboration decision - Referenced throughout post as core decision principle - Links: [[Human-AI-Collaboration]] [[Creative-Process-Design]] #### [Z2] Reader respect requires human authorship for creative content - Principle that LLM generation violates implicit author-reader contract - Supported by perplexity research and engagement assumptions - Links: [[Content-Ethics]] [[Author-Authenticity]] --- ### ✏️ Fleeting Queue - [F1] Test Claudaborative Editing plugin for WordPress writing workflow — https://pento.net/claudaborative-editing/ #fleeting #inbox - [F2] Research perplexity measurement tools for detecting LLM content — mentioned in post #fleeting #inbox - [F3] Explore "creativity location" framework for other WordPress AI features — core principle #fleeting #inbox