Today something a little bit different. I’m a big fan of journalists who cover the AV sector. I’ve been watching the smart car sector for what feels like decades. From Apple’s failure to build a smart car to Xiaomi’s success, it’s been quite a journey. In the later half of the 2020s, there’s finally light at the other end of the tunnel. Advanced autonomy, improved sensors, millions of miles more experience, integrated software and embedded AI have all improved considerably in the last decade.

That world of robots, delivery drones, robo-taxis and consumer ambient intelligence: well, it’s finally coming.

A Physical AI Wave Begins in 2027

I’m hugely optimistic about the coming Physical AI wave. When Rivian had their AI & autonomy day back in December, 2025 I had the distinct impression this is what the entire world is heading towards. The physical AI devices of 2027, emphasis on humanoid robotics, and the evolution of autonomous vehicles finally will appear tangible by the end of the decade.

Rivian’s CEO also recently appears on the No Priors Podcast, a fairly good AI startup podcast. It’s been a long road for Rivian but I see them turning the corner this year where the company has moved from being a niche, high-end truck maker to a serious mass-market contender, primarily driven by the launch of the R2 platform. I’ve always liked the look of their vehicles but finally they are more affordable too, the Rivian R2 is a new, more affordable midsize electric SUV built on a dedicated, highly efficient, and compact platform, starting around $45,000.

But it’s the AI & Autonomous work they are doing that is the deal maker for me.

Video: The Leap | Autonomy & AI Day | Rivian - YouTube.

Autonomous Vehicles are Closer to Mainstream Adoption

Rivian is one of the most promising EV startups in the U.S., at a major turning point this year in 2026 as they pivot more to AI and autonomy, which I find fascinating. They are backed by the likes of Amazon and Volkswagen Group and I believe will become a major alternative to Tesla even as the demand for electric vehicles has waned. I more or less see 2027 as the date civilization enters the age of Physical AI and I think Rivian represents an interesting case study in its emergence.

As Waymo, the leading AV moonshot by Google expands, it’s an exciting time in autonomous vehicle evolution.

I watched the entire Rivian AI and Autonomy Day live, and I was impressed. I asked of the Newsletter The AV Market Strategic, for his insights on Rivian and a deep dive on their AI & Autonomy strategy.

The AV Market Strategist

If you are interested in smart car developments, this is a great publication. Autonomous vehicles are improving and expanding to more cities near you.

It’s, finally happening.

The AV Market Strategist

The AV Market Strategist breaks down strategic developments, market dynamics, and competitive moves in the autonomous vehicle industry to reveal where the market is truly heading.

By Daniel Abreu Marques

• Wayve Raises $1.5B, Waymo Now In 10 Cities And Uber Builds the AV Platform Layer

• Waymo’s Shocking Data Release, Uber Invest $100M in Robotaxi Infrastructure

• Tesla All-In Autonomy: 7 New Robotaxi Cities, $20b CAPEX, $2b Investment in xAI

Learn More

In March, 2026 we’re going to learn a lot more about Rivian’s R2 in just a couple of weeks. Spring, 2026 is going to be their coming out party.

Daniel’s deep dive is a serious look into Rivian and their prospects.

Rivian’s AI & Autonomy Highlights ✨

Read more

Good Morning,

How three centuries of booms were built on credit, and how they break. This is part II in the series on: Bubbles of the past and how they compare to the AI bubble of today.

Listen 18 minutes

• Read Part I here. How the AI ‘bubble’ compares to history 🏛️

• Read the article that started it all here.

• Today’s article is part II in the series.

The Context of Bubbles and the AI Tech Revolution

“We tend to frame historical bubbles as IQ tests because that lets us feel smarter in hindsight. Everyone was delusional. But the historical record is more complicated than that. Many episodes are misremembered. The South Sea Bubble, for example, wasn’t really about South America. It was a retail-facing financial instrument that allowed the public to participate in a large-scale conversion of British government debt into equity.” - Howe Wang.

The Historical Lens of Recurring Cycles of Investment in Mass Infrastructure

“The Future on Margin series lays out a recurring architecture. First, there is a legible future, something the public can see and imagine: railroads crossing continents, homes lit by electricity, the internet entering daily life. Second, there is a financing regime that pulls that future forward: installments, margin, holding-company pyramids, junk bonds, vendor credit. Third, there is mass access. Those financing tools are made available to the general public early, with low thresholds that turn belief into position size. Part I walked us through Mississippi, South Sea, frontier land, and railway manias. Part II looks at utilities, RCA, and the dot-com credit machine.”

writes the Procure.FYI Substack, which explores enterprise AI adoption, infrastructure buildouts, and the financing mechanics behind global technology trends as well as running a consultancy.

Works

In recent times he’s been closely following the AI Infrastructure, semiconductor industry and datacenter roll-out. Previously, Howe worked on ML and data teams at Lyft, and earlier as an investment analyst and a global infrastructure economist.

• From $8M to $21B: How Geopolitics and Lip-Bu Tan’s Bet on Data Connectivity Created an AI Titan

• An Overview of a Global and Dispersed Networking Ecosystem

• Mellanox, NVIDIA, and the GPU Era - A Networking Story Spanning China, the US, and Israel

• The Rise Nuclear and Rise of Small Modular Nuclear Reactors (SMRs)

Follow Michael Burry for more contrarian context on the AI Bubble

Michael Burry, the investor who shot to fame for calling the housing crash before 2008, has launched a Substack newsletter after deregistering his hedge fund, aiming to lay out in detail his increasingly bearish thesis on artificial intelligence.

Michael Burry’s position on the AI bubble is super fascinating (see bonus section at the end):

• Burry has consistently compared the current AI enthusiasm to historical manias, calling OpenAI the “Netscape of our time” (referring to the dot-com pioneer whose hype preceded the 2000 crash). Burry emphasizes return on invested capital (ROIC) as the key metric for evaluating AI companies, arguing that massive capital expenditures (especially on hardware and infrastructure) have not yet translated into meaningful profit returns, pushing Big Tech toward less efficient, capital-heavy business models.

• Burry has further warned that hyperscalers and AI firms are understating depreciation on hardware (e.g., chips with shorter useful lives due to rapid obsolescence), potentially overstating earnings by hundreds of billions across the industry from 2026–2028. He predicts this could lead to a drawn-out market decline, especially given passive index funds’ heavy exposure to AI names.

“….During the Industrial Revolution and the Services Revolution, the impacts on labor were so great that mandatory schooling was instituted and expanded to keep young people out of the labor pool for longer. We certainly have not seen anything like that (Yet as of 2026).” - Michael Burry.

Let’s now continue part II of our historical journey of the future on margins.

Read Part I, First

The Future on Margin (Part II)

How three centuries of booms were built on credit, and how they break

Part II: Utilities, RCA, and the Dot-Com Credit Machine (1924 - 2002)

See Part I Here

By the 1920s, the instruments were more sophisticated, holding companies, margin accounts, a central bank managing credit, but the recipe was unchanged: a future you can picture, plus a balance sheet that makes it cheap to own that future today.

