<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Ron McFarland PhD on Medium]]></title>
        <description><![CDATA[Stories by Ron McFarland PhD on Medium]]></description>
        <link>https://medium.com/@highervista?source=rss-1f7645bf9aea------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*Qz5pxxJPs8QiETh9li-ZYA.png</url>
            <title>Stories by Ron McFarland PhD on Medium</title>
            <link>https://medium.com/@highervista?source=rss-1f7645bf9aea------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Thu, 19 Nov 2026 21:55:51 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@highervista/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[The First Job Trap: What CS, IT, and AI Graduates Need to Know About Underemployment]]></title>
            <link>https://highervista.medium.com/the-first-job-trap-what-cs-it-and-ai-graduates-need-to-know-about-underemployment-6fad5177af2d?source=rss-1f7645bf9aea------2</link>
            <guid isPermaLink="false">https://medium.com/p/6fad5177af2d</guid>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[large-language-models]]></category>
            <category><![CDATA[computer-science]]></category>
            <category><![CDATA[information-security]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <dc:creator><![CDATA[Ron McFarland PhD]]></dc:creator>
            <pubDate>Thu, 19 Nov 2026 16:29:34 GMT</pubDate>
            <atom:updated>2026-11-19T16:31:45.465Z</atom:updated>
            <content:encoded><![CDATA[<p>by Ron McFarland, Ph.D.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*at-gRx-275SDsm7g1q-7hA.png" /><figcaption>The first job trap</figcaption></figure><h3>Introduction</h3><p>For my students in CS, IT, and AI, and for all other AI learners.</p><p>Picture a new graduate in a borrowed blazer, diploma still rolled in its cardboard tube, résumé listing Python, Java, and a capstone project built on too little sleep. Six months later, that graduate is stocking shelves, driving deliveries, or resetting passwords at a help desk that never asked for a degree. The scene is common. In the third quarter of 2026, 42.5% of employed recent college graduates aged 22 to 27 worked in jobs that typically do not require a bachelor’s degree (Federal Reserve Bank of New York [FRBNY], 2026).</p><p>The second number is the one that should keep students up at night. Among graduates whose first job did not require a degree, 73% were still underemployed 10 years later (Burning Glass Institute &amp; Strada Institute for the Future of Work, 2024). The first job is not a waiting room. It is a set of rails, and the train tends to keep running in the direction it left the station.</p><p>This article is written for students in computer science (CS), information technology (IT), and artificial intelligence (AI). It lays out what the research shows, where tech graduates actually stand, how a career path built around AI works, and which courses and certifications help a degree become a career-level first job.</p><h3>The 42% problem: degrees working below their level</h3><p>Roughly two in five recent graduates are working below their degree, and that share hovered near 42% through all of 2026. The New York Fed tracks bachelor’s degree holders aged 22 to 27 and counts them as underemployed when they work in an occupation where most workers do not need a four-year degree (FRBNY, 2026). The rate was 41.5% in the first quarter of 2026, 42% in the second, and 42.5% in the third (FRBNY, 2026).</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*z1NfcN3F-b2QPffnBfe3OA.png" /><figcaption>Figure 1. Data from FRBNY (2026). By-major rates are 2024 annual data; the average line is the Q3 2026 national rate.</figcaption></figure><p>Major matters more than almost anything else. Tech majors sit at the low end of the range: in the Fed’s 2024 data by major, about 19% of computer science graduates were underemployed, while criminal justice approached 66% (Cerullo, 2026; FRBNY, 2026). The tech graduate’s problem is getting through the door at all, which the section on CS and IT graduates below covers.</p><p>Two research teams measure this differently. The Fed’s 42% is a snapshot of every employed graduate aged 22 to 27. The Burning Glass Institute and Strada follow individual graduates over time and find 52% underemployed one year after graduation (Burning Glass Institute &amp; Strada Institute for the Future of Work, 2024). The methods differ; the warning does not.</p><h3>The 73% trap: why the first job sets the track</h3><p>The first job after graduation is the strongest signal of where a graduate will stand a decade later. Researchers at the Burning Glass Institute and Strada traced the careers of tens of millions of bachelor’s degree holders through online résumés and census microdata (Burning Glass Institute &amp; Strada Institute for the Future of Work, 2024). Of graduates whose first job did not require a degree, 73% were still underemployed 10 years later. Even a full decade out, 45% of all graduates remained underemployed (Burning Glass Institute &amp; Strada Institute for the Future of Work, 2024).</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Mk0EvVS3P8YUQR_V8q3H4A.png" /><figcaption>Figure 2. Data from Burning Glass Institute &amp; Strada Institute for the Future of Work (2024) and Hennerich (2025).</figcaption></figure><p>The rails run the other way, too. Seventy-nine percent of graduates who began in a college-level job were still in one 10 years after graduation (Hennerich, 2025). Economists call the stubborn half of this pattern a scarring effect: an early setback leaves a mark that outlasts the bad market that caused it.</p><p>The mark shows up in the paycheck. Recent graduates in college-level jobs earned about $60,000 a year on average, roughly 50% more than underemployed recent graduates (Burning Glass Institute &amp; Strada Institute for the Future of Work, 2024). The same study found one of the clearest escape hatches: graduates with at least one internship had 48.5% lower odds of being underemployed. For a CS or IT student, that internship is the difference between a first résumé line that reads “junior developer” and one that reads “retail associate.”</p><h3>Where CS and IT graduates stand: a good room behind a narrow door</h3><p>Tech graduates face a different problem from the average graduate. The jobs they land usually use the degree, but fewer of them are landing jobs at all. In the New York Fed’s 2024 data by major, recent computer engineering graduates had a 7.8% unemployment rate and computer science graduates 7.0%, against a national rate of about 4.3% (Cerullo, 2026; FRBNY, 2026). Those who do get hired are paid well: median early-career wages were $87,000 for computer science and $90,000 for computer engineering (FRBNY, 2026).</p><p>AI is part of the reason the door has narrowed. Stanford economists studying ADP payroll records found that from late 2022 to mid-2026, employment for workers aged 22 to 25 in the most AI-exposed occupations, including software development and customer service, fell about 11%. Experienced workers in those same jobs grew (Brynjolfsson et al., 2026).</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/992/1*iKoeW5qloPDCp000QnBV4Q.png" /><figcaption>Figure 3. Data from Brynjolfsson et al. (2026), ADP payroll records, June 2026 update.</figcaption></figure><p>The decline came mainly from fewer hires, not layoffs (Brynjolfsson et al., 2026). Companies are not firing their junior developers; they are hiring fewer new ones. The routine coding, testing, and documentation that once trained beginners is exactly the kind of work AI tools now assist with, though the authors stress that their findings describe a pattern rather than prove a cause.</p><p>The same study holds the most useful clue for students. In occupations where AI is mostly used to automate tasks, young workers lost ground. Where AI is mostly used to augment workers, helping them do more, no similar decline appeared (Brynjolfsson et al., 2026). The lesson is not to flee AI-exposed fields. It is to become the person who directs the tool rather than the person the tool replaces.</p><h3>What employers want now: AI skills and senior-level judgment</h3><p>Employers are paying a steep premium for AI skills, and they are asking even entry-level hires to bring judgment once expected only of seniors. PwC’s 2026 Global AI Jobs Barometer, built from more than one billion job ads in 27 countries and territories, found that workers with AI skills earned an average wage premium of 62%, up from 57% the year before (PwC, 2026). Jobs requiring specific AI skills grew 69%, against 9% for the job market as a whole.</p><p>U.S. job-posting data tell the same story at a smaller scale. Lightcast’s analysis of more than 1.3 billion postings found that roles listing AI skills advertised salaries 28% higher, roughly $18,000 a year more, than roles without them (Lightcast, 2025).</p><p>The more important finding for new graduates is who gets hired. In 2.4 million U.S. entry-level postings, the most AI-exposed entry-level roles were seven times more likely to require senior-level, human-intensive skills such as judgment, leadership, and adaptability (PwC, 2026). AI writes the first draft; employers want a junior hire who can tell whether it&#39;s right.</p><p>The federal outlook confirms where growth is heading. The Bureau of Labor Statistics projects the fastest growth among tech occupations for data scientists and research scientists, the roles closest to building and directing AI systems.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*fI2qKi2ciFUNNA5KbD6yWQ.png" /><figcaption>Figure 4. Data from U.S. Bureau of Labor Statistics (2026a, 2026b, 2026c, 2026d, 2026e), Occupational Outlook Handbook, updated August 27, 2026.</figcaption></figure><p>The bottom bar matters as much as the top one. Network and computer systems administrator jobs are projected to shrink 4% over the decade (U.S. Bureau of Labor Statistics [BLS], 2026e). Roles built on routine configuration are thinning; roles built on data, security, and AI are expanding.</p><h3>The AI career path, step by step</h3><p>An AI career is not a single job title. It is a ladder with four rungs, and the hardest step is the first: landing a first job that actually requires the degree. Everything above that rung gets easier, because each role builds evidence for the next.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*tCX3Vdhaw1eSbfJaTqxGAg.png" /><figcaption>Figure 5. A typical AI career path in four stages and three gates. Author’s synthesis of the sources in this article.</figcaption></figure><p><strong>Student.</strong> College is where the foundation is poured: programming, data structures, algorithms, linear algebra, probability, and statistics. It is also where the résumé gets its first real lines. Internships carry the most weight of anything a student can control, lowering the odds of underemployment by nearly half (Burning Glass Institute &amp; Strada Institute for the Future of Work, 2024).</p><p><strong>Entry level.</strong> Most AI careers do not start with the title “AI engineer.” They start as junior software engineers, data analysts, cloud associates, or security analysts who use AI tools daily and learn how real systems fail. Software developers remain a large and well-paid occupation, with a May 2025 median of $135,980 and about 106,100 openings a year across developers, testers, and quality assurance analysts (BLS, 2026a).</p><p><strong>Mid-level.</strong> Here, the title often changes to machine learning engineer, MLOps engineer, data engineer, or AI application developer. The work shifts from writing features to running models in production: deploying them, monitoring for drift, and keeping cost and security under control.</p><p><strong>Senior.</strong> Senior engineers and architects design whole systems and decide where AI should and should not be trusted. Research-heavy roles such as computer and information research scientist typically require a master’s degree and pay a May 2025 median of $140,300 (BLS, 2026b).</p><h3>Courses that pay off</h3><p>The courses that matter most are the ones AI cannot shortcut: the math underneath the models, the engineering that keeps them running, and the judgment that decides when to trust them. Employers screening entry-level candidates in AI-exposed roles now look for senior-level skills such as judgment and adaptability (PwC, 2026), and the courses below build both the technical floor and that judgment.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*8TiOw_sGsfFufVHtvcosYA.png" /><figcaption>Table 1: Courses that pay off</figcaption></figure><p>A capstone or senior project should tie several of these together into something deployed: a small retrieval-augmented chatbot over a real document set, a model served behind an API with monitoring, or a security tool that flags suspicious log entries. A working link in a résumé says more than a course title.</p><h3>Certifications, and why the degree still matters</h3><p>A degree and a certification answer two different questions an employer asks. The degree answers “Can this person learn hard things and think in systems?” The certification answers, “Can this person use the platform we run on today?&quot; The strongest first-job candidates bring both.</p><p>The degree is still the ticket to the room. The Bureau of Labor Statistics lists a bachelor’s degree as the typical entry-level education for software developers, data scientists, and information security analysts (BLS, 2026a, 2026c, 2026d). It is also the dividing line on which the underemployment data are built: a graduate who lands a job that requires the degree is far more likely to stay on the college-level track (Burning Glass Institute &amp; Strada Institute for the Future of Work, 2024).</p><p>Certifications add current, verifiable proof. BLS notes that many employers prefer information security candidates who hold a security certification (BLS, 2026d). In a 2025 Pearson VUE survey, 32% of certification holders reported a raise after certifying, and CompTIA found that nine in ten companies had invested moderately or heavily in skills-based hiring and development (White, 2026). Certification exams are also becoming more hands-on, with simulated systems that test what a candidate can do rather than what they can recall (White, 2026).</p><p>Certifications in the AI field change quickly. Microsoft retired its Azure AI Engineer Associate (AI-102) and Azure AI Fundamentals (AI-900) credentials on June 30, 2026 (Microsoft, 2026), and AWS is updating its Machine Learning Engineer exam. Check the vendor’s page before you study.