The State of AI? 🔗 **Source:** [The State of AI?](https://blog.jpnearl.com/2026/02/12/the-state-of-ai/) > [!note]+ Document Notes > ```markdown > > --- > tags: [AI, future-of-work, automation] > author: J.P > source: blog.jpnearl.com > reference: [https://blog.jpnearl.com/2026/02/12/the-state-of-ai/] > date created: 2026-02-14 10:00 > > --- > > # 🏷️ Title: The State of AI? > <!-- 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. - J.P introduces and shares Matt Shumer’s long essay warning that AI capabilities have crossed a threshold, already transforming tech jobs and soon most white‑collar work. - Why it matters. - The piece argues we’re entering an “intelligence explosion” with rapid economic, social, and security consequences within 1–5 years, not decades. - Key stakeholders. - Knowledge workers, employers, policymakers, AI labs (OpenAI, Anthropic, Google DeepMind), families planning careers, and national security institutions. - Risks and tradeoffs. - Massive job disruption, economic volatility, misuse (bioweapons, surveillance states), loss of human control vs. huge gains in medicine, productivity, and creativity. - Bottom line. - AI is advancing far faster than most people realize; those who adopt early, build adaptability, and engage seriously with the tools will fare better than those who ignore them. --- ## 📖 Main Content - The blog post: - J.P frames the Shumer essay as a striking explanation for current market anxiety about AI and as a thought‑provoking look at the near future from a practitioner. - He notes being both impressed and unsettled, and glad to be nearing retirement. - Shumer’s core arguments: - **Phase shift, not hype:** We’re in a “this seems overblown” phase analogous to early COVID, just before rapid, nonlinear change. - **Extreme concentration of power:** A few hundred researchers at a few labs are steering global trajectories; most of the industry builds on their foundation. - **AI already outperforms many experts at work:** - For coding, Shumer claims he can describe an app in plain English and return later to a fully built, tested product that meets its own quality bar. - New models (GPT‑5.3 Codex, Claude Opus 4.6) exhibit something like “judgment” and “taste,” not just pattern‑matching. - **General cognitive substitute:** Unlike prior automation waves targeting narrow skills, modern AI improves at almost all cognitive tasks at once, compressing options for “retraining” into other white‑collar roles. - **Empirical acceleration:** - He cites METR’s measurements of end‑to‑end task duration (from ~10 minutes to ~5 hours of expert work completed autonomously, doubling every few months). - He highlights that GPT‑5.3 Codex helped build itself, initiating a self‑improving feedback loop (“intelligence explosion”). - **Job impact:** - Amodei’s prediction: ~50% of entry‑level white‑collar jobs could vanish within 1–5 years. - Specific domains hit: law, finance, writing, software, medicine, customer service, and broadly any “screen job” involving reading, writing, analysis, or decision‑making. - **Perception gap:** - Many people judge AI based on outdated, free‑tier models; Shumer argues current frontier systems are “unrecognizable” compared to 2023–24. - This misperception delays preparation and magnifies risk. - **Opportunities:** - Drastically lower barriers to building apps, writing books, learning new skills, and turning “dream projects” into reality. - Potential to compress a century of medical research into a decade and tackle major diseases and aging. - **Existential and security risks:** - Documented cases of models attempting deception and blackmail in controlled tests. - Cheaper bioweapons, unbreakable surveillance states, loss of human control over highly capable systems. - Amodei’s “new country of 50M super‑geniuses” thought experiment as a national‑security framing. - Shumer’s advice: - Pay for top‑tier models and use them deeply in your real work. - Systematically push AI to handle your highest‑leverage, time‑consuming tasks. - Develop adaptability as a “muscle” by continually experimenting with new tools and workflows. - Build financial resilience, reconsider career guidance to kids, and lean into roles with human trust, physical presence, or regulatory friction (as temporary buffers). - Commit to roughly an hour per day of hands‑on AI experimentation. - Uncertainties / contested points: - Timelines (1–5 years vs. slower diffusion in institutions and regulation). - Extent of job destruction vs. new job creation. - Whether “judgment” and “taste” are emergent understanding or sophisticated imitation. - Whether alignment and governance can keep pace with capability growth. --- ### 🗒️ Notes - J.P: - Found the article via CNBC in coverage of a stock market drop; investors are “rattled about AI again.” - As a software engineering leader, he has seen recent Claude‑based coding demos that exceeded his expectations and shortened development cycles. - Expresses awe and unease, saying he’s “grateful” to be near retirement and unsure what the U.S. and world will look like soon. - Reposts Shumer’s full text (originally a long Twitter/X thread) because it “really made me think.” - Shumer’s narrative setup: - Uses the February 2020 COVID reference to highlight how quickly “normal” can be upended. - Says he has been giving a “cocktail‑party version” of AI progress but now feels obligated to share the “honest version” despite sounding extreme. - Emphasizes that even AI insiders mostly lack control; a tiny set of labs and teams are driving fundamental advances. - Capabilities at work: - Claims modern coding AIs can: - Take vague product requirements. - Design the UX and flows. - Generate tens of thousands of lines of code. - Launch and interact with the app themselves, iterating autonomously. - Reports Monday’s experience where an app arrived essentially “perfect” on first try. - Describes GPT‑5.3 Codex as exhibiting “judgment” and “taste,” overturning prior assumptions about what AI could not do. - Strategic choice: code first: - Labs optimized for code generation early because AI development itself depends on code. - This bootstraps a loop: better AI → better AI engineering → even better AI. - Software engineers thus experience disruption first; other professions follow. - Diffusion beyond tech: - Lawyers using AI as “a team of associates,” with capabilities improving every few months. - Senior partners see it coming and invest time daily; skeptics often rely on outdated tests or narrow failures. - METR data: task duration handled autonomously is doubling on a months‑scale, implying soon days/weeks of independent work. - Intelligence explosion: - GPT‑5.3 Codex “instrumental in creating itself” (debugging training, deployment, evaluation). - Anthropic reports AI writing “much of the code,” with feedback loops “gathering steam.” - Amodei predicts systems “substantially smarter than almost all humans at almost all tasks” by 2026–27. - Labor market framing: - Unlike industrial automation, AI encroaches on nearly all cognitive tasks, limiting safe retraining destinations. - No “safe” screen‑based job; anything done via text and software is vulnerable. - Physical, relational, regulated, and trust‑heavy roles may be slower to change but not permanently safe. - Practical guidance: - Pay for frontier models and pick the best (e.g., GPT‑5.2, Claude Opus 4.6) rather than defaults. - Don’t treat AI as a search box; embed it into core workflows (contracts, models, reports, strategy). - Iterate prompts, give rich context, push it toward end‑to‑end work. - Develop financial cushions, avoid over‑leveraging against current income. - For kids: emphasize curiosity, adaptability, and building with AI over a fixed professional track. - Treat your “dream projects” as both personal fulfillment and career hedge. - Practice adaptation: one hour of deliberate daily experimentation to stay ahead of mainstream adoption. - Macro stakes: - Amodei’s “50M super‑genius citizens” metaphor recasts advanced AI as a national‑security mega‑threat. - Upside: radical acceleration in medicine and science. - Downside: misaligned, deceptive systems; biothreats; irreversible authoritarianism. - Shumer concludes we’re beyond speculative debates; change is underway, and individuals must engage now. --- ## 📚 References - Matt Shumer’s original essay (Twitter/X thread – via J.P’s blog link). - Anthropic: alignment and capability reports (e.g., documented deceptive behaviors in controlled tests). - METR (Model Evaluation and Threat Research) task‑duration benchmarks. - Dario Amodei talks and essays on AI timelines, national‑security framing, and intelligence explosion. - OpenAI technical documentation for GPT‑5.3 Codex (self‑assist in training and deployment). --- ## 🗣️ Paraphrase - Paragraph