V. Pyramids of Control: Utility Holding Companies, 1924 - 1932

Everyone could see where it was going - every home lit, every factory humming - but the path there stretched across decades. Electrification was a diffusion process, driven by relentless engineering logic: build larger networks, increase utilization, lower unit costs.

Chicago offers a clean early case because it also provides a central figure. Samuel Insull arrived there in 1892 after rising inside Thomas Edison’s circle and absorbing a crucial lesson: electricity only works economically as a system. His operating model emphasized large central stations, interconnection, and steady load rather than fragmented local plants. The effect was dramatic. Delivered prices in Insull’s system fell from roughly $0.20 per kilowatt-hour in 1892 to about $0.025 by 1909: scale economics at work. National adoption followed, but slowly. U.S. generating capacity nearly doubled from about 22 million kilowatts in 1922 to 43 million kilowatts by 1930, yet penetration remained uneven. As late as 1935, only about 10% of American farms were electrified. The technology was spreading, but the country was still early in the physical rollout.

Exhibit 12: Annual Monthly Mean, Household Electricity Consumption, 1921-1950 Chart showing steady growth in household electricity consumption from near zero in 1921 to ~160 kwh/month by 1950, with both U.S. national average (solid line) and Riverside, California data (dotted line)

Riverside Consumption Levels, 1921-1950 Detailed consumption chart showing how new appliances drove electricity usage: “Lights & Iron” (~20 kwh/mo), Radio transition period (~57 kwh/mo by 1929), Refrigerator adoption (~101 kwh/mo by 1938), and War & power transition reaching ~148 kwh/mo by 1946.

• Source: Ronald C. Tobey, Technology as Freedom: The New Deal and the Electrical Modernization of the American Home (Berkeley: University of California Press, 1996), Figures 6.1 and 6.2. Data from U.S. Bureau of the Census, Historical Statistics of the United States (1976), 1:827, Series 108 (household electricity consumption) and Series 109 (percentage of dwelling units with electric service).

The 1920s compressed that long engineering curve into a short capital-markets event. Electrification was visibly advancing, utilities were consolidating, and the sector looked unusually stable compared to more cyclical industries. What turned that stability into a boom was not optimism alone, but financial engineering.

Wall Street’s instrument was the public-utility holding company. Instead of owning a regulated operating utility, the entity that owned the wires, billed customers, and answered to state commissions, investors bought equity in companies that owned utilities. Those holding companies could issue bonds, preferred stock, and common stock to acquire additional utility securities, which in turn supported further issuance. Layered repeatedly, this produced pyramiding: control flowed upward through thin equity slices, while leverage accumulated throughout the structure. Operating utilities were regulated; holding companies often were not, exploiting gaps between state oversight and interstate finance.

Exhibit 13:Utilities and Other Sector CAPE Ratio Chart (1920s-1930s) Chart showing cyclically-adjusted price-to-earnings ratio reaching record level above 60 in Q3 1929, followed by 75%+ decline through Q2 1932. Demonstrates extreme valuation disconnection from fundamentals.

• Source: NBER Working Paper 20370. https://www.nber.org/system/files/working_papers/w20370/w20370.pdf

Insull became the emblem of this architecture not because he invented it, but because of its scale. By 1930, one contemporary account described roughly $27 million of capital investment at the top controlling more than $500 million of underlying assets. A small amount of equity exercised control over a vast asset base financed by multiple layers of debt and preferred claims. The arrangement appeared resilient as long as refinancing was easy and security prices held.

Valuations reflect how far this logic was carried. The Utilities sector cyclically adjusted price-to-earnings ratio (CAPE), which smooths earnings over a 10-year real average, reached a record level just above 60 in Q3 1929. Barron’s warned that summer about unrestrained speculation in utilities and an accelerating merger wave that far outpaced underlying economics.

When the system unwound, it did so through financing rather than technology. After the 1929 crash, the market for new issues weakened and credit conditions tightened, cutting off the refinancing channel that holding-company structures quietly depended on. Insull’s control rested heavily on pledged stock, so falling prices immediately translated into declining collateral values. That decline triggered margin pressure, forced sales, and challenges to control itself. At that point, the market price of utility securities stopped functioning as a measure of optimism and became an operational constraint: it determined whether debt could be rolled, obligations met, and control maintained.

Exhibit 14: TIME Magazine Cover - Samuel Insull TIME Magazine cover featuring Samuel Insull dated May 14, 1934, after his utility empire’s collapse and subsequent indictment. Captures the transformation from “utility king” to cautionary tale.

• Source: TIME Magazine, May 14, 1934. http://content.time.com/time/covers/0,16641,19340514,00.html

Exhibit 15: Source: American Business History, “From Hero to Hated: America’s Most Tragic Entrepreneur.” https://americanbusinesshistory.org/from-hero-to-hated-americas-most-tragic-entrepreneur/

Once that mechanism flipped, the unwind was nonlinear. Even though underlying electricity demand remained relatively stable, the pyramided capital structure could not survive without continuous access to capital markets. Between Q3 1929 and Q2 1932, the utilities sector lost more than 75% of its real total return. Insull’s empire entered receivership in 1932, wiping out hundreds of thousands of shareholders and bondholders. Insull was pushed out of management and left the U.S. He was indicted on fraud/embezzlement-related charges, then fought extradition. The political response, the Public Utility Holding Company Act of 1935, codified the lesson by dismantling multi-tier holding companies and re-anchoring the sector to simpler, geographically constrained, regulated balance sheets.

VI. Buying the Tollbooth: Radio and RCA - United States, 1927 - 1932

Radio is another clean example of how a real transformation can still produce a classic, levered equity blow-off. The technology was genuinely new, adoption was compounding fast, and the buildout required real capital. Factories had to be expanded, parts supply chains had to be built, and thousands of households adopted radio equipment as utility. Radio advertising was tiny in the late 1920s, then rose fast: $4.8M (1927), $14.1M (1928), $26.8M (1929), $40.5M (1930), $56.0M (1931), $61.9M (1932), and it kept growing into the late 1930s.

RCA became the stock that concentrated that whole story. It was not just “a radio company.” It sat on the industry’s choke points: patents, licensing, and the network economics of broadcasting. Investors love a gatekeeper because it makes the end state feel simpler. You do not have to pick the winning set manufacturer or the winning station. You buy the company that looks like the tollbooth.

Exhibit 16: Radio Equipment Sales Growth Chart (1922-1929) Chart showing explosive growth in radio equipment sales from $60M (1922) to $843M (1929), demonstrating the genuine technological adoption underlying the stock speculation.

• Source: Economic History Association, “The History of the Radio Industry in the United States to 1940.” https://eh.net/encyclopedia/the-history-of-the-radio-industry-in-the-united-states-to-1940/

Exhibit 17: Radio Advertising Revenue Growth (1927-1932) Data showing radio advertising revenue climbing from $4.8M (1927) to $61.9M (1932), even as RCA stock collapsed—illustrating the divergence between technology adoption and equity prices.