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*QOIwGeQ2JoaoybQToGqQsA.png" /><figcaption>Table 2: Certifications that pay off</figcaption></figure><p>A sound sequence is one starter certification during college, one associate certification in the platform your target employer uses before or soon after graduation, and a professional certification only after real production experience. Collecting a stack of beginner badges signals breadth; one associate certification paired with a deployed project signals readiness.</p><h3>An action plan for avoiding the trap</h3><p>The research points to a short list of moves that separate graduates who start on the college-level track from those who do not.</p><p>1. <strong>Land at least one internship, ideally two.</strong> It is the single best-documented protection against underemployment (Burning Glass Institute &amp; Strada Institute for the Future of Work, 2024). Start applying in sophomore year; campus IT, research labs, and local businesses count.</p><p>2. <strong>Build in public.</strong> Keep two or three AI projects on GitHub with a clear README, a live demo, and a short note on what the model gets wrong.</p><p>3. <strong>Use AI tools, and show that you can check them.</strong> Employers want juniors with judgment (PwC, 2026). In interviews, describe a time you caught an AI tool’s mistake.</p><p>4. <strong>Earn one starter and one associate certification</strong> in the cloud platform your target employers use, timed to finish before graduation.</p><p>5. <strong>Aim the first job carefully.</strong> A degree-level role in data, security, cloud, or software that pays less is usually a better long-term bet than a higher-paying job that doesn&#39;t use the degree, because the first job tends to set the track (Burning Glass Institute &amp; Strada Institute for the Future of Work, 2024).</p><p>6. <strong>If you do start underemployed, plan the exit early.</strong> About 27% of graduates escape underemployment within a decade (Burning Glass Institute &amp; Strada Institute for the Future of Work, 2024). Those odds improve with a certification, a portfolio, and a job search that never fully stops.</p><h3>Conclusion</h3><p>The numbers are sobering but not fatal. Roughly 42% of recent graduates are working below their degree, and nearly three in four who start there are still there a decade later. CS and IT graduates are better placed than most: when they get hired, the job almost always uses what they learned, and it pays well. Their challenge is the narrower door into that first job, a door that AI has made smaller at the bottom and wider at the top.</p><p>The students who get through will look a little different from the class of five years ago. They will carry a degree that proves they can think, a certification that proves they can build on today’s platforms, an internship that proves someone already trusted them, and a portfolio that proves they can make AI do useful work and catch it when it is wrong. That combination does not guarantee the first job. It does make it far more likely that the first job is a real one, and in this market that decision shapes the next ten years.</p><h3>References</h3><p>Amazon Web Services. (n.d.-a). <em>AWS Certified AI Practitioner</em>. Retrieved November 18, 2026, from <a href="https://aws.amazon.com/certification/certified-ai-practitioner/">https://aws.amazon.com/certification/certified-ai-practitioner/</a></p><p>Amazon Web Services. (n.d.-b). <em>AWS Certified Machine Learning Engineer — Associate</em>. Retrieved November 18, 2026, from <a href="https://aws.amazon.com/certification/certified-machine-learning-engineer-associate/">https://aws.amazon.com/certification/certified-machine-learning-engineer-associate/</a></p><p>Brynjolfsson, E., Chandar, B., &amp; Chen, R. (2026). <em>Canaries in the coal mine? Six facts about the recent employment effects of artificial intelligence</em> [Working paper]. Stanford Digital Economy Lab. <a href="https://digitaleconomy.stanford.edu/wp-content/uploads/2025/08/Canaries_BrynjolfssonChandarChen.pdf">https://digitaleconomy.stanford.edu/wp-content/uploads/2025/08/Canaries_BrynjolfssonChandarChen.pdf</a></p><p>Burning Glass Institute &amp; Strada Institute for the Future of Work. (2024). <em>Talent disrupted: Underemployment, college graduates, and the way forward</em>. <a href="https://www.burningglassinstitute.org/s/Talent-Disrupted-03052024.pdf">https://www.burningglassinstitute.org/s/Talent-Disrupted-03052024.pdf</a></p><p>Cerullo, M. (2026, February 24). These college majors have the highest earning potential and lowest unemployment rates, study finds. <em>CBS News</em>. <a href="https://www.cbsnews.com/news/college-majors-unemployment-rate-wage">https://www.cbsnews.com/news/college-majors-unemployment-rate-wage</a></p><p>Federal Reserve Bank of New York. (2026). <em>The labor market for recent college graduates</em>. Retrieved November 18, 2026, from <a href="https://www.newyorkfed.org/research/college-labor-market">https://www.newyorkfed.org/research/college-labor-market</a></p><p>Hennerich, H. (2025, August 13). <em>The jobs and degrees underemployed college graduates have</em>. Federal Reserve Bank of St. Louis. <a href="https://www.stlouisfed.org/open-vault/2025/aug/jobs-degrees-underemployed-college-graduates-have">https://www.stlouisfed.org/open-vault/2025/aug/jobs-degrees-underemployed-college-graduates-have</a></p><p>Lightcast. (2025). <em>Beyond the buzz: Developing the AI skills employers actually need</em>. <a href="https://lightcast.io/resources/research/beyond-the-buzz-developing-the-ai-skills-employers-actually-need">https://lightcast.io/resources/research/beyond-the-buzz-developing-the-ai-skills-employers-actually-need</a></p><p>Microsoft. (2026). <em>Retired certification exams</em>. Microsoft Learn. Retrieved November 18, 2026, from <a href="https://learn.microsoft.com/certifications/retired-certification-exams">https://learn.microsoft.com/certifications/retired-certification-exams</a></p><p>PwC. (2026, June 15). <em>AI reshapes global labour market into two distinct paths, rewarding human skills: PwC 2026 Global AI Jobs Barometer</em> [Press release]. <a href="https://www.pwc.com/gx/en/news-room/press-releases/2026/pwc-2026-ai-jobs-barometer.html">https://www.pwc.com/gx/en/news-room/press-releases/2026/pwc-2026-ai-jobs-barometer.html</a></p><p>U.S. Bureau of Labor Statistics. (2026a). Software developers, quality assurance analysts, and testers. In <em>Occupational outlook handbook</em>. U.S. Department of Labor. <a href="https://www.bls.gov/ooh/computer-and-information-technology/software-developers.htm">https://www.bls.gov/ooh/computer-and-information-technology/software-developers.htm</a></p><p>U.S. Bureau of Labor Statistics. (2026b). Computer and information research scientists. In <em>Occupational outlook handbook</em>. U.S. Department of Labor. <a href="https://www.bls.gov/ooh/computer-and-information-technology/computer-and-information-research-scientists.htm">https://www.bls.gov/ooh/computer-and-information-technology/computer-and-information-research-scientists.htm</a></p><p>U.S. Bureau of Labor Statistics. (2026c). Data scientists. In <em>Occupational outlook handbook</em>. U.S. Department of Labor. <a href="https://www.bls.gov/ooh/math/data-scientists.htm">https://www.bls.gov/ooh/math/data-scientists.htm</a></p><p>U.S. Bureau of Labor Statistics. (2026d). Information security analysts. In <em>Occupational outlook handbook</em>. U.S. Department of Labor. <a href="https://www.bls.gov/ooh/computer-and-information-technology/information-security-analysts.htm">https://www.bls.gov/ooh/computer-and-information-technology/information-security-analysts.htm</a></p><p>U.S. Bureau of Labor Statistics. (2026e). Network and computer systems administrators. In <em>Occupational outlook handbook</em>. U.S. Department of Labor. <a href="https://www.bls.gov/ooh/computer-and-information-technology/network-and-computer-systems-administrators.htm">https://www.bls.gov/ooh/computer-and-information-technology/network-and-computer-systems-administrators.htm</a></p><p>White, S. K. (2026, March 2). IT certifications take a turn for the practical. <em>CIO</em>. <a href="https://www.cio.com/article/4135858/it-certifications-take-a-turn-for-the-practical.html">https://www.cio.com/article/4135858/it-certifications-take-a-turn-for-the-practical.html</a></p><p><strong>About the Author</strong></p><p>Ron McFarland, Ph.D. With a career spanning over 30 years in computer science, cybersecurity, and artificial intelligence, Dr. Ron McFarland brings unparalleled expertise to his series on stock market trading and AI. Having earned a Ph.D. in Computer Science and a postdoctoral fellowship in cybersecurity, he combines academic knowledge with decades of hands-on experience in artificial intelligence, machine learning, cloud security, database systems, and risk frameworks. Dr. McFarland has taught advanced AI and machine learning courses at the university level, equipping students and professionals to harness cutting-edge technology — a foundation that translates seamlessly to AI’s transformative role in stock trading.</p><p>Beyond his academic and consulting achievements, Dr. McFarland is a prolific writer, having authored several books and dozens of technical articles on AI, blockchain, and cybersecurity. His ability to simplify complex concepts makes his work accessible to readers of all levels, from beginners to advanced traders. As a former Dean and professor who has led programs in computer science, technical analysis, and programming.</p><p>YouTube Channel: <a href="https://www.youtube.com/@HighervistaLLC">https://www.youtube.com/@HighervistaLLC</a></p><p>Amazon Author’s Page: <a href="https://amazon.com/author/ronmcfarland">amazon.com/author/ronmcfarland</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=6fad5177af2d" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[18 Power Words for Learning Anything with AI]]></title>
            <link>https://highervista.medium.com/18-power-words-for-learning-anything-with-ai-1ea17d3f23f0?source=rss-1f7645bf9aea------2</link>
            <guid isPermaLink="false">https://medium.com/p/1ea17d3f23f0</guid>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[large-language-models]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[information-technology]]></category>
            <category><![CDATA[computer-science]]></category>
            <dc:creator><![CDATA[Ron McFarland PhD]]></dc:creator>
            <pubDate>Thu, 19 Nov 2026 16:14:56 GMT</pubDate>
            <atom:updated>2026-11-19T16:14:56.464Z</atom:updated>
            <content:encoded><![CDATA[<p><em>A visual prompting guide for computer science, IT, and AI students</em></p><p>By Ron McFarland, PhD</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*MG8Tpgunubv1Pxl-8LKDkA.png" /></figure><p>For my students in CS/IT and AI, and others interested in AI.</p><p>It is close to one in the morning in a residence hall study lounge. The vending machine hums in the corner, the fluorescent tube above the table flickers every few seconds, and a second-year networking student has read the same paragraph about TCP encapsulation four times. The words are all familiar: frame, segment, header, and payload. Yet they refuse to assemble into anything she can see. She opens a chat window, types “cross section of a network packet,” and a moment later the abstraction splits open like a geode. An Ethernet frame forms the outer crust, an IP packet nests inside it, a TCP segment inside that, and at the center sits the small, bright kernel of the web request she actually cares about. The paragraph she has been fighting suddenly has a shape.</p><p>That moment is the idea behind a list that has been circulating among teachers and AI enthusiasts: eighteen “power words” you can type, along with a topic, into ChatGPT, Claude, or Grok to make the tool explain something visually. The original list used examples from grade-school science—volcanoes, ant faces, butterflies, and clouds. This article takes those eighteen words and points all eighteen at the subjects you are studying now: networking, operating systems, algorithms, databases, security, cloud infrastructure, and machine learning.</p><p>For each word, you will find what it asks the AI to do, why that kind of picture helps understanding or memory, a worked IT or CS example, a prompt you can copy, and the one thing you should verify before you trust the result.</p><h3>Why a Picture Helps You Learn Code and Systems</h3><p>Computing is unusually invisible. You cannot hold a packet, look inside a running process, or watch a hash table resize. Most of what you learn arrives as text: documentation, man pages, lecture slides, and forum answers. Learning science suggests this is a handicap you can fix. Dual coding theory proposes that the mind keeps two linked systems, one for language and one for imagery, and that material encoded in both is easier to understand and recall (Clark &amp; Paivio, 1991). Mayer’s (2021) research on multimedia learning builds on that idea, showing across many experiments that people learn more deeply from words and pictures together than from words alone.</p><p>Diagrams also do computational work for you. Larkin and Simon (1987) explained that a diagram organizes information by location: everything you need to reason about one component sits next to it on the page, so you search less and calculate less in your head. Hegarty (2011) adds that well-designed visual displays let you offload mental work onto the display itself, which matters when you are juggling seven protocol layers or the state of four threads. A well-chosen picture is not decoration. It is a data structure for your brain.</p><p>Each of the eighteen power words below is a shortcut to a different kind of picture: layers, a miniature world, a route through rooms, a zoom sequence, an unfolding process, a comparison, an inventory, a deck of self-test cards, a web of ideas, a labeled diagram, an assembly pulled apart, a route map, a conversation, a production line, an identification guide, a transformation, a hidden skeleton, or a branching story.</p><h3>Know What Your AI Tool Can Actually Draw</h3><p>ChatGPT and Grok generate images directly. OpenAI released ChatGPT Images 2.0 in April 2026, built on its gpt-image-2 model, and says it handles small text, interface elements, and dense diagrams far better than earlier image generators (OpenAI, 2026). xAI followed in August 2026 with Grok Imagine Image 2.0, which it describes as built for precise image generation and editing with crisp text rendering (xAI, 2026). These tools shine with the illustrative words on this list—cross section, isometric diorama, pop-up book, museum exhibit, knolling, exploded view, comic strip, and X-ray—because those words describe a picture style.