The Majority AI View 🔗 **Source:** [The Majority AI View](https://www.anildash.com//2025/10/17/the-majority-ai-view/) > [!note]+ Document Notes > tags: [AI, Tech industry, Ethics, Governance, Hype cycles] > author: Anil Dash > source: anildash.com > reference: [https://www.anildash.com//2025/10/17/the-majority-ai-view/] > date created: 2025-10-20 10:00 > > # 🏷️ Title: The Majority AI View > <!-- deep-read:sections start --> ## 🧭 Table of Contents - [📝 Summary](#-summary) - [📖 Main Content](#-main-content) - [🗒️ Notes](#️-notes) - [📚 References](#-references) - [🗣️ Paraphrase](#️-paraphrase) - [📰 Newsletter](#-newsletter) - [✅ Action Items](#-action-items) - [🧪 Hypotheses](#-hypotheses) - [📊 Methodology](#-methodology) - [❓ Questions](#-questions) - [🔮 Future Research](#-future-research) - [💡 Implications](#-implications) - [🧠 Learning Styles Assessment](#-learning-styles-assessment) - [📊 Accessibility Assessment](#-accessibility-assessment) - [🌍 DEIB Assessment](#-deib-assessment) - [✅ Logical Validity](#-logical-validity) - [🛡️ DARVO Analysis](#-darvo-analysis) - [🎯 Conclusion](#-conclusion) ### 🔗 References - Include links cited in or relevant to the article. ## 📝 Summary - What happened: The author argues most working technologists see AI/LLMs as useful but overhyped, coercively pushed, and discussed in ways that ignore valid critiques and alternatives. - Why it matters: Hype crowds out pragmatic adoption and “good AI” paths (consent, sustainability, decentralization), narrowing the future. - Key stakeholders: Tech workers, managers, Big AI firms, creators, policymakers, users. - Risks and tradeoffs: Chilling effects from layoffs and conformity; reputational, regulatory, and social harms if hype dictates strategy. - Bottom line: Treat AI as normal tech—subject to standard scrutiny, controls, and sober evaluation—rather than a cult or inevitability. ## 📖 Main Content - Core arguments: - Working technologists broadly agree: “Technologies like LLMs have utility,” but overhype and pressure “make it very difficult to focus on legitimate uses.” - Media amplifies Big AI narratives while sidelining insiders’ critiques—even though many early warnings came from within big companies. - Power dynamics and layoffs chill dissent; moderate viewpoints feel career-risky. - Ignoring the moderate majority constrains better futures: consentful data, greener systems, decentralized architectures. - Context and definitions: - LLMs: Large language models that generate text and can “hallucinate” incorrect outputs. - “Normal technology”: Framing AI like other software—evaluated with standard risk, ethics, and governance practices. - Uncertainties or contested claims: - The exact size of the “majority” is asserted qualitatively. - Feasibility timelines for consentful, sustainable, decentralized AI vary by domain. ### 🗒️ Notes - “Most people who actually have technical roles... are fluent in the latest technologies like LLMs” but aren’t the “big, loud billionaires.” - “Technologies like LLMs have utility, but the absurd way they've been over-hyped... make it very difficult to focus on legitimate uses.” - Desire: treat AI as “normal technology” (hat tip to Arvind Narayanan and Sayash Kapoor). - Media largely repeats Big AI narratives; insider critiques get muted, despite whistleblowers and fired researchers. - Extremist leadership narrows possibilities; alternatives are possible: consent-based data, sustainability, decentralization. - Layoffs and fear enforce conformity; moderate stances feel risky. - “Very few agree with the hype bubble... the mainstream of tech culture is thoughtful, nuanced and circumspect.” ## 📚 References - The Majority AI View (https://www.anildash.com//2025/10/17/the-majority-ai-view/) - AI Snake Oil by Arvind Narayanan and Sayash Kapoor (https://www.aisnakeoil.com/) - OpenAI (https://openai.com/) - Anthropic (https://www.anthropic.com/) - Google AI (https://ai.google/) ## 🗣️ Paraphrase - Paragraph A: Most practitioners know LLMs and find them useful in places, but hype and coercion obstruct real value. - Paragraph B: Media boosts Big AI stories, sidelining insider critiques, which limits better paths like consentful and sustainable AI. ## 📰 Newsletter - This week: a grounded take on AI adoption for builders. The piece argues AI is “normal tech,” not a religion—focus on clear use cases, consentful data, sustainability, and decentralization. For WordPress folks: pilot AI where it measurably improves editorial workflows, translation, or support triage, and skip the hype. Useful reads: AI Snake Oil’s pragmatic analyses and Core Performance updates. Keep governance tight: data provenance, model evals, rollback