• Source: Economic History Association, “The History of the Radio Industry in the United States to 1940.” https://eh.net/encyclopedia/the-history-of-the-radio-industry-in-the-united-states-to-1940/

Then came the credit pathway. The late-1920s stock market made it easy to own growth with borrowed money. The mechanism was margin and broker loans. Investors could buy shares with a partial cash down payment and borrow the rest through brokers, funded by a swelling pool of call loans and other short-term credit. In the run-up, broker loans tied to carrying securities rose sharply, from roughly $4.4B (Jan 1928) to around $8.5B (Oct 1929). That structure does two things. It amplifies the upside on the way up, and it turns a price decline into forced selling on the way down.

Exhibit 18: RCA Stock Price Chart (1922-1932) RCA’s dramatic rise from ~$5 (1922) to $113 (1929 peak) and subsequent 98% collapse to $2.50 (1932), showing classic boom-bust pattern in a “gatekeeper” technology stock.

• Source: Bryan Taylor, “RCA and the Roaring Twenties,” Finaeon. https://finaeon.com/rca-and-the-roaring-twenties/

Exhibit 19: RCA Advertisement (April 1, 1926) Vintage RCA advertisement showing the marketing of radio technology to consumers during the adoption phase. Illustrates how real products underpinned the speculative bubble.

• Source: Saturday Evening Post, “RCA’s Home Entertainment Center” (2017 retrospective). https://www.saturdayeveningpost.com/2017/06/rcas-home-entertainment-center/

In 1929, RCA, at its valuation peak, bought Victor Talking Machine, a manufacturing and distribution empire in consumer audio. The acquisition itself is a normal strategic move in a fast-growing industry. The telling part is how naturally equity becomes currency. RCA reclassified 1,500,000 common shares into 7,500,000 shares, and then issued large blocks of stock in connection with the deal. A stock split and a bigger share count do not change fundamentals overnight. They make it easier to distribute, easier to finance, and easier to use as acquisition money. In boom markets, that happens, a lot.

Then comes the regime change. Once stock prices start falling, collateral falls with them. Brokers and lenders pull back. Margin turns from accelerant into trap. RCA’s stock, split adjusted, traded around $5 in 1922, hit $113 at a point in 1929, and was back around $2.5 in 1932. Even after the crash, parts of the real radio business kept growing. Advertising dollars rose into the early 1930s. But the stock had already done its job in the market cycle. The trade was priced for perfection and financed for momentum. Once the financing window narrowed, the market stopped paying “end state” prices for medium-term progress.

VII. Credit Disguised as Demand: The Internet Boom - United States, 1995 - 2002

Exhibit 20: Capital Expenditures on Telecommunications Equipment in the United States (1996-2006) US telecom equipment capital expenditures rising from ~$40B (1996) to peak of ~$87B (2000), then collapsing to ~$37B (2002) before stabilizing at ~$40B (2006). Lucent’s SNO revenues tracked this pattern, climbing from ~$9B (1996) to ~$25B (2000), then plummeting to ~$7B (2003-2006).

• Source: William Lazonick and Edward March, “The Rise and Demise of Lucent Technologies,” Business and Economic History On-Line, Vol. 9 (2011), p. 35. Data from US Census Bureau, 1996-2006; Lucent Technologies 10-K filings, 1996-2006. http://www.thebhc.org/publications/BEHonline/2011/lazonickandmarch.pdf

By late 1999, the internet trade had crossed the threshold that matters in every mania: the end state became legible. You could picture it. Shopping moved online. Advertising followed eyeballs. Every company needed a website, then e-commerce, then “infrastructure.” The promise was visible in daily life, and easily turned into a valuation boom. But the boom did not run on imagination alone. It ran on a credit pathway that let investors and companies pull future demand into the present.

Exhibit 21: Broker-Dealer Receivables from Customers - Margin Loans and Other Receivables (1997-2004) Chart showing broker-dealer margin loans and customer receivables rising from ~$110B in Q2 1997 to peak of $263B in Q1 2000 (coinciding with NASDAQ’s March 10, 2000 peak at 5,048)

• Source: Federal Reserve, Flow of Funds Accounts of the United States, Table D.2 (Broker-Dealers).

On the market side, the mechanism was leverage against rising prices. NYSE margin debt climbed from roughly $100 billion in 1996 to $278 billion at the March 2000 peak—nearly tripled. Investors controlled large positions with small cash down, brokers financed the rest. And the IPO market functioned as a credit window with a stock-exchange wrapper—public capital for companies that couldn’t have borrowed a dime, available until the day it wasn’t. In 1999 alone, 477 companies went public, raising $69 billion, many at valuations that assumed a decade of growth. When the window closed in 2000, it closed fast: 2001 saw fewer than 80 deals. On the real-economy side, the buildout was capital-intensive and funded like a boom. Telecom capex grew at a 23% compound annual rate from 1995 to 2000, fueled by high-yield debt. Telecom companies—Global Crossing, 360networks, Williams—loaded up on junk bonds against incomplete buildouts, betting the new-issue window would stay open long enough to reach breakeven. It didn’t. When credit spreads widened in 2000, forty-seven telecom startups went bankrupt within three years. The technology still worked. The financing didn’t.

Exhibit 22: Barron’s “Burning Up” Cover (March 20, 2000) Iconic Barron’s magazine cover published 10 days after NASDAQ peaked at 5,048, featuring flames consuming money and headline “Burning Up; Warning: Internet Companies are Running Out of Cash—Fast.” One of history’s most prescient market-top calls.

• Source: Barron’s, March 20, 2000. https://elinversordebolsillo.com.ar/en/burning-up/

Then there was vendor financing. Lucent opened a $2 billion credit line for WinStar in 1998, booking revenue while quietly becoming WinStar’s lender. By fiscal 2000, Lucent had committed up to $8.1 billion in customer credit and guarantees. Nortel had extended $3.1 billion. Cisco, $2.4 billion. The equipment makers had become shadow banks. When Lucent refused WinStar an additional $90 million in early 2001, WinStar went bankrupt and Lucent wrote off $700 million. Multiply that across the industry. At the top, equity became acquisition currency. Cisco did 23 deals in 1999 - 2000, almost all in stock. JDS Uniphase bought SDL for $41 billion in shares at the absolute peak, then wrote off $50 billion within a year.

So what broke in 2000 - 2002 wasn’t one earnings miss or one fraud. It was the financing stack - high-yield debt, vendor credit, the IPO window, and stock-as-currency - that had underwritten the pace of the buildout and the willingness to price distant demand as near-term reality. The proximate trigger was the Fed: Greenspan raised rates six times between June 1999 and May 2000, from 4.75% to 6.5%. But rate hikes alone don’t collapse a system. What collapses a system is a structure that requires perpetual confidence. Stock prices fall, so collateral values fall, so margin calls force selling, so prices fall further. Credit ratings get cut, so debt markets close, so companies issue dilutive equity at the worst moment. One borrower defaults, so lenders reprice the whole sector, so spreads widen, so more borrowers default. The loop is reflexive: one wobble in a tight enough structure and the whole thing starts to unwind. Once that architecture cracked, projects got canceled, inventories piled up, and capex that had compounded at 23% annually fell at 28% for three years. The internet kept spreading. But the market stopped paying end-state multiples for medium-term progress, and stopped extending the credit that made the carry trade feel effortless.