</p><p>Claude takes a different route. Since March 2026, it has been able to build interactive charts, diagrams, and visualizations inline in a conversation and adjust them as you keep talking (Anthropic, 2026). Those visuals are constructed rather than painted, which makes them strong for structured outputs—mind maps, schematics, flowcharts, and comparison tables—where every label must be exact. Transit maps, assembly lines, and choose-your-own-adventure trees belong in this group too.</p><p>Any text-based assistant can also hand you a diagram as code. Mermaid, for example, is an open-source JavaScript tool that renders Markdown-style text definitions as flowcharts, sequence diagrams, mind maps, and more (Mermaid, n.d.). Because the diagram is plain text, you can paste it into a GitHub README, keep it under version control, and fix a wrong label with one keystroke. For CS/IT/AI students, that is often more useful than a polished image. When a diagram has to be correct, add “as a Mermaid diagram” or “as an SVG” to your prompt.</p><h3>The Formula: Power Word + Topic + Constraints</h3><p>Researchers at Vanderbilt University describe prompts as a form of programming that shapes both an LLM’s output and the interaction, and they catalog reusable “prompt patterns” for that reason (White et al., 2023). Treat the power word as the pattern and fill in the arguments:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Ys1q7ucnsIGA1aEO86mPjw.png" /><figcaption>Power words Template 1</figcaption></figure><p>The last sentence matters most. Asking the model to flag its own uncertainty gives you a starting list for verification, which the final section of this article covers.</p><h3>The Eighteen Power words</h3><h3>1. Cross Section: See What Is Inside</h3><p>A cross section slices a thing open so you can see its layers in place. In the original list, it reveals the molten rock inside a volcano. In computing, the natural candidates are things built in layers that hide one another: a network packet, a container image, a storage device, or the layers of a neural network.</p><p>Take encapsulation. When your browser sends an HTTPS request, the request is wrapped in TLS, then in a TCP segment, then in an IP packet, then in an Ethernet frame, each layer adding its own header like rings in a tree. A textbook presents that sequence one sentence at a time. A cutaway shows the nesting all at once, which is exactly the advantage Larkin and Simon (1987) attributed to diagrams: related information sits side by side instead of being spread across a paragraph.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*i4moZO1f6TsisefX4CI-bQ.png" /><figcaption>Try this prompt: Prompt 1</figcaption></figure><p>Other uses: a cross-section of a Docker image’s read-only layers beneath the writable container layer; a solid-state drive from the controller down to the NAND flash cells; a convolutional neural network showing its feature maps at each depth.</p><p><strong>What to check: </strong>header sizes and field names. An IPv4 header without options is 20 bytes, and so is a TCP header without options. Image models can produce confident labels with wrong numbers, so compare the picture against a real packet capture in Wireshark.</p><h3>2. Isometric Diorama: A Small 3D World</h3><p>An isometric diorama is a tabletop model seen from above at an angle, so every building and road is visible at once. The original example shows how a town cleans its water. In IT, the best subjects are systems with many parts working in parallel: a home network, a data center, or a Kubernetes cluster.</p><p>Picture a Kubernetes cluster as a small town. Each worker node is a city block. Pods are apartments that can be torn down and rebuilt overnight. A service is the post office, which keeps the same address even when the residents change. The ingress controller is the town gate where outside traffic enters, and the control plane is city hall, with etcd as the records office that keeps the town&#39;s official state. Seen this way, a beginner can answer questions that are hard to answer from text: Where does a request come in? What happens when one apartment disappears?</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*3NsqUNx3f9PcrUs6Wg5EcQ.png" /><figcaption>Try this prompt: Prompt 2</figcaption></figure><p><strong>What to check: </strong>where the metaphor breaks. Pods don&#39;t live at a fixed address; the scheduler places them and can move them to a different node. Follow up with: “Where does this town metaphor mislead me about how Kubernetes really behaves?” A diorama is a teaching model, not an architecture document.</p><h3>3. Memory Palace: A Room for Every Fact</h3><p>The memory palace, known in psychology as the method of loci, is one of the oldest memory techniques on record. You place each item you want to remember at a specific spot along a familiar route, then mentally walk the route to recall the items in order. It is not folklore. In a study published in <em>Neuron</em>, people with ordinary memories who trained in the method of loci for six weeks roughly doubled the number of words they could recall from a list, and their brain connectivity shifted toward the patterns seen in world-class memory athletes (Dresler et al., 2017).</p><p>The technique works best for ordered lists, and computing has plenty of them: the seven OSI layers, the stages of the boot process, the phases of the software development life cycle, and the steps of an incident response plan.</p><p>Imagine a seven-story building. On the ground floor, the physical layer, copper cables run across the floor like vines. On the second floor, Data Link clerks stamp MAC addresses on envelopes in a mailroom. The third floor, Network, is a map room with IP routes pinned to every wall. The fourth, Transport, is a loading dock where a foreman counts boxes and resends any that go missing—TCP&#39;s reliability in one image. Session, Presentation, and Application fill the floors above, ending in a rooftop café where a web browser orders its page. The stranger and more specific each room, the better it sticks.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*2YdyfTJrsfWtCjpp4eH4fw.png" /><figcaption>Try this prompt: Prompt 3</figcaption></figure><p><strong>What to check: </strong>that each vivid image encodes the right fact. A memorable but wrong picture is worse than none, because it sticks. Confirm protocol-to-layer assignments against your course text, since models sometimes blur the OSI model with the four-layer TCP/IP model.</p><h3>4. Microscopic View: Zoom In on the Tiny</h3><p>A microscopic view enlarges what is too small to see. The original example zooms in on an ant’s face and legs. In computing, the most valuable microscope zooms down through levels of abstraction, from a line of code you wrote to the hardware that executes it.</p><p>Start with one line of Python: total = a + b. Zoom in once and you see bytecode, the instructions CPython’s virtual machine actually runs. Zoom again, and you reach machine code, perhaps a single ADD instruction on an x86 processor. Zoom in again, and you are inside the arithmetic logic unit, where adders are built from logic gates. One more level and you reach transistors switching on and off. Seeing those five levels as successive magnifications teaches something no single course does: abstraction is a stack of lenses, and every bug you will ever chase lives at one of them.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Nka8dm3KRm_6V-SQOmHJBA.png" /><figcaption>Try this prompt: Prompt 4</figcaption></figure><p><strong>What to check: </strong>the bytecode. Instruction names change between Python versions; Python 3.11 and later use a general BINARY_OP instruction where older versions used BINARY_ADD. You do not have to trust the AI. Run python -m dis on the line and compare. Checking a picture against a real tool is a habit worth building every time.</p><h3>5. Pop-Up Book: Make the Lesson Unfold</h3><p>A pop-up book turns a process into a sequence of pages, each one rising off the paper as you open it. The original example shows an egg becoming a butterfly. This word is best for processes with distinct stages, where the state of the system changes between steps.</p><p>Git is an ideal candidate, because most students learn its commands long before they understand what each command changes. Page one: a file edited in the working directory. Page two: git add lifts a snapshot of that file into the staging area. Page three: git commit folds the snapshot into a commit object with an author, a message, and a pointer to its parent. Page four: the branch label slides forward to the new commit. Page five: git push carries the commit to the remote repository. Each page shows exactly what moved, and nothing else.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*XxDXbWzbVi1QCCJOHF-29Q.png" /><figcaption>Try this prompt: Prompt 5</figcaption></figure><p>Other uses: the TLS handshake; a CPU’s fetch-decode-execute cycle; one iteration of a neural network’s training loop (forward pass, loss, backpropagation, weight update); the life of a Linux process from fork to exit.</p><p><strong>What to check: </strong>by running the steps in a throwaway repository. git status, git log — oneline — graph, and git cat-file -p HEAD show you the real objects, so you can confirm the pages match the machine.</p><h3>6. Museum Exhibit: Ideas in Glass Cases</h3><p>A museum exhibit puts related objects in separate cases so you can walk past them and compare. The original example sets cloud types side by side. This word is your comparison tool, and comparison is one of the most productive ways to learn. Schwartz and Bransford (1998) found that college students who first analyzed contrasting cases — examples that differ in deliberately chosen ways — got more out of a later lecture than students who simply read a summary of the differences. Noticing distinctions yourself prepares you to understand why they matter.</p><p>Sorting algorithms make a classic example. In the first case, bubble sort swaps neighbors one pair at a time: simple, stable, and O(n²) on average. In the second, merge sort splits and merges: guaranteed O(n log n) and stable, but it needs extra memory. In the third, quicksort partitions around a pivot: usually fast in practice at O(n log n) on average, though it degrades to O(n²) with poor pivots. In the fourth, heapsort builds a heap: O(n log n) even in the worst case and in place, but not stable (Cormen et al., 2022). Every placard lists the same facts in the same order, so your eye does the comparing.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*BVUHYU2JmOoxGCX99Ha4KA.png" /><figcaption>Try this prompt: Prompt 6</figcaption></figure><p>Other exhibits: SQL versus NoSQL databases; IaaS, PaaS, and SaaS; symmetric versus asymmetric encryption; supervised, unsupervised, and reinforcement learning.</p><p><strong>What to check: </strong>every complexity figure against your textbook. Comparison tables are where small errors hide, because one wrong cell looks exactly like the correct cells around it.</p><h3>7. Knolling: Lay Out Every Part</h3><p>Knolling means arranging related objects in parallel lines or at right angles on a flat surface so that everything is visible at once. The word has a specific origin. Andrew Kromelow coined it in 1987 while working as a janitor in Frank Gehry’s studio, where he lined up stray tools at right angles that echoed the furniture the studio was designing for Knoll; the artist Tom Sachs, who worked there at the same time, adopted the practice and wrote down its rules (Heathcote, n.d.). Its power is inventory: nothing hides behind anything else.</p><p>For IT students, knolling is a natural fit for hardware. Ask for a knolled layout of a desktop PC build, and you get the CPU, cooler, motherboard, memory modules, NVMe drive, graphics card, power supply, and cables lined up on a workbench, each labeled. You can see everything that belongs in the case before you open it.</p><p>Knolling also works on things that are not physical. Lay out the parts of a URL — scheme, host, port, path, query string, and fragment — in a neat row, and the anatomy of every web address you have typed becomes visible. Do the same for the header, payload, and signature of a JSON Web Token, or for each field in a line of ls -l output.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*PdxyGatRB_8TlHp9m0rHXQ.png" /><figcaption>Try this prompt: Prompt 7</figcaption></figure><p><strong>What to check: </strong>compatibility and labels. An image model may place a DDR4 module next to a DDR5 motherboard, or label the port as part of the path. Confirm the URL breakdown with your browser’s developer tools or the URL parser in your programming language.</p><h3>8. Make-Sticky-Notes: Facts You Can Test Yourself On</h3><p>This word turns key facts into short, bright notes. Used well, it is the best-supported technique on the list — but only if the notes become questions. Roediger and Karpicke (2006) showed that students who practiced retrieving material from memory retained far more a week later than students who spent the same time restudying it, even though the restudy group felt more confident. A major review of ten common study techniques rated practice testing among the most useful and rated highlighting and rereading among the least (Dunlosky et al., 2013). A wall of attractive notes that you only reread is highlighting in disguise.</p><p>So ask for notes with a question on the front and the answer on the back. Good candidates are the facts you must recall instantly on a certification exam or at a terminal prompt: well-known port numbers (22 for SSH, 53 for DNS, 443 for HTTPS, 3389 for RDP); Linux permission values (read is 4, write is 2, execute is 1, so chmod 755 gives the owner full rights and everyone else read and execute); HTTP status code families; and the Big-O cost of common data-structure operations.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*2bNCKRMr07pSt50ccdEB4Q.png" /><figcaption>Try this prompt: Prompt 8</figcaption></figure><p><strong>What to check: </strong>the list against the official exam objectives for the version you are taking, since objectives change between versions. And notice the final line of the prompt. It asks the AI to quiz you rather than tell you, which is the difference between studying and watching someone else study.</p><h3>9. Make a Mind Map: Link Your Ideas</h3><p>A mind map places one central idea in the middle and branches related ideas outward, so you can see how concepts connect. The original example maps how beans and water change the taste of coffee. Node-and-link diagrams have a solid research record. A meta-analysis of 55 studies found that learning with concept maps was associated with better knowledge retention across many settings (Nesbit &amp; Adesope, 2006), and a controlled study of medical students found that mind mapping improved recall of written material a week later (Farrand et al., 2002).