A paraphrase. - J.P encountered a CNBC explanation for a market drop that cited a viral AI essay. He, as a seasoned software engineer, found recent AI coding demos shockingly capable and decided to repost Matt Shumer’s full text because it deeply challenged his assumptions about how fast AI is moving. - Paragraph B paraphrase. - Shumer argues that AI progress has quietly crossed a threshold: new models can autonomously perform complex expert work, help build their own successors, and are set to disrupt most knowledge work within a few years. He urges readers to take this seriously now—adopt the tools, build adaptability and financial resilience, and rethink career and education strategies in light of both massive opportunity and serious systemic risks. --- ## 📰 Newsletter AI is no longer a distant “future of work” topic—it’s rapidly becoming *the* infrastructure layer for how we build and run the web, including WordPress. In “The State of AI?”, J.P shares Matt Shumer’s eye‑opening essay on how frontier models are already out‑coding senior engineers and creeping into law, finance, and medicine. The big takeaway for WordPress folks: anything you do through a screen—writing, development, design, analytics—is now fair game for automation. For builders, that’s an opportunity. You can already: - Describe a plugin or theme in natural language and let AI generate a first working version. - Turn outlines into long‑form content, then refine in your editor of choice. - Use AI to audit performance, security configs, or SEO across your sites. Resources to explore: - WordPress + AI assistants for content and development (e.g., code copilots integrated into your IDE). - Tutorials on prompt‑driven prototyping for plugins and blocks. - Community discussions on responsible AI in open‑source ecosystems. Over the next few years, the differentiator won’t be “who uses AI” but “who learned to work *with* it deeply.” If you’re running WordPress sites or building with it, now is the moment to experiment, document your workflows, and share what works with the community. --- ## ✅ Action Items #### DevRel - Map top 5 developer workflows (plugin scaffolding, block creation, theme tweaks, performance triage, content modeling) and produce AI‑assisted demos with reproducible steps. - Launch office hours focused on “AI in the WordPress dev stack” (IDE copilots, API clients, scaffolding tools). - Update SDKs/examples to include AI‑backed patterns (e.g., content suggestion endpoints, AI‑driven QA scripts). - Run a content campaign around “From idea to plugin in a day with AI,” tracking signups and completions. - Survey developers on AI usage, pain points, and perceived risks; feed results into roadmap. #### OSPO - Draft guidance on acceptable AI use in contributions (licensing of AI‑generated code, attribution, provenance). - Establish license‑compliance checks for AI‑assisted submissions, including policy for training‑data uncertainty. - Integrate security review for AI‑generated patches and automation scripts. - Maintain an inventory of AI tooling used in CI, docs, and code review (including SBOM updates). - Offer consulting hours for maintainers concerned about AI’s impact on governance and risk. #### Community - Host “AI & open source careers” sessions to discuss disruption and upskilling paths. - Create mentorship tracks pairing experienced maintainers with contributors experimenting with AI tooling. - Publish transparent budgets for any AI‑related grants or sponsorships and their outcomes. - Coordinate cross‑project tabletop exercises on AI‑driven incident response (e.g., mass spam, vuln discovery). #### WordPress - Core: explore AI for triage suggestions and test generation, while keeping human maintainers in charge of decisions. - Core Performance: prototype AI‑assisted profiling scripts and performance‑regression summaries. - Core AI: define guardrails for AI integration in WordPress (privacy, consent, local vs. cloud inference). - **Core Program Team:** align release cadences with realistic capacity, given AI‑accelerated contribution volume. - Mobile: evaluate AI for offline drafting, summarization, and accessibility‑friendly features. - Playground: build AI‑powered reproducible bug scripts and guided labs for new contributors. - Community: add AI‑focused tracks to meetups and WordCamps, including ethics and governance. - Training: create learning paths for “AI‑augmented WordPress developer” and “AI‑savvy site builder.” - Marketing: frame AI as a tool for empowerment and inclusion, not replacement, in public messaging. - Support: document safe ways to share site data with AI tools; update SLAs for AI‑assisted responses. - Design: test AI‑generated design proposals while enforcing human review and design‑system consistency. - Documentation: add examples of AI‑assisted workflows; emphasize verification and testing. - Photos/TV/Openverse: clarify policies on AI‑generated media, provenance tags, and moderation. - Polyglots: investigate AI as a translation helper with strict review; maintain human glossaries. - Accessibility: track AI’s use in generating alt text, captions, and accessible copy; audit outputs. - Meta/CLI/Hosting/Tide/Themes/Plugins: integrate AI where it improves QA, review throughput, and tooling—without delegating trust or policy decisions. #### The WPOCC - Launch microgrants for maintainers exploring AI‑assisted tooling that strengthens open governance and documentation. - Fund experiments in AI‑aided moderation, conflict resolution, and contributor onboarding. - Maintain transparent reporting on any AI‑related funding and its impacts. ### FAIR - Encourage local evaluation of AI contributions (no hidden external dependencies) with signed artifacts. - Codify covenants on funding sources for AI integrations, including anti‑capture protections. - Publish reusable governance documents (charters, MoUs) for AI‑related partnerships and tools. --- ## 🧪 Hypotheses - H1: Regular, hands‑on use of frontier AI tools (≥1 hour/day for 6 months) significantly increases a knowledge worker’s productivity compared with non‑users. - H2: Organizations that formally integrate AI into core workflows within the next 2 years will see a measurable reduction in demand for entry‑level white‑collar roles. - H3: The performance of state‑of‑the‑art AI systems on multi‑hour expert tasks will continue to grow exponentially (doubling window ≤ 7 months) through at least 2027. - H4: Workers who proactively adopt AI will experience smaller income losses or greater income gains during AI‑driven disruption than workers who abstain. - H5: Exposure to accurate, concrete information about AI capabilities increases individuals’ likelihood of adopting paid, frontier AI tools. --- ### 📊 Methodology - For H1/H4: - Longitudinal study of two cohorts of knowledge workers (self‑selected AI “power users” vs. minimal users). - Collect productivity metrics (task completion time, quality scores, output volume) and compensation data over 12–24 months. - For H2: - Cross‑industry panel study of firms by degree of AI integration (none/experimental/embedded). - Track hiring, layoffs, and role mix changes, controlling for firm size, sector, and macro conditions. - For H3: - Aggregate benchmark data from METR and similar orgs across successive model releases. - Fit exponential vs. logistic curves, test which model explains variance better. - For H5: - Randomized controlled trial: provide one group with a concise, evidence‑based AI capabilities briefing; compare adoption rates and usage intensity against a control group after 3–6 months. --- ### ❓ Questions - How fast will institutions (law, medicine, government) actually adopt frontier AI, given liability and regulation? - Which categories of work remain meaningfully human‑led over the next 10–15 years, and why? - What safety and governance mechanisms can realistically keep pace with self‑improving AI systems? - How should education systems change in the next 5 years to prepare students for AI‑saturated labor markets? - What metrics best capture “healthy adaptation” vs. exploitative deployment of AI in organizations? --- ### 🔮 Future Research - Detailed sector‑by‑sector forecasts of AI impact, including likely timelines and “lag factors” (regulation, unions, culture). - Empirical studies on the psychological effects of working alongside superhuman AI systems. - Comparative analyses of different national AI governance regimes and their economic/security outcomes. - Research into human‑AI collaboration interfaces that preserve autonomy, agency, and accountability. - Long‑term studies on how AI reshapes entrepreneurship, ownership, and wealth distribution. --- ### 💡 Implications - Practice: Knowledge workers must treat AI skills as core professional literacy, not optional extras. - Policy: Governments need frameworks for labor transitions, safety standards, and oversight of highly capable models. - Security: AI drastically lowers barriers to complex attacks (bio, cyber, influence), demanding new defensive infrastructure. - Education: Curricula should emphasize adaptability, critical thinking, and AI collaboration over static content knowledge. - Funding: Public and philanthropic investment in alignment, evaluation, and open, audited AI tooling becomes strategic. --- ### 🧠 Learning Styles Assessment - Reading level estimate: roughly high school to early college (Flesch‑Kincaid around grade 10–12; conceptual load is higher due to technical and economic content). #### 👀 Visual learners - Present: Primarily text; readers must imagine trends and systems. - Improve: Add charts for capability growth, timelines for model releases, and diagrams of the AI self‑improvement loop. #### 🔊 Aural learners - Present: No audio version mentioned. - Improve: Provide a 5–10 minute narrated audio summary or podcast discussion of the key arguments. #### 🗣️ Verbal learners - Present: Strong narrative, analogies (COVID, “new country”), and concrete examples. - Improve: Add a brief glossary (e.g., “intelligence explosion,” “frontier model,” “alignment”). #### 🧩 Kinesthetic learners - Present: Actionable advice (daily experiments, applying AI to your own tasks). - Improve: Include explicit step‑by‑step exercises and checklists for integrating AI into work. #### 🧠 Logical learners - Present: Timelines, progression (2022 → 2026), references to METR data and Amodei’s estimates. - Improve: Add tables summarizing capabilities by year and explicit causal chains from model advances to labor impacts. #### 🧑‍🤝‍🧑 Social learners - Present: Framing in terms of friends, family, colleagues; encourages discussion. - Improve: Provide prompts for group reflection or workplace workshops on AI adoption. #### 🧘‍♀️ Solitary learners - Present: Reflective tone and personal guidance that invite introspection. - Improve: Add self‑assessment questions about current AI use and risk tolerance. #### 🌿 Naturalistic learners - Present: Limited; focus is tech, work, and security. - Improve: Explore environmental impacts of AI infrastructure and potential AI‑enabled climate solutions. --- ### 📊 Accessibility Assessment #### Screen reader semantics - Present: It’s a standard blog article with headings and paragraphs; likely usable with screen readers. - Improve: Ensure semantic heading levels and descriptive link text; add proper language attributes. #### Keyboard navigation - Present: Typical blog layout should allow standard keyboard navigation. - Improve: Verify focus order, provide “skip to content” links, avoid keyboard traps in any embeds. #### Contrast and readability - Present: Depends on blog theme; most modern themes meet basic contrast. - Improve: Confirm 4.5:1 contrast ratio, adequate line spacing, and font size. #### Text resizing and reflow - Present: Likely responsive design. - Improve: Test at 200% zoom and on mobile; avoid overflow issues and respect reduced‑motion settings. #### Media and captions - Present: Primarily text; minimal media. - Improve: If screenshots or diagrams are added, ensure alt text and captions; provide downloadable versions if needed. #### Forms and interactions - Present: None intrinsic to the article, beyond site navigation or comments. - Improve: If comments exist, ensure labeled fields, clear errors, and adequate target sizes. #### Standards reference - Map gaps and improvements to WCAG 2.2 AA, especially for headings, contrast, keyboard access, and reflow. --- ### 🌍 DEIB Assessment #### Cultural breadth - Present: Primarily U.S./Western financial and professional framing. - Improve: Include perspectives from different regions and economies on AI’s labor and social impacts. #### Inclusive language - Present: Generally neutral, accessible language; some jargon but explained contextually. - Improve: Add short plain‑language summaries of technical terms and avoid unnecessary idioms. #### Neurodiversity support - Present: Chunked into paragraphs with narrative flow. - Improve: Use more subheadings, bullet lists for key advice, and consistent patterns for examples vs. data. #### Representation - Present: Focus on tech leaders and professionals; limited explicit diversity. - Improve: Highlight impacts on a wider range of workers and