Same Architecture, Different Century

Exhibit 23: Stylized Adoption vs Valuation Curve

Every boom in this essay follows the same sequence. A technology arrives whose end state is easy to picture: a continent connected by rail, every home electrified, commerce rewired by the internet. That clarity is the first ingredient. The second is a financing regime that lets you own the future now. When a vivid future meets cheap credit, markets don’t wait for a 40-year diffusion curve. They compress it into three to five years of price action.

Early on, the fundamentals feel decisive: adoption metrics, earnings revisions, technical milestones. The market is building a story, and the evidence keeps arriving on schedule. But three or four years in, once the financing stack is fully extended, the question quietly changes. The market stops asking, “Is this working?” and starts asking, “Can this still be financed?” At that point, everyone with exposure becomes a salesman, because the structure needs belief to remain liquid.

Then sentiment cracks. It rarely takes much: a rate hike, a missed quarter, a magazine cover that does the math, a single default that forces lenders to reprice an entire sector. The trigger can be small: credit tightens, refinancing stops, and the bridge to the future can’t be funded anymore. The technology keeps working, but the market stops financing the optimism.

The people who underwrote railroad bonds and bought RCA at the top were usually right about the direction of history. The future they believed in arrived. It just didn’t arrive on the timeline their leverage demanded. So the question in any boom isn’t whether the technology is transformational, it usually is. The question is whether you’re still in the phase where that matters.

Return to Part I

Michael Burry related Bonus Section🌊

Read more

This article is about one of the most accomplished AI authors I know on Substack. Promoting global voices and emerging writers is a privilege. 🙏

Sangeet Choudary is the best-selling co-author of Platform Revolution and the author of the new book Reshuffle. He has advised CEOs at more than 40 Fortune 500 companies as well as pre-IPO tech firms. It’s certainly a great time for awesome AI books.

According to :

Newsletter: Platforms, AI, and the Economics of BigTech

• If you enjoy this piece, please subscribe to his newsletter:

Platforms, AI, and the Economics of BigTech

Your guide to strategy and policy in the age of platform ecosystems, BigTech dominance, and AI.

By Sangeet Paul Choudary

The Newsletter is a great guide to strategy, policy and discovery in the age of platform systems, BigTech’s dominance, and AI.

🔮 Get to know the Author and about the New Book - 2 min :52

Trending Articles:

1. Figma - The untold story

2. Chegg, ChatGPT, and the changing nature of change

3. Tesla, SpaceX, and the economics of Stack Thinking

4. Don't sell shovels, sell treasure maps

Pledge your Support to SPC

"This feels like a generational book, not a mere bestseller." -

Prashant Kohl

• What if we’ve misunderstood the real power of AI, not as a tool for doing tasks faster, but as the missing mechanism for making complex systems finally work together? In Reshuffle, Sangeet Paul Choudary reframes one of the most hyped technologies of our time not as a smarter brain, but as a better glue.

Get the new Book

Reviews of Reshuffle: Who Wins When AI Restacks the Knowledge Economy by Sangeet Paul Choudary highlight its insightful and systems-level perspective on AI’s transformative impact on the knowledge economy. Below is a summary of key themes and sentiments from available reviews:

• Praise for Systems Thinking and Clarity: Reviewers commend the book for moving beyond simplistic views of AI as a tool for task automation.

• Practical and Strategic Insights: The book is described as a "lucid roadmap" for navigating technological disruption, particularly for executives and leaders.

• Historical Analogies and Broader Perspective: The use of historical parallels, like the shipping container revolution or the barcode’s impact on retail, is frequently highlighted as a strength.

• Reframing AI’s Role: The book is lauded for shifting the conversation from AI as a smarter tool to AI as a driver of new economic logic.

Sangeet is currently based in Dubai, UAE, read some of his recent posts on LinkedIn. Sangeet Paul Choudary is a business executive, advisor, and best-selling author. His ratings on Amazon and Goodreads are very high. I note his originality and layered insights.

Who else to read in AI book authors?

• Amazon Best-Sellers in AI and Machine Learning currently feature books by , , , and many others. Also by Parmy Olson:

• Sangeet is currently a Senior Fellow at the University of California, Berkeley, and a Scholar at Dartmouth College. The book, Reshuffle: Who wins when AI restacks the knowledge economy.

Why AI experts get their predictions on jobs so wrong

By , August, 2025.

The task-based fallacy that fools today’s top AI experts

This article is based on the book Reshuffle and contains extracts from the book.

British computer scientist Geoffrey Hinton, often dubbed the “Godfather of AI” famously predicted in 2016 that AI would soon replace radiologists. Much of the job involves interpreting medical images, a task that lends itself well to machine learning. Algorithms, trained on vast image datasets, can detect patterns of disease with remarkable precision. Yet, nearly a decade down, radiology hasn’t disappeared; it has flourished. What’s striking is that at most tasks in radiology, AI already outperforms humans. So why are radiologists still in demand?

Hinton’s prediction about radiologists might have missed the mark, but it stands as a larger lesson: expert pronouncements on AI’s future often oversimplify the complex dynamics of work, reducing roles to collections of tasks that machines can supposedly supplant. And that trend hasn’t slowed. Dario Amodei, Sam Altman, Eric Schmidt, and others have all made similar pronouncements about other jobs.

Some of this can be attributed to hype, but even the most well-meaning of pronouncements - sometimes by researchers who have little to gain financially from such statements - also fall into the same trap.

And that is the oversimplistic explanation that jobs are essentially just sets of tasks.

The task-based fallacy

We often think of jobs in terms of the tasks people perform, not least because the roles we’re familiar with - teacher, designer, plumber - are themselves descriptive of the tasks they are associated with.

Such a task-based view of work often leads us to ask which tasks will lose value once AI can perform them more efficiently and cost-effectively. In response, we try to shift toward other tasks that AI hasn’t yet mastered. However, this mindset puts us in a constant race against the machine, always chasing the next task that AI can’t take over.

Our jobs, however, don’t exist solely around the performance of certain tasks. Jobs, instead, are part of larger systems - workflows, organizations, industries - and are structured to resolve constraints in those systems. When that constraint collapses, the logic for the role no longer holds.

The typist’s dilemma

People often assume their jobs are safe because the tasks involved seem resistant to automation. But what if that’s not enough? If we look beyond individual tasks and consider the broader system that makes a role necessary, we find that a role can vanish, even while the tasks it once involved continue.

Consider what happened to typists in the late twentieth century. Before the advent of word processors, workflows were designed around typists. Entire teams coordinated with typing pools. Errors were expensive and required retyping entire documents. Formatting had to be done within the physical limitations of the typewriter. The typist’s role was structured around managing these constraints.