</p><p>Here is the computing version of the coffee question: what makes a web page load slowly? Put “page load time” in the center. Branch out to the network (DNS lookup, TCP and TLS handshakes, distance to the server), the server (database queries, application code, caching), the payload (image size, JavaScript bundle, compression), and the browser (render-blocking CSS and scripts, layout, and paint). A vague complaint becomes a measurable checklist.</p><p>Because Claude and other text models can return a mind map as Mermaid code, you can render it, edit it, and commit it to a repository:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*NRE9Mel2oVuJ_R_28FUmAw.png" /><figcaption>Mindmap</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*A8a7y9IBTIjHR1Cwwfd57g.png" /><figcaption>Try this prompt: Prompt 9</figcaption></figure><p><strong>What to check: </strong>that the map becomes yours. Much of the benefit comes from building and revising a map, not from receiving one. Ask the AI for a skeleton, then add, move, and delete branches yourself as your understanding grows.</p><h3>10. Make-Schematic: Label Every Part</h3><p>A schematic is a clean, labeled drawing of parts and their connections. The original example suggests a car engine. For IT and CS students, the schematic is the native language of the field: network topology diagrams, system architecture diagrams, entity-relationship diagrams, and logic circuit diagrams are all schematics.</p><p>Try a three-tier web application. A browser connects over HTTPS on port 443 to a load balancer, which spreads requests across two web servers. Both call an application server, which checks a Redis cache and reads and writes a PostgreSQL database. Drawn as a schematic, the request path, the single points of failure, and the places where a firewall rule belongs all become visible. This is the advantage Larkin and Simon (1987) described: when each component sits next to the things it connects to, the next question is right there on the page.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*WkpodhOZySPPulKmyc8Ubg.png" /><figcaption>Flowchart</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Co4lxIoL-afQnnPonwSvbg.png" /><figcaption>Try this prompt: Prompt 10</figcaption></figure><p><strong>What to check: </strong>default ports (PostgreSQL listens on 5432 and Redis on 6379 by default) and the direction of every arrow. Then ask the follow-up the diagram makes possible: “If the app server fails, does the whole site go down, and how would I fix that?”</p><h3>11. Exploded View: Pull It Apart in Place</h3><p>An exploded view separates an object’s parts along invisible axes so you can see how they fit together and the order in which they assemble. It differs from knolling in one important way. Knolling lays parts out flat as an inventory; an exploded view keeps every part lined up with the place it came from, so the relationships survive. It is the drawing on the first page of every repair manual.</p><p>For hardware students, the obvious subject is a laptop: keyboard deck lifted away, then the battery, the cooling assembly, the motherboard, the storage, and the bottom panel, each floating above the next. For software students, the same idea applies to a stack. Explode a Linux system, and you see applications at the top, then system libraries, then the system call interface, then the kernel with its scheduler, memory manager, file systems, network stack, and device drivers, and finally the hardware. The gaps between the layers are where the interfaces live.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*qhCfLNazpv9Weh9OVQ78jw.png" /><figcaption>Try this prompt: Prompt 11</figcaption></figure><p><strong>What to check: </strong>where the model puts each component. A common slip is placing device drivers in user space; most Linux drivers run in the kernel or as loadable kernel modules. For hardware, compare the drawing with the manufacturer’s service manual before you pick up a screwdriver, since real laptops often have glued batteries and hidden screws that a generated image will not show.</p><h3>12. Transit Map: Routes, Stops, and Transfers</h3><p>A transit map strips geography away and keeps only what a traveler needs: lines, stops, and the stations where you change lines. That simplification is why a subway map is readable when a street map is not. Anything in computing that moves along a path can be drawn this way.</p><p>Consider a single click in an online store built from microservices. The request boards at the API gateway, rides the Orders line to the inventory service, transfers at a message queue to the Payments line, and stops at a notification service before the confirmation returns. In a transit map, a student can see which stations every trip passes through—and therefore which outage would stop every train.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*STEB5RVrk-wXH0vU46x0dw.png" /><figcaption>Try this prompt: Prompt 12</figcaption></figure><p>Other routes: an ETL data pipeline from source systems to a data warehouse; a packet’s hops across the internet; a student’s path through prerequisite courses toward a degree.</p><p><strong>What to check: </strong>the route against reality. In a real system, distributed tracing tools record the actual path a request took. If you have access to a trace from a class project or lab, compare it with the map; generated maps tend to make systems look tidier than they are.</p><h3>13. Comic Strip: Let the Parts Talk</h3><p>A comic strip turns a process into characters having a conversation, panel by panel. That suits computing better than it first appears, because protocols are conversations: one machine says something, the other replies, and both follow strict rules about who speaks when. Cryptographers have used characters this way for decades — Alice and Bob trading messages while Eve listens in.</p><p>Draw the TCP three-way handshake as three panels. In the first, the client says, “SYN: I’d like to talk, and I’ll start counting from here.” In the second, the server answers, “SYN-ACK: I heard you, and here’s where I’ll start counting.” In the third, the client replies, “ACK: Got it. Let’s go.” A sequence that reads like alphabet soup in a textbook becomes a polite exchange you will remember.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ha4dN_e4Hf-V2bLRlIk6iA.png" /><figcaption>Try this prompt: Prompt 13</figcaption></figure><p>Other strips: Alice, Bob, and an eavesdropper explaining public-key encryption; a browser and a DNS resolver tracking down an IP address; two threads arguing over a shared variable to show a race condition.</p><p><strong>What to check: </strong>the protocol details hidden in the jokes. In the handshake, each acknowledgment number is the other side’s sequence number plus one; image models sometimes get that wrong or swap who sends what. Capture a real handshake in Wireshark and compare the flags and numbers panel by panel.</p><h3>14. Assembly Line: Watch the Product Change</h3><p>An assembly line shows raw material moving past a series of stations, with the product changing at each one. It is the right picture whenever data is transformed in stages, and each stage hands its output to the next.</p><p>A compiler is the classic case. Source code arrives at the first station, where the lexer chops it into tokens. The parser assembles the tokens into a syntax tree. Semantic analysis inspects the tree for type errors. The optimizer reshapes it to run faster, and the code generator stamps out machine instructions at the end of the line. Showing what the code looks like between each station is what makes this picture worth asking for.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*4oYU_DBWgAsscAiTLWk47w.png" /><figcaption>Try this prompt: Prompt 14</figcaption></figure><p>Other lines: a CI/CD pipeline from commit through build, test, and deploy; a machine learning pipeline from raw data through cleaning, feature engineering, training, and evaluation.</p><p><strong>What to check: </strong>the intermediate forms with real tools. gcc -S writes the assembly a C compiler actually produces, and Python’s built-in ast module will print the real syntax tree for a line of Python. Real compilers also merge or split these stages in ways a neat five-station diagram hides.</p><h3>15. Field Guide: Learn to Spot the Species</h3><p>A field guide teaches identification. Each page shows one species with its distinguishing marks, where it lives, and how to tell it from the lookalikes. Much of IT is identification work: recognizing an attack, an error, a log pattern, or a design pattern quickly enough to act on it.</p><p>Phishing is the best-known example. A field guide page for a credential-harvesting email would show the field marks: a sender address that does not match the organization it claims to be, urgent language that pushes you to act before thinking, a link whose real destination differs from its display text, and an unexpected request to sign in or open an attachment. Seen across several species — invoice scams, fake password resets, gift-card requests from a “boss” — the patterns start to jump out.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*vCL2UbfL2SXnfLP-b809gw.png" /><figcaption>Try this prompt: Prompt 15</figcaption></figure><p>Other guides: common Python or Java error messages and their usual causes; malware families by behavior; the Gang of Four design patterns; suspicious entries in a web server log.</p><p><strong>What to check: </strong>your guide against current guidance from your school’s or employer’s security team, because attack styles change quickly. Use mock examples only, and never click links or open attachments in real suspicious messages while studying them.</p><h3>16. Before and After: See What the Change Did</h3><p>A before-and-after picture shows the same thing in two states with the differences highlighted. It is a contrasting case built for one purpose: making a change visible. That connects it to the research on contrasting cases discussed under the museum exhibit, where noticing differences prepared students to understand the principle behind them (Schwartz &amp; Bransford, 1998).</p><p>Database normalization is a natural fit. The “before” table crams customer names, addresses, order dates, and product names into one wide grid, with the same customer’s address repeated on every order row. The “after” picture splits it into customers, orders, and products tables linked by keys, and the duplicated data disappears. Seeing the two side by side explains why normalization exists better than any definition of third normal form.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*rhOD49a0PE9g8u76_jlE8w.png" /><figcaption>Try this prompt: Prompt 16</figcaption></figure><p>Other pairs: a 200-line function before and after refactoring; a firewall rule set before and after removing an overly broad “allow any” rule; a neural network’s predictions before and after training.</p><p><strong>What to check: </strong>the “after” with the rules, not the picture. For normalization, confirm that every non-key column depends on the whole key and on nothing but the key. For a refactoring, run the tests before and after; a change that looks cleaner but breaks behavior is not an improvement.</p><h3>17. X-Ray: See the Skeleton Under the Skin</h3><p>An X-ray shows the hidden structure beneath a familiar surface while keeping the surface faintly visible, so you can see how the two line up. It differs from a cross section, which slices a thing open; an X-ray looks straight through it, which makes it ideal for interfaces that hide machinery.</p><p>A web page is the perfect patient. On the surface: a login form, a logo, a button. Under the X-ray: the DOM tree that structures the page, the CSS boxes that position each element, and the network requests that fire when you press the button — a POST to an authentication endpoint, a token coming back, a redirect. For students learning front-end development or web security, seeing the skeleton and skin together connects what users see with what the code does.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*b7b1eCBfsgxXRcMiJPWZxQ.png" /><figcaption>Try this prompt: Prompt 17</figcaption></figure><p><strong>What to check: compare the X-ray against your browser’s developer tools, which show you the real thing</strong>. The Elements panel shows the actual DOM, and the Network panel shows every request the page really makes. Practice on pages you build yourself or on sites you are authorized to test.</p><h3>18. Choose-Your-Own-Adventure: Every Choice Leads Somewhere</h3><p>A choose-your-own-adventure story asks a question at the bottom of each page and sends you to a different page depending on your answer. Under the playful surface, it is a decision tree, which makes it the perfect disguise for troubleshooting. Help desk work, network diagnosis, and debugging all come down to asking the right question next.</p><p>Begin with a complaint every IT student will hear: “My laptop can’t reach the internet.” Page one asks whether the Wi-Fi icon shows a connection. If yes, turn to the page that asks whether the laptop has a valid IP address. An address beginning with 169.254 sends you down the DHCP path; a valid address sends you on to ping the default gateway, then a public IP address, then a domain name. Every path ends at a fix or a handoff. Because the reader makes the choices, the story trains diagnostic thinking instead of memorizing a checklist.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*7Mqwtj2pbcv5p4fuZsaxTA.png" /><figcaption>Flowchart TD</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Or0mXAf8gvnyKbualUJT3w.png" /><figcaption>Try this prompt: Prompt 18</figcaption></figure><p><strong>What to check: walk every path</strong>, ideally with a classmate playing the confused user. Look for dead ends, loops, and pages that ask two questions at once. A troubleshooting tree is only as good as its worst branch.</p><h3>Quick Reference</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*CTFR0KUtJunFv89ua8mnEQ.png" /></figure><h3>Stacking Power Words</h3><p>The words become more powerful in combination. A museum exhibit of knolled parts compares three server builds side by side. A memory palace whose rooms are cross sections lets you walk through a protocol stack one layer at a time. A schematic followed by sticky notes turns a system you have drawn into questions you can answer on tomorrow’s bus ride. A transit map followed by an X-ray shows a request’s route and then what each stop is really doing underneath. For a brand-new topic, try this sequence:</p><p>1. <strong>Isometric diorama</strong> to see the whole system at a glance.</p><p>2. <strong>Schematic</strong> to pin down the exact parts and connections.</p><p>3. <strong>Cross-section</strong> or <strong>microscopic view</strong> to look inside the part that confuses you.</p><p>4. <strong>Museum exhibit</strong> to compare it with its alternatives.