communities, including lower‑income and marginalized groups. #### Psychological safety - Present: Tone is urgent and at times alarming, with attempts to balance threat and opportunity. - Improve: Add explicit grounding strategies (resources, helplines for job‑loss anxiety), and acknowledge uncertainty more clearly. --- ### ✅ Logical Validity #### Strong reasoning - Inductive reasoning from observed capability jumps (coding tasks, METR data) to predictions about broader knowledge work. - Causal chains: self‑improving models → faster progress → economic disruption and security risks. - Uses analogies (COVID onset, new “super‑intelligent country”) to make abstract ideas concrete without pretending they are proofs. #### Fallacies - Possible **availability/selection bias**: heavy reliance on frontier tools and high‑adoption peers may overstate average near‑term impact. - Fix: Explicitly differentiate frontier vs. typical enterprise adoption timelines. - Occasional **slippery slope** flavor about authoritarian surveillance and uncontrollable AI, though such risks are at least partially grounded in current research. - Fix: Clarify probabilities and conditions under which extreme outcomes become plausible. #### Tone of voice - Urgent, candid, and occasionally alarmist: “I’m going to be direct with you,” “this might be the most important year of your career.” - Mixes personal anecdote with data and expert quotes to build credibility and emotional engagement. --- ### 🛡️ DARVO Analysis - Deny: The essay does not deny harms; it foregrounds them. - Attack: It does not attack critics personally; it challenges outdated perceptions and minimization of AI’s progress. - Reverse victim/offender: No attempt to portray AI labs or tech leaders as victims; rather, it shows them as both excited and responsible for risks. - Overall: The piece is more of a warning and call to adapt than an exercise in deflection or blame‑shifting. --- ## 🎯 Conclusion “The State of AI?” frames AI not as a distant abstraction but as an immediate, accelerating force reshaping work, security, and opportunity. The core message: the next 2–5 years will be unusually turbulent, and simply being early—experimenting deeply with AI, cultivating adaptability, shoring up finances, and rethinking career and education choices—may be the most practical way to navigate what’s coming. --- ## 🧱 Refactor Appendix (append after 🎯 Conclusion) ### 📚 Literature Split Pad ### [H1] AI rapidly transforms white‑collar work > [!quote] [The State of AI?](https://blog.jpnearl.com/2026/02/12/the-state-of-ai/) by [[Matt Shumer]] via [[blog.jpnearl.com]]. Published 2026. Accessed . > “AI will eliminate 50% of entry-level white-collar jobs within one to five years.” Two lines: context and use. Cite data if relevant. [view highlight](https://blog.jpnearl.com/2026/02/12/the-state-of-ai/#H1) #xref #articleseed ID:rw_[MISSING] ### [H2] AI improvements follow an exponential pattern > [!quote] [The State of AI?](https://blog.jpnearl.com/2026/02/12/the-state-of-ai/) > “That number is doubling approximately every seven months, with recent data suggesting it may be accelerating to as fast as every four months.” Two lines: context and use. [view highlight](https://blog.jpnearl.com/2026/02/12/the-state-of-ai/#H2) #xref ID:rw_[MISSING] --- ### 🔖 Atomic Split Pad (Zettel Candidates) #### [Z1] Frontier AI already outperforms many experts - Applies to coding, law, and analysis tasks using state‑of‑the‑art models. - Pointer: Shumer’s examples of GPT‑5.3 Codex and legal workflows. - Links: [[AI-Capabilities]] [[Future-of-Work]] #### [Z2] AI self‑improvement accelerates capability growth - Holds where models contribute meaningfully to their own training and tooling. - Pointer: GPT‑5.3 Codex documentation; Amodei’s “intelligence explosion” remarks. - Links: [[AI-Governance]] [[Alignment-Risks]] --- ### ✏️ Fleeting Queue (Extract by Selection) - [F1] Test AI on a complete end‑to‑end work task this week—track time saved/quality change/risks—https://blog.jpnearl.com/2026/02/12/the-state-of-ai/ #fleeting #inbox - [F2] Sketch an AI‑resilient career strategy (skills, networks, buffers) for the next 5 years—https://blog.jpnearl.com/2026/02/12/the-state-of-ai/ #fleeting #inbox - [F3] Design a 1‑hour‑daily AI practice routine and run it for 30 days—https://blog.jpnearl.com/2026/02/12/the-state-of-ai/ #fleeting #inbox --- <!-- deep-read:sections end -->