Then the word processor arrived.

At first glance, it seemed like a productivity tool that would speed up the work of typists. But word processors collapsed the costs that had once justified the typist role. The cost of fixing errors dropped to zero. Revisions could be made instantly, and formatting could be automated.

The task of typing didn’t disappear - humans were still required to perform it, in fact, more than ever before - but the job of the typist no longer made sense. The role existed not to perform the task but to manage the constraint created by the high cost of managing errors and revisions. Once word processors eliminated those costs, the job bundle lost its economic rationale. Typing became a distributed task, absorbed into the broader workflows of knowledge work. Typists weren’t outperformed by better typists equipped with new tools; they were displaced by a system in which typing no longer required a dedicated role.

The task-centric view sees AI as a tool that improves how individual tasks are performed. Work remains structurally unchanged. AI is simply layered on top to improve speed or lower costs. In this framing, the main risk is that a smarter tool might replace the person doing the task. The proposed solution is straightforward: people should learn to use the tool to stay relevant and keep their jobs.

The system-centric view, on the other hand, looks at how AI reshapes the organization of work itself. It focuses on how tasks fit into broader workflows and how their value is determined by the logic of the overall system. In this view, even if tasks persist, the rationale for grouping them into a particular job, or even performing them within the company, may no longer hold once AI changes the system’s structure.

A crash-course on AI by Steph Curry

To truly understand how AI affects jobs, we must look beyond individual tasks to comprehend AI’s impact on our workflows and organizations. And to understand how AI changes organizations, we need to understand how AI changes the very basis of competition for companies - and, with that, the reason those organizations exist in the first place.

To do that, we need to shift our lens from focusing on whether AI can perform a task to understanding how AI changes the entire system that gives that task its purpose and value.

The task-centric view treats jobs as stable bundles of tasks, which evolve as the constituent tasks are gradually replaced or enhanced over time. Automation and augmentation are seen as the only two effects.

In contrast, the system-centric view doesn’t see jobs as fixed roles. It sees them as temporary groupings of tasks that make sense in a specific system. When AI changes the system, new roles emerge in response to a new system. The impact of AI on jobs is, then, determined not just by AI’s impact on a job’s constituent tasks, but on AI’s impact on how those tasks are organized to help firms compete more effectively.

Top 5 pieces from my newsletter

⚗️ The many fallacies of ‘AI won’t take your job but someone using AI will’: The top article on Sangeet’s substack with more than 100,000 reads and nearly a thousand likes. Highly recommend!

😂 How to intellectually debate AI while completely missing the point: A central fallacy that Sangeet tackles in his book Reshuffle to dismiss both AI boomers and doomers.

💡 Humans as luxury goods in the age of AI: A powerfully argued piece on the human advantage in the age of AI.

😲 You think you are AI-first, but you probably aren’t: This piece completely flips how we think about building an AI-first startup.

🦄 When answers get cheap, good questions are the new scarcity: Where does new advantage lie in an age of abundant LLM-generated answers.

• The article continues below and uncovers how AI reshuffles everything!

Read more

Here are some things I’m watching and some AI news rundown vibes:

• Grok 4 Voice mode is a substantial upgrade making voice AI realistic for mass adoption in the 2025 to 2030 period.

• Moonshot AI’s Kimi K2 marks the point and first time where an open-weight Agentic model surpasses closed-model proprietary models.

• Anthropic’s MCP is reaching mass adoption and along with Claude Code appears to be pivotal for the future of Agentic AI.

• That cool Japanese AI startup Sakana AI introduces a method to teach reasoning to LLMs with small RL-trained teacher models.

• Tencent unveils open-source LLM plus new benchmarks for agent evaluation.

• Genspark $10M run rate in 9 days with an AI agent, after its pivot. They bill themselves now as an all-in-one Agentic workspace where they recently released AI pods.

• ElevenLabs unveils 11.ai voice assistant. It enables you to create tasks and summarizes your schedule.

• Baidu has recently unveiled its open-source multimodal family.

Lately I’ve been thinking about the future of work with AI agents as they will get better rapidly we presume in the months and years ahead.

🎧 AI Spectrum Tidbits 💡 (AI news sound bytes)

I had the pleasure of speaking with of the Newsletter 2nd Order Thinkers. It’s a relatively new Newsletter:

2nd Order Thinkers 🦄

2nd Order Thinkers

Write for smart, curious minds like you who are allergic to hype. You are keen on deep thought and refuse to be spoon-fed easy answers.

By Jing Hu

• Jing Hu dissect AI’s real effects on society, ethics, and business in 2nd Order Thinkers. 🚨

She is an AI journalist with a scientific and corporate background with a background in product. Featured in Forbes: The Human Costs of AI Transformation. She deals with many topics related to the impact of AI on human systems.

2nd Order Thinkers - Articles

• I Studied 200 Years’ Of Tech Cycles. This Is How They Relate To AI Hype.

• AI Makes Us All Speak American?!

• Is Apple the Only One Getting AI Right!?

• Claude 4 Artificially Conscious But Not Intelligent?

• OpenAI's Creativity Heist and What Survives

Her writing is easy to read but her topic choices are fairly profound and sometimes contrarian which I like. According to a report by Forrester Generative AI and via AI agents will replace approximately 2.4 million job positions in the United States by the year 2030. That seems like a reasonable number to me considering the pace of technological change in 2025. Curiously, the firm’s 2023 Generative AI Jobs Impact Forecast [paywall] predicts that the tech will reshape more jobs than it replaces.

I asked Jing for her honest and researched take on the impact of AI agents on the future of jobs.

Agentic News Tidbits

• Manus launches an agentic browser that saves login state across agent sessions. LinkedIn to a major partnership with Alibaba, Manus has moved operations it seems to Singapore.

• Anthropic launches Project Vend: Claude runs a real-world shop for a month (e.g. will Ecommerce stores be run by AI agents soon? 🤔)

• Hugging Face releases SmolLM3 - smol, multilingual, long-context reasoner.

  

• Perplexity launches its AI-powered web browser Comet.

• Goldman Sachs is piloting its first autonomous coder in what is being called by CNBC “a major AI milestone for Wall Street.” It turns out pesky Cognition’s AI coding agent Devin has scored a major customer! Devin on Wall Street? Strange🤯.

• AWS AI Agent Marketplace Launch: Amazon Web Services (AWS) is set to launch an AI Agent Marketplace on July 15, 2025, at the AWS Summit in New York City, with Anthropic as a key partner. This will likely strengthen the partnership between Amazon and Anthropic whereby MCP and Claude Code have made enormous progress in the first half of the year. Amazon is thus likely to fund Anthropic more soon.

So how close are we to AI agents replacing more and more of our tasks? What is the probability and what are the capabilities these systems lack? We are supposedly at State 3 of OpenAI’s hierarchy of AGI. Would you trust an agent to take actions on your behalf in 2025?