</p><p>5. <strong>Sticky notes</strong> to test yourself until you no longer need the pictures.</p><h3>Verify, Don’t Just Admire</h3><p>There is a risk built into all of this, and it grows as image models improve. A beautiful, confident picture feels true. NIST’s risk-management guidance for generative AI lists confabulation — confidently stated content that is false or erroneous — among the core risks of these systems (National Institute of Standards and Technology, 2024). In a picture, confabulation looks like a mislabeled header, a reversed arrow, or a port number that is off by one. Because the rest of the image is polished, the error does not stand out.</p><p>Computing students have an advantage: your subject can talk back. You can run python -m dis, open Wireshark, inspect a Git object, call a URL parser, or read a man page. Make it a rule that every picture that teaches you a fact gets checked against documentation or a running system before that fact goes into your notes.</p><p>The second risk concerns you rather than the AI. In a field experiment with nearly a thousand high school math students, those given an unrestricted GPT-4 assistant performed better during practice but worse on later exams once the AI was taken away, while students using a version designed to give hints instead of answers largely avoided that loss (Bastani et al., 2025). The power words in this article are meant to work like the second tutor. Use them to see, question, and test your understanding, not to skip the work of building it. When a visual makes something click, close it and try to redraw it from memory. If you can, you have learned it. If you cannot, you have found exactly what to study next.</p><p>Finally, follow your course’s policy on AI tools, and disclose AI assistance when it shaped work you submit.</p><h3>Back in the Study Lounge</h3><p>The vending machine still hums, and the fluorescent tube still flickers, but the student is no longer staring at a paragraph. She has a cutaway of a packet, a seven-story building that holds the OSI model, and a stack of digital sticky notes that will quiz her on the bus in the morning. Nothing about networking changed overnight. What changed is that she asked her tools to show her instead of tell her — and then checked their work.</p><p>Eighteen words. Pick one, add your topic, and start drawing.</p><h3>References</h3><p>Anthropic. (2026, March 12). <em>Claude now creates interactive charts, diagrams and visualizations</em>. Claude. <a href="https://claude.com/blog/claude-builds-visuals">https://claude.com/blog/claude-builds-visuals</a></p><p>Bastani, H., Bastani, O., Sungu, A., Ge, H., Kabakçı, Ö., &amp; Mariman, R. (2025). Generative AI without guardrails can harm learning: Evidence from high school mathematics. <em>Proceedings of the National Academy of Sciences, 122</em>(26), Article e2422633122. <a href="https://doi.org/10.1073/pnas.2422633122">https://doi.org/10.1073/pnas.2422633122</a></p><p>Clark, J. M., &amp; Paivio, A. (1991). Dual coding theory and education. <em>Educational Psychology Review, 3</em>(3), 149–210. <a href="https://doi.org/10.1007/BF01320076">https://doi.org/10.1007/BF01320076</a></p><p>Cormen, T. H., Leiserson, C. E., Rivest, R. L., &amp; Stein, C. (2022). <em>Introduction to algorithms</em> (4th ed.). MIT Press.</p><p>Dresler, M., Shirer, W. R., Konrad, B. N., Müller, N. C. J., Wagner, I. C., Fernández, G., Czisch, M., &amp; Greicius, M. D. (2017). Mnemonic training reshapes brain networks to support superior memory. <em>Neuron, 93</em>(5), 1227–1235.e6. <a href="https://doi.org/10.1016/j.neuron.2017.02.003">https://doi.org/10.1016/j.neuron.2017.02.003</a></p><p>Dunlosky, J., Rawson, K. A., Marsh, E. J., Nathan, M. J., &amp; Willingham, D. T. (2013). Improving students’ learning with effective learning techniques: Promising directions from cognitive and educational psychology. <em>Psychological Science in the Public Interest, 14</em>(1), 4–58. <a href="https://doi.org/10.1177/1529100612453266">https://doi.org/10.1177/1529100612453266</a></p><p>Farrand, P., Hussain, F., &amp; Hennessy, E. (2002). The efficacy of the “mind map” study technique. <em>Medical Education, 36</em>(5), 426–431. <a href="https://doi.org/10.1046/j.1365-2923.2002.01205.x">https://doi.org/10.1046/j.1365-2923.2002.01205.x</a></p><p>Heathcote, E. (n.d.). <em>Knolling</em>. R / D. Retrieved November 18, 2026, from <a href="https://readingdesign.org/knolling">https://readingdesign.org/knolling</a></p><p>Hegarty, M. (2011). The cognitive science of visual-spatial displays: Implications for design. <em>Topics in Cognitive Science, 3</em>(3), 446–474. <a href="https://doi.org/10.1111/j.1756-8765.2011.01150.x">https://doi.org/10.1111/j.1756-8765.2011.01150.x</a></p><p>Larkin, J. H., &amp; Simon, H. A. (1987). Why a diagram is (sometimes) worth ten thousand words. <em>Cognitive Science, 11</em>(1), 65–100. <a href="https://doi.org/10.1111/j.1551-6708.1987.tb00863.x">https://doi.org/10.1111/j.1551-6708.1987.tb00863.x</a></p><p>Mayer, R. E. (2021). <em>Multimedia learning</em> (3rd ed.). Cambridge University Press. <a href="https://doi.org/10.1017/9781316941355">https://doi.org/10.1017/9781316941355</a></p><p>Mermaid. (n.d.). <em>About Mermaid</em>. Retrieved November 18, 2026, from <a href="https://mermaid.js.org/intro/">https://mermaid.js.org/intro/</a></p><p>National Institute of Standards and Technology. (2024). <em>Artificial intelligence risk management framework: Generative artificial intelligence profile</em> (NIST AI 600–1). U.S. Department of Commerce. <a href="https://doi.org/10.6028/NIST.AI.600-1">https://doi.org/10.6028/NIST.AI.600-1</a></p><p>Nesbit, J. C., &amp; Adesope, O. O. (2006). Learning with concept and knowledge maps: A meta-analysis. <em>Review of Educational Research, 76</em>(3), 413–448. <a href="https://doi.org/10.3102/00346543076003413">https://doi.org/10.3102/00346543076003413</a></p><p>OpenAI. (2026, April 21). <em>Introducing ChatGPT Images 2.0</em>. <a href="https://openai.com/index/introducing-chatgpt-images-2-0/">https://openai.com/index/introducing-chatgpt-images-2-0/</a></p><p>Roediger, H. L., III, &amp; Karpicke, J. D. (2006). Test-enhanced learning: Taking memory tests improves long-term retention. <em>Psychological Science, 17</em>(3), 249–255. <a href="https://doi.org/10.1111/j.1467-9280.2006.01693.x">https://doi.org/10.1111/j.1467-9280.2006.01693.x</a></p><p>Schwartz, D. L., &amp; Bransford, J. D. (1998). A time for telling. <em>Cognition and Instruction, 16</em>(4), 475–522. <a href="https://doi.org/10.1207/s1532690xci1604_4">https://doi.org/10.1207/s1532690xci1604_4</a></p><p>White, J., Fu, Q., Hays, S., Sandborn, M., Olea, C., Gilbert, H., Elnashar, A., Spencer-Smith, J., &amp; Schmidt, D. C. (2023). <em>A prompt pattern catalog to enhance prompt engineering with ChatGPT</em> (arXiv:2302.11382). arXiv. <a href="https://doi.org/10.48550/arXiv.2302.11382">https://doi.org/10.48550/arXiv.2302.11382</a></p><p>xAI. (2026, August 7). <em>Image 2.0</em>. <a href="https://x.ai/news/grok-imagine-image-2">https://x.ai/news/grok-imagine-image-2</a></p><p><strong>About the Author</strong></p><p>Ron McFarland, Ph.D. With a career spanning over 30 years in computer science, cybersecurity, and artificial intelligence, Dr. Ron McFarland brings unparalleled expertise to his series on stock market trading and AI. Having earned a Ph.D. in Computer Science and a postdoctoral fellowship in cybersecurity, he combines academic knowledge with decades of hands-on experience in artificial intelligence, machine learning, cloud security, database systems, and risk frameworks. Dr. McFarland has taught advanced AI and machine learning courses at the university level, equipping students and professionals to harness cutting-edge technology — a foundation that translates seamlessly to AI’s transformative role in stock trading.</p><p>Beyond his academic and consulting achievements, Dr. McFarland is a prolific writer, having authored several books and dozens of technical articles on AI, blockchain, and cybersecurity. His ability to simplify complex concepts makes his work accessible to readers of all levels, from beginners to advanced traders. As a former Dean and professor who has led programs in computer science, technical analysis, and programming.</p><p>YouTube Channel: <a href="https://www.youtube.com/@HighervistaLLC">https://www.youtube.com/@HighervistaLLC</a></p><p>Amazon Author’s Page: <a href="https://amazon.com/author/ronmcfarland">amazon.com/author/ronmcfarland</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=1ea17d3f23f0" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[AI and Government: What the Evidence Shows (2025–2026)]]></title>
            <link>https://highervista.medium.com/ai-and-government-what-the-evidence-shows-2025-2026-95cc32acfe3b?source=rss-1f7645bf9aea------2</link>
            <guid isPermaLink="false">https://medium.com/p/95cc32acfe3b</guid>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[information-technology]]></category>
            <category><![CDATA[software-development]]></category>
            <category><![CDATA[large-language-models]]></category>
            <dc:creator><![CDATA[Ron McFarland PhD]]></dc:creator>
            <pubDate>Thu, 19 Nov 2026 15:06:02 GMT</pubDate>
            <atom:updated>2026-11-19T15:06:02.966Z</atom:updated>
            <content:encoded><![CDATA[<p>Briefing prepared November 2026</p><p>By Ron McFarland, Ph.D.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*bCE1d_bnYcgVp3scAqSAmA.png" /><figcaption>AI and Government</figcaption></figure><h3>Author’s Introduction and Notes</h3><p>Artificial intelligence is changing how governments spend money, deliver services, defend networks, and handle personal information. I prepared this briefing to explain several changes supported by published research, official records, company disclosures, or established news organizations. Since I am writing this to support my CS/IT and AI students (and any other readers interested in AI), each section defines key terms, describes what the evidence shows, and cites the source in APA 7th edition style. Facts were verified and are current as of November 2026. Drop me a note at my email if you have any questions or additions (especially as this article ages).</p><p>I prepared this briefing as a fact-check of a viral Instagram interview that linked AI investment, government bailouts, and a “control grid” of digital identity and digital currency. I left out the interview’s predictions and claims about companies’ hidden motives because I could not test them or find significant research supporting them, so I left those claims out of this article. The table below lists only the testable claims and what the record shows.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*EgRKDuiQaUN_y8H0XEnRhw.png" /><figcaption>Table 1: Claims found and fact-checks</figcaption></figure><h3>The AI Investment Boom</h3><h3>How Much of US Growth Comes From AI?</h3><p>Gross domestic product (GDP) measures the total value of goods and services an economy produces, and economists can estimate how much each type of spending adds to its growth. ING’s chief international economist, James Knightley, calculated technology investment’s share of US year-over-year GDP growth in the second quarter of 2026 in three ways (Knightley, 2026). The broadest measure, which counts all information-processing equipment, software and data-center construction, came to 50.2%. A stricter measure came to 44%. When Knightley subtracted computers, peripherals and semiconductors imported from abroad, the share fell to 36%, which he called the fairest measure available (Knightley, 2026).</p><p>The difference matters because Americans buy imported hardware made elsewhere, so it does not add to US production. Knightley concluded that tech investment accounts for about one-third of current US growth (Knightley, 2026). The “half of GDP growth” figure is therefore real but sits at the top of a range.</p><h3>What “Circular Financing” Means</h3><p>Circular financing describes deals in which money flows from a supplier to a customer and then back to the supplier as sales revenue. On September 22, 2025, OpenAI and NVIDIA announced a letter of intent to deploy at least 10 gigawatts of NVIDIA systems, with NVIDIA intending to invest up to $100 billion in OpenAI as each gigawatt is deployed (NVIDIA, 2025). In the same period, OpenAI committed $300 billion to Oracle for computing capacity, and Oracle announced roughly $40 billion in purchases of NVIDIA chips (Dellinger, 2025). Reporting on these deals described money cycling among a small group of companies, with each one’s revenue depending on the others (Jadhav, 2025).</p><p>The scale is large but not unlimited. UBS estimated that the NVIDIA–OpenAI arrangement could represent up to 13% of NVIDIA’s projected 2026 revenue (UBS, 2025).</p><h3>The November 2026 Selloff</h3><p>Stock prices for AI companies fell sharply in early November 2026. On November 5, the Nasdaq Composite dropped 5.5%, and the S&amp;P 500 fell 4.4%, while the Dow Jones Industrial Average lost more than 1,800 points (Associated Press, 2026a). Reuters reported that the Nasdaq’s decline was its deepest one-day drop since April 2025 and that investors were reacting to Democratic gains in Congress and the prospect of closer scrutiny of AI, especially data center construction (Reuters, 2026). After that day, the Nasdaq stood 11.9% below its record high, which meets the common definition of a market correction (Associated Press, 2026b). A correction is a fall in stock prices; it does not by itself measure whether AI companies’ businesses are failing.</p><h3>Public Money and Public Costs</h3><p>Public money reaches the AI and chip industries through three documented channels: government ownership stakes, state tax exemptions, and electricity prices paid by households and businesses.</p><h3>Government Ownership Stakes</h3><p>In August 2025, the US government agreed to buy 433.3 million shares of Intel at $20.47 each, a 9.9% stake worth $8.9 billion (RTTNews, 2025). The purchase was funded by converting $5.7 billion in CHIPS and Science Act grants that had been awarded but not yet paid, plus $3.2 billion from the Secure Enclave program (RTTNews, 2025). In June 2026, President Trump said the government might take equity stakes in AI companies (Ha, 2026). He later told TIME that he would not nationalize frontier AI labs but might take stakes, and Anthropic and the administration denied discussing a stake in Anthropic (Stan, 2026).