plans. ## ✅ Action Items #### DevRel - Publish pragmatic AI integration guides with measurable KPIs and rollback plans. - Run workshops on evaluation sets, prompt hygiene, and error handling. - Maintain examples and quickstarts focused on high-SNR use cases. - Track campaign attribution to real developer adoption, not vanity metrics. - Moderate forums to welcome skeptical, evidence-based feedback. #### OSPO - Define contribution policies for AI-assisted code and provenance. - Enforce license scanning and AI training-data compliance. - Establish vuln intake and model/security embargo procedures. - Maintain SBOMs including model artifacts and data sources. - Offer consultation hours on AI governance and legal guardrails. - Fund upstream projects enabling consentful datasets and verification. - Report AI governance KPIs to executives quarterly. #### Community - Maintainers: add evaluation gates; document non-goals. - Contributors: first-issue rotations on testing/evals. - Sponsors: fund consentful data pipelines and audits. - Ecosystem: practice coordinated disclosures for AI incidents. #### WordPress - Core: define AI-related triage and acceptance criteria. - Core Performance: track latency/cost regressions from AI features. - Core AI: guardrails, eval sets, phased pilots, sunset clauses. - Core Program Team: set cross-team rituals and retros. - Mobile: offline-first fallbacks; error UX for AI failures. - Playground: reproducible AI bug labs and sandboxes. - Community: meetup kits on pragmatic AI. - Training: lesson paths on evaluation and governance. - Marketing: avoid hype; publish case studies with metrics. - Support: SLAs for AI misclassification and escalations. - Design: AI feature discoverability, consent prompts. - Documentation: examples-first, failure modes, limits. - Photos/Openverse: provenance, opt-out/opt-in pathways. - Accessibility: review AI impacts on WCAG conformance. - Meta/CLI/Hosting/Tide/Themes/Plugins: provenance checks, telemetry, scoring transparency. #### The WPOCC - Programs: stipends for eval-set maintainers. - Projects: governance clinics; “AI incident” tabletop drills. - Funding: matched grants for consentful data initiatives. ### FAIR - Technical independence: local signatures, provenance checks. - Policies: funding transparency, sunset clauses, anti-capture. - Productized governance: charters, operator agreements, scorecards. ## 🧪 Hypotheses - H1: Teams using “AI as normal tech” governance see fewer production incidents than hype-driven adopters. - H2: Consentful training data reduces legal risk without materially degrading task performance in targeted domains. - H3: Clear evaluation sets improve AI feature retention and NPS. - H4: Public layoffs correlate with reduced employee willingness to voice AI concerns. - H5: Decentralized AI deployments reduce single-vendor outage impact. ### 📊 Methodology - H1/H3: Controlled rollout with A/B teams; track incident rate, NPS, retention. - H2: Compare models trained on consentful vs. mixed data; measure accuracy, legal events. - H4: Survey analysis pre/post layoff events; anonymity and longitudinal tracking. - H5: Chaos experiments simulating vendor outages; measure recovery time. ### ❓ Questions - Where do AI features actually outperform status quo with acceptable risk? - How to operationalize consent and provenance at scale? - What incentives shift media coverage toward practitioner evidence? ### 🔮 Future Research - Standardized evaluation suites for CMS/editorial AI tasks. - Energy and carbon accounting standards for AI features. - Federated, decentralized inference patterns for the web. ### 💡 Implications - Practice: ship evals, kill-switches, and measurements by default. - Policy: require provenance, opt-out/opt-in, and energy disclosures. - Security: integrate AI incident response into existing playbooks. - Education: train contributors on limitations and failure modes. - Funding: prioritize consentful datasets and open evaluations. ### 🧠 Learning Styles Assessment - Reading level: Flesch-Kincaid Grade ≈ 9–10 (conversational, opinionated). #### 👀 Visual learners - Improve: add diagrams of AI governance flows and data provenance. #### 🔊 Aural learners - Improve: 3-minute audio summary with key quotes. #### 🗣️ Verbal learners - Present: strong analogies (“normal technology”); add glossary of terms. #### 🧩 Kinesthetic learners - Improve: hands-on evaluation exercises and rollback drills. #### 🧠 Logical learners - Present: causal links; add decision trees and KPIs. #### 🧑‍🤝‍🧑 Social learners - Present: community framing; add moderated Q&A templates. #### 🧘‍♀️ Solitary learners - Improve: self-checklists for AI readiness. #### 🌿 Naturalistic learners - Improve: lifecycle carbon comparisons and green patterns. ### 📊 Accessibility Assessment #### Screen reader semantics - Improve: ensure descriptive alt text for any AI UI. #### Keyboard navigation - Improve: guarantee tab reachability for AI controls. #### Contrast and readability - Present: prose-readable; ensure 4.5:1 in UI mockups. #### Text resizing and reflow - Improve: respect prefers-reduced-motion in AI interactions. #### Media and captions - Improve: transcripts for any demos; avoid autoplay. #### Forms and interactions - Improve: inline validation and forgiving inputs for prompts. #### Standards reference - Map gaps to WCAG 2.2 AA: 1.4.