In 2025, it appears like OpenAI has slipped behind the likes of Anthropic and even Google in a product timeline around Agentic AI systems. Meanwhile we are all anticipating both DeepSeek-R2 and GPT-5 to understand this better. I expect them both in the coming weeks. Even as Meta frantically builds their Meta Superintelligence Lab.

Which jobs might be most susceptible?

• Let’s think about this more deeply, how will AI agents impact jobs and the future of work?

Support Emerging Indie Journalists

Will AI Agents really Automate Jobs at Scale?

By , July, 2025.

AI is coming for your job.

At least, that's what a lot of big-shot CEOs and pundits keep insisting.

Read more

You may have noticed AI Supremacy (Sitemap 2025) hasn’t covered Agentic AI much, I am still waiting for consensus on frameworks, protocols and leading hubs to manifest. However I think we can start to dip our toes into the topic.

Will Agentic AI alter how we think of business models, defensive moats and AI vs. SaaS growth norms? Will it alter the very nature of entrepreneurship and innovation? That remains to be seen.

of Decoding Discontinuity Newsletter has written a very interesting series on Agentic AI. I wanted to bring some attention to it and asked her for a macro piece on the topic. That’s the topic of our deep dive today. We’ll be lowering our frequency of posts during the July-August period.

1. Agentic Era Part 1. A Strategic Inflection Point Where Orchestration and Distribution - Not Model Power - Define AI Moats

2. Agentic Era Part 2: How the Architectural Battle Between Model Maximalists and Code Craftsmen is Shaping AI's Future

3. Agentic Era Part 3: How MCP and A2A Form the Invisible Operating System of the Autonomous AI Future

4. Agentic Era Part 4: The Exponential Economics of Agentic Applications

5. The Agentic Era Part 5: Building the Agentic Cloud

“ Agentic AI leads to what I call orchestration economics and looking at historical growth will not help determine future winners.” - Raphaelle d'Ornano

A bit similar to how I admire many business thinkers on Substack like or (Check out his YouTube), I find many of her ideas very stimulating.

Decoding Discontinuity

Frameworks for analyzing the financial impact of emerging technologies such as GenAI.

Decoding Discontinuity

A newsletter that provides frameworks for analyzing the financial and strategic impact of emerging technologies like GenAI.

By Raphaëlle d'Ornano

“I specialize in decoding technological disruptions that reshape markets, empowering companies and investors to capitalize on transformative opportunities.” -

⚙️ Read more about AI and the coming Machine Economy: over 250+ featured authors. and others like in the spotlight:

Read top AI publications

🗼 Biography

Raphaëlle D’Ornano, is the founder of D’Ornano + Co., a strategic advisory firm, and of Decoding Discontinuity, a research and investment platform exploring AI’s transformative impact on business models.

• Her proprietary Durable Growth Moat™ framework evaluates how companies sustain competitive advantages through AI-driven shifts, emphasizing architectural resilience and financial adaptability. Passionate about the discontinuity brought by GenAI, Raphaëlle challenges leaders and investors to rethink strategy and unlock enduring value in an AI-powered future. She is based in New York.

🌟 Featured Works

• From Models to Agents: How Manus Rewrites the AI Competition

• The AI Infrastructure Boom Faces the Jevons Fallacy

• DeepSeek: GenAI’s Punctuated Equilibrium?

• The SaaS Renaissance Illusion: Why AI-Driven Growth May Mask Fundamental Business Model Threats

• The Great AI Discontinuity: When Exponential Growth Becomes Visible

This is going to be a very in-depth deep dive that will give you new perspectives on Agentic AI and its role in the Generative AI shift. I like some of her more nerdy and contrarian ideas and original analysis. I think you might as well.

🔊 For paid readers we also have an audio version so you can listen to this on your commute.

The Agentic Discontinuity: How AI’s Evolution from Tools to Autonomous Colleagues Transforms Value Creation

Read more

If you feel covering AI’s human impact on society and people…

Read more

In the first six months of 2025, many of the best Product Newsletters have inadvertently become AI Newsletters too (why is that? 🤔), such is the demand for tips for product managers at the intersection of Generative AI and AI agents into their workflows.

The Intersection of Product and AI is Lucrative and Useful for Professionals

What Product Newsletters learned as the hype of Gen AI continued to grow since ChatGPT went live (about the last 2.5 years), was that talking about AI was a key to growth. And for good reason!

Estimates suggest there are over 200,000 product managers in the U.S. in 2024. PMs like SWEs have been scrambling to adapt to being AI-natives as professionals. Some of the best sources for AI at the intersection of Product Managers and Product related roles are on this platform are:

10. + so many others. See the Long list.

out of Spain is experimenting with building AI tools, and is involved heavily in Venture Capital, content marketing and the AI opp is not lost on him, so much so he even started a Newsletter recently called The AI Opportunity, one of the fastest growing Newsletters in the domain. He’s also found product-market fit for his VC and Newsletter activities on LinkedIn. His Flagship Newsletter is actually called just that, Product Market Fit. He’s consistently near the top of the Business rising leaderboard.

His flair for systems thinking with AI tools really stands out including various viral marketing strategies embedded into his VC, media and startup activities:

Product Market Fit

Product Market Fit

Weekly actionable insights & resources to grow and fund your startup used by best in class companies like Notion, Stripe & OpenAI.

By Guillermo Flor

Guillermo’s unique interests and expertise positions him as someone on the bleeding edge of Startups in the era of AI. This includes creating a suite of Newsletters around his obsessive area of activity. Pay very close attention to his LinkedIn post strategy, if you want to grow your LinkedIn reach.

Startup, VC, Product and AI Knowledge Base & Resources

Guillermo Flor is also part of a social movement. In the last year or so, a new wave of Newsletter operators are distributing knowledge geared for Founders on how to grow their startups. Roughly I would include the following in this list: (these are some of the fastest growing Newsletter publications in Tech and Business on the Substack Platform).

Founders are one of the Best Professional Audiences to Monetize

A lot of the success of a media company like a suite of Newsletters depends on one thing, the monetization potential of the target audience.

It turns out Founders and related job seekers in the startup scene are a great target audience for Newsletter monetization, especially if combined with viral LinkedIn posts tactics: [Product Managers wanting to use AI to become more productive are a great intersection too].

• Founders need access to resources easily and the pricing power of such Newsletters is nearly limitless.

• It’s not enough just to talk about Venture Capital or startup trends, you need to provide real resources of value.

• A lot of early adopters of AI tools dabble in vibe coding and perhaps even have or will found a company.

The Builders for Founders

These people don’t just share knowledge, they share access to resources, networks, access to investors, VC lists, growth-tactics, pitch decks, investor decks, templates, databases, tips and tricks, established playbooks, growth strategy, best practices, startup ideation strategy, lists of angel investors in their region, guides for AI-native lean/agile startups & more:

• + many others.