</p><h3>The “Backstop” Controversy</h3><p>A government “backstop” is a guarantee that the government will cover losses if a borrower cannot repay. At a Wall Street Journal event on November 5, 2025, OpenAI’s chief financial officer, Sarah Friar, suggested the US government could backstop the company’s infrastructure loans to lower its borrowing costs (Gülen, 2025). She retracted the remark the same day, writing that OpenAI was not seeking a government backstop. White House AI adviser David Sacks then wrote that there would be no federal bailout for AI, and OpenAI CEO Sam Altman wrote that the company does not have or want government guarantees for its data centers (Gülen, 2025).</p><h3>State Tax Exemptions</h3><p>Many states exempt data-center equipment from sales and use taxes to attract construction. A November 2025 report by Good Jobs First, a nonprofit that tracks business subsidies, found that at least 36 states offer such subsidies, but only 11 disclose which companies receive them (Hardy, 2025). States that calculated their return on these exemptions found they lose between 52 and 70 cents for every dollar exempted, and Virginia alone forgoes nearly $1 billion a year in state and local sales and use taxes (Hardy, 2025).</p><p>Not everyone agrees these exemptions are subsidies. The Cato Institute argues that many of them remove taxes on business inputs, such as equipment, that sound tax policy would not tax in the first place (Michel, 2026).</p><h3>Electricity Bills and the Capacity Market</h3><p>PJM Interconnection runs the electric grid for 67 million people in 13 states and Washington, D.C. (Crosley, 2026). It holds a “capacity auction” that pays power-plant owners in advance to guarantee enough supply for future peak demand, and the cost of those payments falls on electricity customers (Crosley, 2026; Kunkel, 2025). The capacity price rose from $28.92 per megawatt-day for 2024–25 to $329.17 for 2026–27, and the 2026–27 price would have been higher without a price cap (Kunkel, 2025).</p><p>PJM’s independent market monitor, Monitoring Analytics, estimated that data-center demand caused 63% of the price increase in the 2025–26 auction, adding $9.3 billion to customer costs (Crosley, 2026). In a later report on the most recent auction, the monitor found that forecasts of data centers not yet built were the main driver: speculative forecasts raised capacity prices by 61%, while existing and committed data centers accounted for 4% (Ohio Manufacturers’ Association, n.d.).</p><h3>Personal Data and AI Advertising</h3><p>Advertising pays for many free online services, and the more a company knows about a user, the more precisely it can target ads. Two major AI companies now connect AI conversations to advertising in different ways.</p><h3>Meta</h3><p>Meta reports that more than 1 billion people use Meta AI every month (Meta, 2025). Beginning December 16, 2025, Meta started using people’s interactions with its AI features, by voice or text, to personalize the content and ads they see (Meta, 2025). For example, someone who chats with Meta AI about hiking may start seeing hiking groups and ads for hiking boots (Meta, 2025). Meta said it would not use conversations about sensitive topics, such as religion, health, and politics, for ad targeting (Rijo, 2025). Users can adjust their ad settings, but a Meta privacy manager told Bloomberg that users cannot opt out of having their AI chats used (<em>Meta to Use AI to Tailor Ads</em>, 2025).</p><h3>OpenAI</h3><p>On January 16, 2026, OpenAI announced plans to test ads in ChatGPT for logged-in adults in the United States on its Free and Go tiers (OpenAI, 2026a). It set out principles that include “answer independence,” meaning ads do not influence ChatGPT’s answers and are always separate and clearly labeled (OpenAI, 2026a). Plus, Pro, Business, Enterprise, and Education subscriptions do not carry ads, and Free users can opt out of ads in exchange for fewer daily messages (OpenAI, 2026b).</p><h3>Teens and AI Companions</h3><p>AI companions are chatbots designed for personal conversation, described to survey respondents as “digital friends or characters you can text or talk with” (Common Sense Media, 2025b). A nationally representative Common Sense Media survey of 1,060 US teens aged 13 to 17, conducted in April and May 2025, found that 72% had used AI companions at least once and 52% used them at least a few times a month (Common Sense Media, 2025a, 2025b). About a third used them for social interaction and relationships, and about a quarter shared personal information with them (Common Sense Media, 2025b). Half of teens said they do not trust AI companions’ information or advice, and 80% of users still spend more time with real friends (Common Sense Media, 2025b). Based on its risk assessment, Common Sense Media recommends that no one under 18 use AI companions (Common Sense Media, 2025b).</p><h3>Digital Currency and Digital Identity</h3><h3>Digital Currency</h3><p>A central bank digital currency (CBDC) is government-issued digital money, distinct from privately issued stablecoins and from the digital bank balances people already use. US policy now prohibits one. Executive Order 14178, signed January 23, 2025, bars federal agencies from establishing, issuing, or promoting a CBDC (Exec. Order №14178, 2025). Congress then wrote a ban into law: the 21st Century ROAD to Housing Act passed the House 358–32 and prohibits the Federal Reserve from issuing a CBDC, or anything substantially similar, through December 31, 2030 (<em>Congress Sends Anti-CBDC Housing Bill</em>, 2026). The act became law on July 11, 2026, after President Trump neither signed nor vetoed it (Cooley, 2026).</p><h3>Digital Identity</h3><p>A mobile driver’s license (mDL) is a digital version of a state-issued license stored on a smartphone. The United States has no federal government-backed digital identity system; states issue these credentials themselves (“Implementing Government-Backed Digital Identity,” 2024). As of September 2026, 21 states offered mDLs accepted at Transportation Security Administration (TSA) checkpoints, and TSA describes digital IDs as more secure than physical cards, which can be counterfeited (Pack, 2026). Colorado officials and TSA still advise travelers to carry a physical ID as a backup (Toomer, 2026).</p><h3>Chip Supply Chains and Critical Minerals</h3><p>Semiconductors depend on specialized minerals, and China dominates the supply of some of them. China produces about 94% of the world’s gallium, used in chips, LEDs, and solar panels, according to a 2024 European Union report (Mishra, 2025). In December 2024, China banned exports of gallium, germanium, and antimony to the United States, leading to shortages for US users (Mishra, 2025). China’s Ministry of Commerce suspended that ban in November 2025 until November 27, 2026; the minerals remain on China’s export-control list, so licenses are still required (Mishra, 2025).</p><h3>AI and Jobs</h3><p>A Stanford study used payroll records rather than surveys. Erik Brynjolfsson, Bharat Chandar, and Ruyu Chen of Stanford analyzed data from the largest US payroll software provider (Brynjolfsson et al., 2025). They found that, after generative AI spread, workers aged 22 to 25 in the occupations most exposed to AI saw a 13% relative decline in employment, even after controlling for conditions at individual firms (Brynjolfsson et al., 2025).</p><p>The pattern was specific. Older workers in the same occupations and young workers in less-exposed fields held steady or grew. Declines were concentrated where AI tends to automate tasks rather than assist workers, and the adjustment showed up mainly in employment levels rather than in wages (Brynjolfsson et al., 2025). The results held when technology firms and remote-friendly occupations were excluded (Brynjolfsson et al., 2025).</p><h3>How Government Uses AI</h3><h3>What Agencies Report</h3><p>Federal agencies must publish annual inventories of how they use AI. The 2025 inventory listed 3,611 use cases across 56 agencies, up from 1,757 in 2024, with 1,818 deployed or in pilot testing; the Department of Health and Human Services reported the most, 447 (Aldane, 2026). Science-related applications were the most common type (<em>Disclosed Government AI Use Increased</em>, 2026). The Center for Democracy &amp; Technology, using a different 2024 baseline, put the growth at nearly 70% and found inconsistent reporting and too little detail about high-impact uses (Anex-Ries, 2026).</p><h3>Fraud Prevention at the Treasury</h3><p>Machine learning is a type of AI that finds patterns in large datasets. The Treasury Department makes about 1.4 billion payments a year (Egan, 2024). It reported preventing and recovering more than $4 billion in fraudulent and improper payments in fiscal 2024, up from $652.7 million the year before (<em>AI Tools Helped Treasury Recover Billions</em>, 2024). The total included $2.5 billion prevented by identifying high-risk transactions, $1 billion in check fraud recovered with machine learning, $500 million from expanded risk-based screening, and $180 million from payment-schedule changes (<em>AI Tools Helped Treasury Recover Billions</em>, 2024; Egan, 2024).</p><h3>Automated Cyber Defense</h3><p>The Defense Advanced Research Projects Agency (DARPA) and the Advanced Research Projects Agency for Health (ARPA-H) ran the AI Cyber Challenge, a competition to build systems that automatically find and fix software flaws. In the final round, teams’ systems found 54 of 63 planted vulnerabilities and patched 43, across more than 54 million lines of code, with patches arriving in an average of 45 minutes (Defense Advanced Research Projects Agency [DARPA], 2025). The systems also discovered 18 real, previously unknown vulnerabilities, and all seven finalist systems are being released as open source (DARPA, 2025).</p><h3>When Automated Government Systems Fail</h3><p>Three well-documented cases show what happens when automated systems make consequential decisions about people with little human review.</p><h3>Michigan’s MiDAS</h3><p>Michigan’s Unemployment Insurance Agency used a $47 million system called MiDAS to decide fraud cases from 2013 to 2015. The state acknowledged that about 40,000 people were accused of fraud by a system that operated without human supervision and had an error rate as high as 93% (AI Incident Database, n.d.). In January 2024, the Michigan Court of Claims approved a $20 million class-action settlement for people who were falsely accused (Carr, 2024).</p><h3>Australia’s Robodebt</h3><p>Australia’s Robodebt scheme averaged people’s annual income from tax records to calculate welfare overpayments, then issued debts that recipients had to disprove (JURIST, 2023; <em>The Flawed Algorithm</em>, 2023). The scheme issued about 470,000 wrongful debts before it ended in 2020 (JURIST, 2023). A Federal Court ruling in 2019 found the debt notices invalid (<em>The Flawed Algorithm</em>, 2023). The Royal Commission’s 2023 report found the scheme unlawful, described it as a “crude and cruel” mechanism and referred officials for possible prosecution (<em>“Crude and Cruel,”</em> 2023; <em>JURIST</em>, 2023). It was meant to save $4.7 billion (Australian) but saved $406 million while costing $971 million to run, and the government settled a class action for $1.2 billion (<em>Gross Betrayal</em>, 2023).</p><h3>The Dutch Childcare Benefits Scandal</h3><p>The Dutch tax authority used a risk-scoring algorithm to flag childcare benefit claims as possible fraud. About 26,000 families were wrongly accused (<em>How a Discriminatory Algorithm</em>, 2021). Amnesty International found that “Dutch citizenship: yes/no” was a factor in the model, so non-Dutch nationals received higher risk scores (Amnesty International, 2021). The prime minister and his cabinet resigned in January 2021 (<em>Dutch Government Resigns</em>, 2021).</p><h3>Immigration Enforcement Technology</h3><p>In April 2025, Immigration and Customs Enforcement (ICE) awarded Palantir a contract worth about $30 million to build “ImmigrationOS,” which extends ICE’s case-management system to track visa overstays and self-deportations in near real time (<em>ICE Just Ordered $30 Million</em>, 2025). The contract describes the system as covering the “immigration lifecycle from identification to removal” (<em>ICE Just Ordered $30 Million</em>, 2025).</p><h3>Cybersecurity</h3><h3>An AI-Orchestrated Espionage Campaign</h3><p>In November 2025, Anthropic, the company that makes the Claude AI assistant used to draft this briefing, published a report on a campaign it detected in mid-September 2025 (Anthropic, 2025). It attributed the campaign with high confidence to a Chinese state-sponsored group it designated GTG-1002. The attackers used Claude Code as an automated agent against about 30 targets, including technology companies, financial institutions, chemical manufacturers, and government agencies in several countries (Anthropic, 2025).</p><p>According to the report, the AI carried out an estimated 80% to 90% of tactical operations, while humans made key decisions such as approving exploitation and data theft (Anthropic, 2025). The AI also “frequently overstated findings and occasionally fabricated data,” such as claiming credentials that did not work, and only a handful of intrusions succeeded (Anthropic, 2025). Anthropic banned the accounts, notified affected organizations, and coordinated with authorities (Anthropic, 2025). This is one company’s account of the misuse of its own product.</p><h3>Deepfake Impersonation of Officials</h3><p>A deepfake is AI-generated audio, video, or images made to imitate a real person. In 2025, an impostor used AI-generated voice and text messages to pose as Secretary of State Marco Rubio and contacted at least three foreign ministers, a US senator, and a governor (Associated Press, 2025). The State Department warned its embassies and consulates, and the FBI had already warned that senior US officials were being impersonated with AI voice messages (Associated Press, 2025).</p><h3>Rules for Government AI</h3><p>The Office of Management and Budget (OMB) set rules for how federal agencies use AI. Memorandum M-25–21 defines “high-impact AI” as systems whose outputs materially affect people’s rights, access to services, or safety (Perkins Coie, 2025). The memo was issued on April 3, 2025. Agencies had 365 days to apply minimum risk-management practices to high-impact systems, including testing before deployment, and must publish an updated inventory of their AI uses at least once a year (REI Systems, 2025).</p><h3>Checklist for Evaluating AI Claims</h3><p>These are the steps used to prepare this briefing, and students can apply them to any viral claim:</p><p>• Sort each statement into a verifiable fact, a prediction, or a claim about someone’s motive. Only the first can be checked.</p><p>• Trace every number to its original source and read how it was defined, for example, whether imports were subtracted.</p><p>• Check the date. Markets, laws, and export controls in this briefing changed within months.</p><p>• Look for the strongest opposing evidence, such as Cato’s argument about data-center tax exemptions.