*, 2.1.*, 3.3.*, 2.2.2. ### 🌍 DEIB Assessment #### Cultural breadth - Improve: include global labor and data-consent cases. #### Inclusive language - Present: jargon-light; add plain-language callouts. #### Neurodiversity support - Improve: chunk content with consistent headings. #### Representation - Improve: quote a broader range of practitioners. #### Psychological safety - Present: names chilling effects; add reporting pathways. ### ✅ Logical Validity #### Strong reasoning - Inductive claim: widespread practitioner sentiment based on author’s interactions; aligns with observed layoff-chill dynamics. - Causal claim: hype crowds out better alternatives; plausible via resource allocation and media incentives. #### Fallacies - Possible hasty generalization: “virtually 100% of tech experts I talk to.” Fix: cite surveys or broader samples. #### Tone of voice - Direct, candid, occasionally exasperated (“stop being so goddamn creepy and weird about the technology!”). ### 🛡️ DARVO Analysis - Deny: Not applicable; the author doesn’t deny harms, he foregrounds them. - Attack: Critiques “authoritarian extremists” and “hype” rather than individuals’ credibility. - Reverse victim/offender: Not evident; workers’ fear is contextualized, not weaponized. ## 🎯 Conclusion - Bottom line: AI should be treated as normal tech—useful, limited, governed. Ignore hype; invest in consent, sustainability, decentralization, and measurement. - Next steps: Define governance, ship evals and kill-switches, pilot narrow use cases, publish metrics, and create safety for dissenting practitioner input. ## 🧱 Refactor Appendix (append after 🎯 Conclusion) ### 📚 Literature Split Pad - Purpose: Stage Konik CEE units that become per-source Literature notes. - Split level: H3. - Include verbatim quotes under 25 words and a [view highlight] link. ### [H1] Practitioners view AI as normal tech > [!quote] [The Majority AI View](https://www.anildash.com//2025/10/17/the-majority-ai-view/) by [[Anil Dash]] via [[anildash.com]]. Published 2025-10-17. Accessed 2025-10-20. > “What we all want is for people to just treat AI as a ‘normal technology’...” Context: practitioner sentiment vs. hype. Use: governance framing. [view highlight](https://www.anildash.com//2025/10/17/the-majority-ai-view/#H1) #xref #articleseed ID:rw_[MISSING] ### [H2] Hype crowds out legitimate use > [!quote] [The Majority AI View](https://www.anildash.com//2025/10/17/the-majority-ai-view/) > “Over-hyped… and the insistence on ignoring the many valid critiques… make it very difficult to focus on legitimate uses…” Context: prioritize measured deployments. Use: evaluation gates. [view highlight](https://www.anildash.com//2025/10/17/the-majority-ai-view/#H2) #xref ID:rw_[MISSING] --- ### 🔖 Atomic Split Pad (Zettel Candidates) - Purpose: Promote reusable claims into Atomic notes. - Split level: H4. - Each item: one idea, present tense, specific, testable. #### [Z1] Governance-first AI reduces incident risk - Scope: production features with evals and rollbacks. - Pointer: article claims hype harms; test via incident metrics. - Links: [[AI Governance]] [[Operational Risk]] #### [Z2] Consentful datasets lower legal exposure - Scope: commercial deployments with publisher data. - Pointer: claims on consent; test via legal events rate. - Links: [[Data Provenance]] [[AI Compliance]] --- ### ✏️ Fleeting Queue (Extract by Selection) - Purpose: Capture quick actions or sparks for Fleeting notes. - Extraction: Select one line → “Refactor: Extract selection to new note (content only).” - Keep each line short and self-contained. - [F1] Draft “AI as normal tech” checklist — context/https://www.anildash.com//2025/10/17/the-majority-ai-view/ #fleeting #inbox - [F2] Map evaluation KPIs to rollback criteria — context/https://www.anildash.com//2025/10/17/the-majority-ai-view/ #fleeting #inbox - [F3] Survey practitioner sentiment post-layoffs — context/https://www.anildash.com//2025/10/17/the-majority-ai-view/ #fleeting #inbox --- <!-- deep-read:sections end -->