What I also like about this new movement is they talk about VC realities in Europe, globally and in specific relation to the Generative AI hype cycle and the business of leaner and more opportunistic times. AI native startups can accelerate growth faster with lower costs therefore have improved margins and access to opportunities in this unique time in technology and business.

AI-native startups scale differently and the new generation of Founders understand this.

AI-First Solopreneurship is a Thing?

Whether we are AI optimists or pessimists by preference or ideology, please recognize there is indeed an “AI opportunity” to secure wealth and build incredible things early, if you apply yourself and have the motivation to do so.

• I believe many young people all over the world adopting ChatGPT aren’t doing it out of fun, they want to leverage it in their professional lives, economically and to build new opportunities.

• If young people graduate into a job market where many entry level positions no longer exist, they will indeed have to embrace a new perhaps entrepreneurial reality.

• In every new paradigm or stage of the internet, there is an incredible window of opportunity. 🚀 (You either seize it or it passes you by).

Let’s now get to the deep dive by and give you a sneak peak about what I’ve been trying to convey:

• As a Founder you can fast-track your personal brand, network and influence by using the right AI tools.

• Guillermo’s content isn’t just practical, it’s also motivational and inspires you to dream big. Let me make a small selection of his content to show you what I mean:

Authors Works:

• Sequoia AI Ascent - AI might be the biggest opportunity in humanity’s history

• The 12 Tactics Behind Lovable’s Insane Growth to $50M ARR

• The Next Big AI Opportunity: Buying Traditional Businesses and Powering Them with AI

• The AI founders Playbook

• 💰The Family Office List

Join The AI Opportunity

The AI Operating System for the Startup Founder

Over the past few years, I’ve had the privilege of working closely with founders, exploring countless AI tools, and experimenting firsthand with ways to scale myself (and others) using artificial intelligence.

In this article, I’m going to share a step-by-step guide on how you can create your own AI Solopreneur Operating System—one that can be just as effective for individuals running a one-person business as it is for small startups looking to move fast.

Why an AI Solopreneur OS?

The AI playbook for the startup founder to grow, raise and scale

• What if you could reverse-engineer success and growth with AI?

• What AI tools would you use as a startup Founder to scale your personal brand and boost your sales?

• What strategies and tactics can an AI-first solopreneur use to prepare to build AI tools?

Subscribe now

Read more

We often associate China’s Generative AI leadership with DeepSeek or Alibaba Qwen. But for all the press we get in the west about OpenAI or Google’s new AI products, China BigTech like …

Read more

Read more

Share

In September, 2024 Audio Overviews went viral creating a moment in time when Google Labs’ NotebookLM had its own “ChatGPT” like moment. That Generative AI could magically turn any topic into a synthe…

Read more

I’m still in battery saving mode so updates may be a little less frequent than usual across my channels. As we kick off 2025, a writer I’ve admired last year agreed to write this guest …

Read more

Listen on the Go 🎧 24 minutes 21 seconds

Read more

TSMC is expanding globally and driving new levels of efficiency. Image from the LinkedIn post here by .

Hey Everyone,

As I cover emerging tech with a focus on startups and venture capital trends (e.g…

Read more

For an 🎧 audio version of this article check below. To find AI policy and alignment 🧪 experts read until the end.

When ChatGPT came into the world nearly 2 years ago, one of the big fears was that Generative AI would be used to produce dangerous misinformation. But it turns out, the internet might be more resilient than we thought it was.

Upgrade to Premium

(☝—💎For a limited time get a Yearly sub for under $5 a month✨— ☝).

While Generative AI can be used both to create misinformation campaigns and exploited to amplify disinformation campaigns, we’ve been rather lucky in this U.S. 2024 election cycle. In a manner of speaking, it could have been worse. More powerful models were delayed till after the election and text-to-video Generative AI only get more powerful and capable next year in 2025 with likely more misinformation and cybersecurity issues arriving in the not so distant future.

Where to read about AI policy?

For AI policy perspectives I recommend a few Newsletters here, here and here and for more philosophical takes go here. But how do the academics, AI policy experts and scholars think of Misinformation? Today we turn to who is one of those scholars I find very accessible. We are learning from the history of AI as it is unfolding. Learning from examples, if you will.

Even as Meta has released recently their text-to-video tool called Meta Movie Gen, and the usual controversies and Ad spends bombard the U.S. Presidential race, including the antics of Elon Musk, we have to assume that text-to-video will explode in the 2025 to 2030 period and that the misinformation challenges ahead are going to be an ongoing concern with potentially unforeseen consequences as American BigTech embraces synthetic content in their products. And if we are the product, what will that mean for the future of the internet?

Leave a comment

Elon Musk participating at a Donald Trump rally. October, 5th, 2024.

Harry Law works on governance and policy at Google DeepMind (GDM). He’s also a PhD candidate at the Department of History and Philosophy of Science at the University of Cambridge and a postgraduate fellow at the Leverhulme Centre for the Future of Intelligence.

In an era of the internet where we are normalizing synthetic media, content that’s made up for example, podcasts that aren’t real, it’s getting very weird considering the hallucinations and black hat use cases that are possible and occuring. It’s not even clear what comes next. While Google frames NotebookLM as a research tool, the reality is digital avatars and fake video are becoming more common and the creation and proliferation of hyperrealistic synthetic content meant to deceive could theoretically become much higher. But were our initial fears about the black-hat use cases of Generative AI and foreign adversaries using them, exaggerated? It might be too soon to say, but Harry has some interesting observations on this.

Read more by Harry

• The marginal risk of AI 

• The cruelty of the Turing test

• What’s in a picture? 

• An introduction to AI history

Harry mostly writes about  AI history, policy, and governance at www.learningfromexamples.com. 

🎧 Listen to the Article: 14:40

Misinformation Revisited: Where is the epistemic security crisis?

, September, 2024, writes in a personal capacity.

This is the year of the election. In 2024, around 1.5 billion people are going to the polls as elections take place in more than 50 countries that between them hold almost half of the world’s population. One of the most consequential of these took place in April in India when 642 million voters participated in an election that saw prime minister Narendra Modi win a third consecutive term with the ruling Bharatiya Janata Party (BJP). 

The signs looked ominous before the vote. Arvind Kejriwal, the chief minister of Delhi and the head of the Aam Aadmi Party (AAP) was arrested in late March to answer corruption charges. After his arrest, Kejriwal generated an AI voice recording from behind bars to share with his supporters, which in turn saw BJP voters create AI generations of their own to make light of the situation at Kejriwal’s expense. 

As the election got underway, you can understand why some observers worried that a deluge of AI-powered misinformation was inevitable. But it didn’t happen. An excellent piece in The Atlantic, which explains both Kejriwal’s story and the muted impact of AI on the political process, convincingly argues that “deepfakes have not been as destructive in India as many had feared”. 

According to Indian fact checker Boom Live, of the 258 election-related fact-checks that the organisation conducted, only twelve involved AI. That isn’t to say there were no instances of AI generated misinformation––there have been reports of false election-result predictions, simulated phone conversations, and fake celebrity criticisms––but rather that the scale and impact of AI on the political environment was much less significant than some predicted.   