</p><p>• Note who produced each source and why. A company describing its own product, an advocacy group, and a federal agency have different incentives.</p><p><strong>Summary</strong></p><p>In this briefing, I looked at what research, official records, and established news reporting show about artificial intelligence’s effects on government, public finances, and security. It began as a fact-check of a viral interview. I only examined claims that evidence (research) supports. As of this writing, AI investment now drives a large share of US economic growth; ING’s economist puts it at “around a third of the current YoY rate of US GDP growth.” That investment is partly financed through circular deals among a few companies. The public pays indirectly through state tax exemptions and higher electricity bills. In PJM, the largest US grid, data centers caused 63% of the price increase in the 2025–2026 auction, adding $9.3 billion in capacity costs that all ratepayers must absorb. Major AI companies have begun using chatbot conversations for advertising, even as seventy-two percent of teens have used AI companions at least once. Contrary to the interview, US law blocks the Fed from issuing a CBDC, or any substantially similar digital asset, through Dec. 31, 2030. Government use of AI has delivered measurable gains in fraud prevention and cyber defense. Earlier automated systems in Michigan, Australia, and the Netherlands show the harm that follows when algorithms penalize people with little human review. AI-enabled espionage and deepfake impersonation of officials make security one of the most pressing issues for students entering the field.</p><h3>References</h3><p>AI Incident Database. (n.d.). <em>Michigan’s unemployment benefits algorithm MiDAS issued false fraud claims to thousands of people</em> (Report №2798). <a href="https://incidentdatabase.ai/reports/2798">https://incidentdatabase.ai/reports/2798</a></p><p><em>AI tools helped Treasury recover billions in fraud and improper payments</em>. (2024, October). Nextgov/FCW. <a href="https://www.nextgov.com/artificial-intelligence/2024/10/ai-tools-helped-treasury-recover-billions-fraud-and-improper-payments/400368/">https://www.nextgov.com/artificial-intelligence/2024/10/ai-tools-helped-treasury-recover-billions-fraud-and-improper-payments/400368/</a></p><p>Aldane, J. (2026, April 22). <em>Number of AI use cases in the US federal government more than doubles in a year, official figures show</em>. Global Government Forum. <a href="https://www.globalgovernmentforum.com/number-of-ai-use-cases-in-the-us-federal-government-more-than-doubles-in-a-year-official-figures-show/">https://www.globalgovernmentforum.com/number-of-ai-use-cases-in-the-us-federal-government-more-than-doubles-in-a-year-official-figures-show/</a></p><p>Amnesty International. (2021, October 25). <em>Dutch childcare benefit scandal an urgent wake-up call to ban racist algorithms</em>. <a href="https://www.amnesty.org/en/latest/news/2021/10/xenophobic-machines-dutch-child-benefit-scandal/">https://www.amnesty.org/en/latest/news/2021/10/xenophobic-machines-dutch-child-benefit-scandal/</a></p><p>Anex-Ries, Q. (2026, April 22). <em>One-year retrospective on the federal government’s implementation of updated AI guidance: Accelerating usage with incomplete safeguards</em>. Center for Democracy &amp; Technology. <a href="https://cdt.org/?p=11461">https://cdt.org/?p=11461</a></p><p>Anthropic. (2025, November). <em>Disrupting the first reported AI-orchestrated cyber espionage campaign</em> [Report]. <a href="https://assets.anthropic.com/m/ec212e6566a0d47/original/Disrupting-the-first-reported-AI-orchestrated-cyber-espionage-campaign.pdf">https://assets.anthropic.com/m/ec212e6566a0d47/original/Disrupting-the-first-reported-AI-orchestrated-cyber-espionage-campaign.pdf</a></p><p>Associated Press. (2025, July 9). <em>Impostor uses AI to impersonate Rubio and contact foreign and U.S. officials</em>. Connecticut Public. <a href="https://www.ctpublic.org/2025-07-09/impostor-uses-ai-to-impersonate-rubio-and-contact-foreign-and-u-s-officials">https://www.ctpublic.org/2025-07-09/impostor-uses-ai-to-impersonate-rubio-and-contact-foreign-and-u-s-officials</a></p><p>Associated Press. (2026a, November 5). <em>A brutal sell-off for AI stocks sends Wall Street reeling, and the S&amp;P 500 tumbles 4.4%</em>. Boston.com. <a href="https://www.boston.com/news/national-news/2026/11/05/a-brutal-sell-off-for-ai-stocks-sends-wall-street-reeling-and-the-sp-500-tumbles-4-4/">https://www.boston.com/news/national-news/2026/11/05/a-brutal-sell-off-for-ai-stocks-sends-wall-street-reeling-and-the-sp-500-tumbles-4-4/</a></p><p>Associated Press. (2026b, November 5). <em>A brutal sell-off for AI stocks sends Wall Street reeling; S&amp;P 500 sees worst day since 2025 ‘Liberation Day’ tariffs</em>. FOX Carolina. <a href="https://www.foxcarolina.com/2026/11/05/brutal-sell-off-ai-stocks-sends-wall-street-reeling-sp-500-sees-worst-day-since-2025-liberation-day-tariffs/">https://www.foxcarolina.com/2026/11/05/brutal-sell-off-ai-stocks-sends-wall-street-reeling-sp-500-sees-worst-day-since-2025-liberation-day-tariffs/</a></p><p>Brynjolfsson, E., Chandar, B., &amp; Chen, R. (2025, August). <em>Canaries in the coal mine? Six facts about the recent employment effects of artificial intelligence</em> [Working paper]. Stanford Institute for Economic Policy Research. <a href="https://siepr.stanford.edu/publications/working-paper/canaries-coal-mine-six-facts-about-recent-employment-effects-artificial">https://siepr.stanford.edu/publications/working-paper/canaries-coal-mine-six-facts-about-recent-employment-effects-artificial</a></p><p>Carr, G. (2024, August 7). <em>Case over the Michigan Unemployment Insurance Agency’s faulty automated system finally settled</em>. Science, Technology, and Public Policy Program, University of Michigan. <a href="https://stpp.fordschool.umich.edu/research/policy-brief/case-over-michigan-unemployment-insurance-agencys-faulty-automated-system">https://stpp.fordschool.umich.edu/research/policy-brief/case-over-michigan-unemployment-insurance-agencys-faulty-automated-system</a></p><p>Common Sense Media. (2025a, July 16). <em>Nearly 3 in 4 teens have used AI companions, new national survey finds</em> [Press release]. <a href="https://www.commonsensemedia.org/press-releases/nearly-3-in-4-teens-have-used-ai-companions-new-national-survey-finds">https://www.commonsensemedia.org/press-releases/nearly-3-in-4-teens-have-used-ai-companions-new-national-survey-finds</a></p><p>Common Sense Media. (2025b). <em>Talk, trust, and trade-offs: How and why teens use AI companions</em>. <a href="https://www.commonsensemedia.org/research/talk-trust-and-trade-offs-how-and-why-teens-use-ai-companions">https://www.commonsensemedia.org/research/talk-trust-and-trade-offs-how-and-why-teens-use-ai-companions</a></p><p><em>Congress sends anti-CBDC housing bill to Trump’s desk for final approval</em>. (2026, June 23). The Block. <a href="https://www.theblock.co/post/405915/anti-cbdc-housing-bill-passes-congress">https://www.theblock.co/post/405915/anti-cbdc-housing-bill-passes-congress</a></p><p>Cooley. (2026). <em>Key financial services provisions in the ROAD to Housing Act</em>. <a href="https://finsights.cooley.com/key-financial-services-provisions-in-the-road-to-housing-act/">https://finsights.cooley.com/key-financial-services-provisions-in-the-road-to-housing-act/</a></p><p>Crosley, B. (2026, February 28). <em>PJM’s $100 billion rate shock: How data centers are rewriting America’s electricity bills</em>. Introl. <a href="https://introl.com/blog/pjm-rate-shock-100-billion-data-center-electricity-2026">https://introl.com/blog/pjm-rate-shock-100-billion-data-center-electricity-2026</a></p><p><em>“Crude and cruel”: Shameful Robodebt findings laid bare</em>. (2023). Information Age. <a href="https://ia.acs.org.au/article/2023/crude-and-cruel-shameful-robodebt-findings-laid-bare.html">https://ia.acs.org.au/article/2023/crude-and-cruel-shameful-robodebt-findings-laid-bare.html</a></p><p>Defense Advanced Research Projects Agency. (2025, August 8). <em>AI Cyber Challenge marks pivotal inflection point for cyber defense</em>. <a href="https://darpa.mil/news/2025/aixcc-results">https://darpa.mil/news/2025/aixcc-results</a></p><p>Dellinger, A. J. (2025, October 13). <em>Is the AI conveyor belt of capital about to stop?</em> Gizmodo. <a href="https://gizmodo.com/is-the-ai-conveyor-belt-of-capital-about-to-stop-2000671017">https://gizmodo.com/is-the-ai-conveyor-belt-of-capital-about-to-stop-2000671017</a></p><p><em>Disclosed government AI use increased by 70% in 2025, per OMB</em>. (2026). FedScoop. <a href="https://develop.fedscoop.com/?p=89731">https://develop.fedscoop.com/?p=89731</a></p><p><em>Dutch government resigns over child benefits scandal</em>. (2021, January 16). The Guardian. <a href="https://www.pressreader.com/usa/the-guardian-usa/20210116/281711207296255">https://www.pressreader.com/usa/the-guardian-usa/20210116/281711207296255</a></p><p>Egan, M. (2024, October 18). <em>AI helped the feds catch $1 billion of fraud in one year</em>. ABC7 Chicago. <a href="https://abc7chicago.com/post/ai-helped-feds-catch-1-billion-fraud-year/15441021/">https://abc7chicago.com/post/ai-helped-feds-catch-1-billion-fraud-year/15441021/</a></p><p>Exec. Order №14178, 90 Fed. Reg. 8647 (2025). <a href="https://www.govinfo.gov/content/pkg/DCPD-202500169/html/DCPD-202500169.htm">https://www.govinfo.gov/content/pkg/DCPD-202500169/html/DCPD-202500169.htm</a></p><p><em>The flawed algorithm at the heart of Robodebt</em>. (2023, August 3). AIHub. <a href="https://aihub.org/2023/08/03/the-flawed-algorithm-at-the-heart-of-robodebt">https://aihub.org/2023/08/03/the-flawed-algorithm-at-the-heart-of-robodebt</a></p><p><em>Gross betrayal, dishonesty and collusion: Ministers ignored or dismissed Robodebt concerns</em>. (2023, July 7). InQueensland. <a href="https://inqld.com.au/news/2023/07/07/gross-betrayal-dishonesty-and-collusion-ministers-ignored-or-dismissed-robodebt-concerns">https://inqld.com.au/news/2023/07/07/gross-betrayal-dishonesty-and-collusion-ministers-ignored-or-dismissed-robodebt-concerns</a></p><p>Gülen, K. (2025, November 7). <em>Altman rejects idea of federal bailout after CFO floats backstop proposal</em>. Dataconomy. <a href="https://dataconomy.com/2025/11/07/altman-rejects-idea-of-federal-bailout-after-cfo-floats-backstop-proposal">https://dataconomy.com/2025/11/07/altman-rejects-idea-of-federal-bailout-after-cfo-floats-backstop-proposal</a></p><p>Ha, A. (2026, June 6). <em>The Trump administration might take an equity stake in OpenAI</em>. TechCrunch. <a href="https://techcrunch.com/2026/06/06/the-trump-administration-might-take-an-equity-stake-in-openai">https://techcrunch.com/2026/06/06/the-trump-administration-might-take-an-equity-stake-in-openai</a></p><p>Hardy, K. (2025, November 12). <em>Most states don’t disclose which companies get data center incentives, report finds</em>. Stateline. <a href="https://stateline.org/2025/11/12/most-states-dont-disclose-which-companies-get-data-center-incentives-report-finds/">https://stateline.org/2025/11/12/most-states-dont-disclose-which-companies-get-data-center-incentives-report-finds/</a></p><p><em>How a discriminatory algorithm wrongly accused thousands of families of fraud</em>. (2021). Vice. <a href="https://incidentdatabase.ai/reports/1404">https://incidentdatabase.ai/reports/1404</a></p><p><em>ICE just ordered $30 million worth of new technology from Palantir to track immigrants</em>. (2025, April). Business Insider. <a href="https://www.yahoo.com/news/ice-just-ordered-30-million-212445164.html">https://www.yahoo.com/news/ice-just-ordered-30-million-212445164.html</a></p><p>Implementing government-backed digital identity in the United States. (2024). <em>Jurimetrics</em>. American Bar Association. <a href="https://www.americanbar.org/groups/science_technology/resources/jurimetrics/2024-winter/implementing-government-backed-digital-identity-united-states">https://www.americanbar.org/groups/science_technology/resources/jurimetrics/2024-winter/implementing-government-backed-digital-identity-united-states</a></p><p>Jadhav, A. (2025, November 4). <em>AI’s trillion dollar deal wheel bubbling around Nvidia, OpenAI</em>. The Register. <a href="https://www.theregister.com/2025/11/04/the_circular_economy_of_ai/">https://www.theregister.com/2025/11/04/the_circular_economy_of_ai/</a></p><p>JURIST. (2023, July). <em>Australia Robodebt royal commission hands down findings, recommends referrals for prosecution</em>. <a href="https://www.jurist.org/news/2023/07/australia-robodebt-royal-commission-hands-down-findings-recommends-referrals-for-prosecution/">https://www.jurist.org/news/2023/07/australia-robodebt-royal-commission-hands-down-findings-recommends-referrals-for-prosecution/</a></p><p>Knightley, J. (2026, August 20). <em>How much is AI contributing to US economic growth?</em> ING Think. <a href="https://think.ing.com/downloads/pdf/opinion/how-much-is-ai-contributing-to-us-economic-growth">https://think.ing.com/downloads/pdf/opinion/how-much-is-ai-contributing-to-us-economic-growth</a></p><p>Kunkel, C. (2025, July 30). <em>Projected data center growth spurs PJM capacity prices by factor of 10</em>. Institute for Energy Economics and Financial Analysis. <a href="https://ieefa.org/resources/projected-data-center-growth-spurs-pjm-capacity-prices-factor-10">https://ieefa.org/resources/projected-data-center-growth-spurs-pjm-capacity-prices-factor-10</a></p><p>Meta. (2025, October 1). <em>Improving your recommendations on our apps with AI at Meta</em>. <a href="https://about.fb.com/news/2025/10/improving-your-recommendations-apps-ai-meta/">https://about.fb.com/news/2025/10/improving-your-recommendations-apps-ai-meta/</a></p><p><em>Meta to use AI to tailor ads, content to your feeds</em>. (2025). Boston 25 News. <a href="https://www.boston25news.com/news/trending/meta-use-ai-tailor-ads-content-your-feeds/ABJAUOO6YRGRLIS2KPCHNKPHSA/">https://www.boston25news.com/news/trending/meta-use-ai-tailor-ads-content-your-feeds/ABJAUOO6YRGRLIS2KPCHNKPHSA/</a></p><p>Michel, A. N. (2026, September 10). <em>Not all data center tax breaks are subsidies</em>. Cato Institute. <a href="https://www.cato.org/blog/not-all-data-center-tax-breaks-are-subsidies">https://www.cato.org/blog/not-all-data-center-tax-breaks-are-subsidies</a></p><p>Mishra, S. (2025, November 10). <em>China suspends ban on gallium, germanium, antimony exports to US</em>. Mining Technology. <a href="https://www.mining-technology.com/news/china-suspends-ban-exports-us/">https://www.mining-technology.com/news/china-suspends-ban-exports-us/</a></p><p>NVIDIA. (2025, September 22). <em>OpenAI and NVIDIA announce strategic partnership to deploy 10 gigawatts of NVIDIA systems</em> [Press release]. <a href="https://nvidianews.nvidia.com/news/openai-and-nvidia-announce-strategic-partnership-to-deploy-10gw-of-nvidia-systems">https://nvidianews.nvidia.com/news/openai-and-nvidia-announce-strategic-partnership-to-deploy-10gw-of-nvidia-systems</a></p><p>Ohio Manufacturers’ Association. (n.d.). <em>Watchdog: Utility load forecasts of unbuilt data centers costing customers $47 billion over three years</em>. <a href="https://www.ohiomfg.com/our-communities/watchdog-utility-load-forecasts-of-unbuilt-data-centers-costing-customers-47-billion-over-three-years/">https://www.ohiomfg.com/our-communities/watchdog-utility-load-forecasts-of-unbuilt-data-centers-costing-customers-47-billion-over-three-years/</a></p><p>OpenAI. (2026a, January 16). <em>Our approach to advertising and expanding access to ChatGPT</em>. <a href="https://openai.com/index/our-approach-to-advertising-and-expanding-access/">https://openai.com/index/our-approach-to-advertising-and-expanding-access/</a></p><p>OpenAI. (2026b). <em>Testing ads in ChatGPT</em>. <a href="https://openai.com/index/testing-ads-in-chatgpt/">https://openai.com/index/testing-ads-in-chatgpt/</a></p><p>Pack, A. (2026, September 4). <em>TSA accepting mobile driver’s licenses at Colorado airports</em>. Passenger Terminal Today. <a href="https://www.passengerterminaltoday.com/?p=49832">https://www.passengerterminaltoday.com/?p=49832</a></p><p>Perkins Coie. (2025). <em>White House unveils updated AI guidelines for federal agencies</em>. <a href="https://perkinscoie.com/insights/update/white-house-unveils-updated-ai-guidelines-federal-agencies">https://perkinscoie.com/insights/update/white-house-unveils-updated-ai-guidelines-federal-agencies</a></p><p>REI Systems. (2025). <em>Navigating the federal AI governance and risk management landscape</em>. <a href="https://reisystems.com/navigating-the-federal-ai-governance-risk-management-landscape">https://reisystems.com/navigating-the-federal-ai-governance-risk-management-landscape</a></p><p>Reuters. (2026, November 5). <em>Blue tsunami wipes out $2.5 trillion from Nasdaq, as investors fret about AI trade</em>. GV Wire. <a href="https://gvwire.com/2026/11/05/blue-tsunami-wipes-out-2-5-trillion-from-nasdaq-as-investors-fret-about-ai-trade/">https://gvwire.com/2026/11/05/blue-tsunami-wipes-out-2-5-trillion-from-nasdaq-as-investors-fret-about-ai-trade/</a></p><p>Rijo, L. (2025, October 25). <em>Meta plans to use AI chat data for ad targeting starting December</em>. PPC Land. <a href="https://ppc.land/meta-plans-to-use-ai-chat-data-for-ad-targeting-starting-december/">https://ppc.land/meta-plans-to-use-ai-chat-data-for-ad-targeting-starting-december/</a></p><p>RTTNews. (2025, August 22). <em>US govt. to buy nearly 10% stake in Intel</em>. Nasdaq. <a href="https://www.nasdaq.com/articles/us-govt-buy-nearly-10-stake-intel/">https://www.nasdaq.com/articles/us-govt-buy-nearly-10-stake-intel/</a></p><p>Stan, A. M. (2026, October 1). <em>Trump tells TIME the US might take stakes in OpenAI and Anthropic</em>. The Next Web. <a href="https://thenextweb.com/news/trump-time-interview-stake-openai-anthropic">https://thenextweb.com/news/trump-time-interview-stake-openai-anthropic</a></p><p>Toomer, L. (2026, September 1). <em>Coloradans can use mobile driver’s licenses to verify identity at airport security</em>. Colorado Newsline. <a href="https://www.newsfromthestates.com/node/425229">https://www.newsfromthestates.com/node/425229</a></p><p>UBS. (2025, October 10). <em>Should recent AI financing deals be a cause for concern?</em> UBS Chief Investment Office. <a href="https://www.ubs.com/global/en/wealthmanagement/insights/chief-investment-office/house-view/daily/2025/latest-10102025.html">https://www.ubs.com/global/en/wealthmanagement/insights/chief-investment-office/house-view/daily/2025/latest-10102025.html</a></p><h3><strong>About the Author</strong></h3><p>Ron McFarland, Ph.D. With a career spanning over 30 years in computer science, cybersecurity, and artificial intelligence, Dr. Ron McFarland brings unparalleled expertise to his series on stock market trading and AI. Having earned a Ph.D. in Computer Science and a postdoctoral fellowship in cybersecurity, he combines academic knowledge with decades of hands-on experience in artificial intelligence, machine learning, cloud security, database systems, and risk frameworks. Dr. McFarland has taught advanced AI and machine learning courses at the university level, equipping students and professionals to harness cutting-edge technology — a foundation that translates seamlessly to AI’s transformative role in stock trading.</p><p>Beyond his academic and consulting achievements, Dr. McFarland is a prolific writer, having authored several books and dozens of technical articles on AI, blockchain, and cybersecurity. His ability to simplify complex concepts makes his work accessible to readers of all levels, from beginners to advanced traders. As a former Dean and professor who has led programs in computer science, technical analysis, and programming.</p><p>YouTube Channel: <a href="https://www.youtube.com/@HighervistaLLC">https://www.youtube.com/@HighervistaLLC</a></p><p>Amazon Author’s Page: <a href="https://amazon.com/author/ronmcfarland">amazon.com/author/ronmcfarland</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=95cc32acfe3b" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[No Treaty, No Button: The Missing Rules for AI Weapons]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://highervista.medium.com/no-treaty-no-button-the-missing-rules-for-ai-weapons-1ca212aaa806?source=rss-1f7645bf9aea------2"><img src="https://cdn-images-1.medium.com/max/757/1*JtlVuodbLsNHYVyKdTxf9w.png" width="757"></a></p><p class="medium-feed-snippet">Who Decides When the Machine Fires? &#xB7; Part 3 of 4</p><p class="medium-feed-link"><a href="https://highervista.medium.com/no-treaty-no-button-the-missing-rules-for-ai-weapons-1ca212aaa806?source=rss-1f7645bf9aea------2">Continue reading on Medium »</a></p></div>]]></description>
            <link>https://highervista.medium.com/no-treaty-no-button-the-missing-rules-for-ai-weapons-1ca212aaa806?source=rss-1f7645bf9aea------2</link>
            <guid isPermaLink="false">https://medium.com/p/1ca212aaa806</guid>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[software-engineering]]></category>
            <category><![CDATA[software-development]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[large-language-models]]></category>
            <dc:creator><![CDATA[Ron McFarland PhD]]></dc:creator>
            <pubDate>Fri, 13 Nov 2026 18:13:43 GMT</pubDate>
            <atom:updated>2026-11-13T18:13:43.716Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[The Last Human in the Loop: AI’s Promise, Peril, and Who Profits]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://highervista.medium.com/the-last-human-in-the-loop-ais-promise-peril-and-who-profits-a8e6aea92aad?source=rss-1f7645bf9aea------2"><img src="https://cdn-images-1.medium.com/max/689/1*99SMn1TIOPZS4uZBuFP_JQ.png" width="689"></a></p><p class="medium-feed-snippet">Who Decides When the Machine Fires? &#xB7; Part 4 of 4</p><p class="medium-feed-link"><a href="https://highervista.medium.com/the-last-human-in-the-loop-ais-promise-peril-and-who-profits-a8e6aea92aad?source=rss-1f7645bf9aea------2">Continue reading on Medium »</a></p></div>]]></description>
            <link>https://highervista.medium.com/the-last-human-in-the-loop-ais-promise-peril-and-who-profits-a8e6aea92aad?source=rss-1f7645bf9aea------2</link>
            <guid isPermaLink="false">https://medium.com/p/a8e6aea92aad</guid>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[software-engineering]]></category>
            <category><![CDATA[systems-engineering]]></category>
            <category><![CDATA[software-development]]></category>
            <category><![CDATA[machine-learning]]></category>
            <dc:creator><![CDATA[Ron McFarland PhD]]></dc:creator>
            <pubDate>Fri, 13 Nov 2026 18:13:03 GMT</pubDate>
            <atom:updated>2026-11-13T18:13:03.659Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[Models That See, Models That Read: How Military AI Picks Targets and How It Fails]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://highervista.medium.com/models-that-see-models-that-read-how-military-ai-picks-targets-and-how-it-fails-cbf9c6e375b3?source=rss-1f7645bf9aea------2"><img src="https://cdn-images-1.medium.com/max/760/1*wD63Fo_0Hh3lkmMvXlmxTg.png" width="760"></a></p><p class="medium-feed-snippet">Who Decides When the Machine Fires? &#xB7; Part 2 of 4</p><p class="medium-feed-link"><a href="https://highervista.medium.com/models-that-see-models-that-read-how-military-ai-picks-targets-and-how-it-fails-cbf9c6e375b3?source=rss-1f7645bf9aea------2">Continue reading on Medium »</a></p></div>]]></description>
            <link>https://highervista.medium.com/models-that-see-models-that-read-how-military-ai-picks-targets-and-how-it-fails-cbf9c6e375b3?source=rss-1f7645bf9aea------2</link>
            <guid isPermaLink="false">https://medium.com/p/cbf9c6e375b3</guid>
            <category><![CDATA[software-development]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[software-engineering]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[large-language-models]]></category>
            <dc:creator><![CDATA[Ron McFarland PhD]]></dc:creator>
            <pubDate>Fri, 13 Nov 2026 17:54:38 GMT</pubDate>
            <atom:updated>2026-11-13T18:04:50.209Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[The Learning Loop: How AI Is Teaching Ukraine’s Drones to Kill]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://highervista.medium.com/the-learning-loop-how-ai-is-teaching-ukraines-drones-to-kill-131cf4f3d1c8?source=rss-1f7645bf9aea------2"><img src="https://cdn-images-1.medium.com/max/701/1*HTwSbO4K2ctpKT5CbunOUg.png" width="701"></a></p><p class="medium-feed-snippet">Who Decides When the Machine Fires? &#xB7; Part 1 of 4</p><p class="medium-feed-link"><a href="https://highervista.medium.com/the-learning-loop-how-ai-is-teaching-ukraines-drones-to-kill-131cf4f3d1c8?source=rss-1f7645bf9aea------2">Continue reading on Medium »</a></p></div>]]></description>
            <link>https://highervista.medium.com/the-learning-loop-how-ai-is-teaching-ukraines-drones-to-kill-131cf4f3d1c8?source=rss-1f7645bf9aea------2</link>
            <guid isPermaLink="false">https://medium.com/p/131cf4f3d1c8</guid>
            <category><![CDATA[inteligencia-artificial]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[large-language-models]]></category>
            <category><![CDATA[software-engineering]]></category>
            <category><![CDATA[software-development]]></category>
            <dc:creator><![CDATA[Ron McFarland PhD]]></dc:creator>
            <pubDate>Fri, 13 Nov 2026 17:46:00 GMT</pubDate>
            <atom:updated>2026-11-13T18:04:09.518Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[The Quiet Point of No Return]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://highervista.medium.com/the-quiet-point-of-no-return-88035c909752?source=rss-1f7645bf9aea------2"><img src="https://cdn-images-1.medium.com/max/1564/1*dgM9WqbbK73uasSgpoIjpg.png" width="1564"></a></p><p class="medium-feed-snippet">Superintelligence, runaway AI agents, and the race to keep control</p><p class="medium-feed-link"><a href="https://highervista.medium.com/the-quiet-point-of-no-return-88035c909752?source=rss-1f7645bf9aea------2">Continue reading on Medium »</a></p></div>]]></description>
            <link>https://highervista.medium.com/the-quiet-point-of-no-return-88035c909752?source=rss-1f7645bf9aea------2</link>
            <guid isPermaLink="false">https://medium.com/p/88035c909752</guid>
            <category><![CDATA[software-engineering]]></category>
            <category><![CDATA[computer-science]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[large-language-models]]></category>
            <dc:creator><![CDATA[Ron McFarland PhD]]></dc:creator>
            <pubDate>Mon, 09 Nov 2026 22:06:01 GMT</pubDate>
            <atom:updated>2026-11-13T17:28:05.768Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[The Clever AI Machine: How AI Agents Turned Free Web Tools into a Weapon]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://highervista.medium.com/the-clever-ai-machine-how-ai-agents-turned-free-web-tools-into-a-weapon-89c1e479b758?source=rss-1f7645bf9aea------2"><img src="https://cdn-images-1.medium.com/max/852/1*rua9XXTqjmNra8CDkytuMw.png" width="852"></a></p><p class="medium-feed-snippet">by Ron McFarland, Ph.D.</p><p class="medium-feed-link"><a href="https://highervista.medium.com/the-clever-ai-machine-how-ai-agents-turned-free-web-tools-into-a-weapon-89c1e479b758?source=rss-1f7645bf9aea------2">Continue reading on Medium »</a></p></div>]]></description>
            <link>https://highervista.medium.com/the-clever-ai-machine-how-ai-agents-turned-free-web-tools-into-a-weapon-89c1e479b758?source=rss-1f7645bf9aea------2</link>
            <guid isPermaLink="false">https://medium.com/p/89c1e479b758</guid>
            <category><![CDATA[large-language-models]]></category>
            <category><![CDATA[software-engineering]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[information-technology]]></category>
            <dc:creator><![CDATA[Ron McFarland PhD]]></dc:creator>
            <pubDate>Sun, 08 Nov 2026 18:49:23 GMT</pubDate>
            <atom:updated>2026-11-13T17:28:44.743Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[Inference in Production: Keeping Fast Models Fast, Reliable, and Affordable]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://highervista.medium.com/inference-in-production-keeping-fast-models-fast-reliable-and-affordable-fa4452bcb4b7?source=rss-1f7645bf9aea------2"><img src="https://cdn-images-1.medium.com/max/2600/1*iibPO7JZlDD3Yj8zL1fyVA.png" width="3200"></a></p><p class="medium-feed-snippet">By Ron McFarland, Ph.D.</p><p class="medium-feed-link"><a href="https://highervista.medium.com/inference-in-production-keeping-fast-models-fast-reliable-and-affordable-fa4452bcb4b7?source=rss-1f7645bf9aea------2">Continue reading on Medium »</a></p></div>]]></description>
            <link>https://highervista.medium.com/inference-in-production-keeping-fast-models-fast-reliable-and-affordable-fa4452bcb4b7?source=rss-1f7645bf9aea------2</link>
            <guid isPermaLink="false">https://medium.com/p/fa4452bcb4b7</guid>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[information-technology]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[software-engineering]]></category>
            <category><![CDATA[large-language-models]]></category>
            <dc:creator><![CDATA[Ron McFarland PhD]]></dc:creator>
            <pubDate>Wed, 04 Nov 2026 21:26:49 GMT</pubDate>
            <atom:updated>2026-11-13T17:29:31.622Z</atom:updated>
        </item>
    </channel>
</rss>