Then there was the July UK election, which I experienced in all of its misery glory from Cambridge, England. In a result that was of no surprise to anyone, the Labour Party returned to power for the first time since 2010 and the incumbent Conservative Party was swept into opposition. In the middle of the election, the BBC ran an article with the headline “TikTok users being fed misleading election news, BBC finds”.        

Now, there are two different versions of ‘AI’ at play here: there is the recommender system that determines which piece of content to promote, and there is the content itself. Taken together, the concern is that AI threatens to influence both the medium and the message in a way that degrades the strength of the polity. 

But what sort of scale are we talking about? As it turns out, nothing too widespread. The report said that these videos have racked up “hundreds of thousands of views”, which—for an app with daily UK views in the billions—is probably best described as a microbe inside a drop in the ocean. 

There is a growing belief amongst those actually studying the prevalence of misinformation that claims about the purported harms of algorithmically-powered misinformation are not supported by studies of misleading content. We might say that hosting the story on the BBC homepage dramatically overplays the scale of the problem, which in turn may give readers the wrong impression about the importance of misinformation in shaping the outcome of the UK election.    

In other words, if we believe misinformation to include misleading but not outright false content, reporting about the supposed effect of ‘misinformation’ could well be described as, well, misinformation. 

The evidence 

In January early this year, the World Economic Forum issued a report saying that in the next two years AI-powered misinformation and disinformation represented a greater risk than “economic downturn[s]”, “extreme weather events”, and even “interstate armed conflict”. The authors canvassed and amalgamated opinions from “1,490 experts across academia, business, government” in order to create an index of top risks over the next 24 months. In pages 18-21 they explain the reasoning behind some of these views: the emergence of “large-scale artificial intelligence (AI) models” that “have already enabled an explosion in falsified information and so-called ‘synthetic’ content”.

As we saw, though, AI hasn’t yet hopelessly degraded our epistemic security—despite the emergence of powerful and widely available tools that have the potential to do so. In one of the better studies looking at misinformation, Harvard tackled the three most common arguments about AI’s impact on the information environment––increased quantity of misinformation, increased quality of misinformation, and increased personalisation of misinformation––and found that each was overpriced. At the moment, they argue, “existing research suggests at best modest effects of generative AI on the misinformation landscape.”

The report is notable in that it deals with the ‘marginal risk’ of generative models. This is the idea that we ought to understand the risk posed by a new technology (in this case AI) in comparison to the existing risk accepted by society by similar technologies (in this case the internet and associated communications technologies). 

There are essentially two elements we need to consider using this framing. First, as above, we can assess whether AI is enabling increased quantity, quality, and personalisation of misinformation compared to existing methods. Second, we’re looking for evidence that AI is driving changes along these axes in the real world—rather than whether models have the capability to exacerbate the problem in principle.    

Fortunately, some studies are doing that. A June study published in Nature gathers together evidence about precisely these issues. As the authors explain: “Here we identify three common misperceptions: that average exposure to problematic content is high, that algorithms are largely responsible for this exposure and that social media is a primary cause of broader social problems such as polarization.”  

The authors cite famous reports from Facebook that content made by Russian bots from the country’s Internet Research Agency reached 126 million American citizens on the platform before the 2016 US presidential election. Sounds big, until the authors remind us that this content represented 0.004% of the posts that US citizens saw in the Facebook newsfeed during the period of study. 

This dynamic––of failing to contextualise the amount of low quality or misleading sources that a person consumes relative to their information diet as a whole––is commonplace in the world of misinformation studies. A well known 2019 study, for example, categorised 490 websites identified as ‘untrustworthy’ – but failed to account for the fact that this made up only 5.9% of US citizens’ visits to news sites on average. When you take television into account, the figure drops to just 0.1% of US citizens’ media diet. 

That the overall surface area for misleading content is small necessarily means that the impact of AI is limited. But how limited? And what, exactly, do we make of the corrosive effects of algorithms on political life that we hear so much about? The most famous idea is that of the ‘filter bubble’ that traps users online in an echo chamber of distorted information.  

The core set of claims here is that recommender systems aim to promote user engagement,  that inflammatory content is more engaging than average, and algorithms subsequently promote it to users to boost view counts. Unfortunately for this story, the existence of large effects on information consumption from algorithms hasn’t actually been established. A major 2023 study even showed that algorithmically-curated feeds supply less untrustworthy content than chronological feeds. 

A recent study found that AI bots trained on individuals' YouTube viewing histories and following the platform's algorithmic recommendations consumed less partisan content than the actual human users did. This is probably because people who engage with extreme or untrustworthy content tend to seek it out themselves. In other words, people generally go looking for misinformation rather than passively falling victim to it.

After all, a landmark 2021 paper demonstrated that extreme content is viewed on YouTube by a small percentage of the population who tend to consume similar content elsewhere. The upshot is that consumption is driven by demand, not algorithms.

Misinformation for thee, but not for me

So, the question is, why exactly does the ability of AI to degrade the information environment seem overpriced? Well, the advent of social media very obviously coincides with political polarisation, a perceived collapse in the quality of political discourse, and the rise of extremist ideologies. As a result, lots of people tend to suspect that there may have been a causal relationship between them.

For AI, the concern is that it will exacerbate these trends and deliver more of the same. But as the above should make clear, it’s not all that clear that social media––where content is created and recommended by AI––really is degrading the health of the polity. Yes, it feels like it is – but there’s not really any hard evidence to back it up. 

There are two stories we can tell ourselves here. Either the marginal risk of AI is high because the actual impact of communications technologies on political discourse is low, or the marginal risk of AI is low because you believe that social media has already poisoned the well.

But is that right? There is, I suppose, a third option: that just like the advent of social media the advent of AI feels riskier than it actually is. This option, which I think probably fits my own world model best, is closely connected to what are essential problems with ‘misinformation’ discourse.

The unfortunate truth is that, while we all have our biases, those who seek to isolate misinformation as a discrete area of study have an unfortunate tendency to believe that misinformation is a thing that happens to someone else. Misinformation for thee, but not for me. 

They like to believe that there is such a thing as an objective science of misinformation. But there can’t be. Putting aside the fact that most people don’t consume much outright false information, popular definitions of misinformation have been broadened to encompass true but misleading content. Based on this definition, you get into the absurd situation whereby ‘misinformation researchers’ end up spreading misinformation (if you wanted to call it that) themselves because ‘true but misleading’ could mean just about anything.  

To be clear, though, that doesn’t mean that I think the information environment is in good health. 

Dan Williams of the University of Sussex makes the case that, even if we reject the beliefs that A) misinformation is the most worrying form of bad information and B) misinformation is easy to identify, that does not mean that society is enjoying a period of epistemic vitality. 

As Williams explains: “First…communication can be - and frequently is - highly misleading without ever involving blatantly false or fabricated content. Second, once you broaden the focus on bad information to include any content that might be misleading even if it does not involve outright falsehoods and fabrications, bad information is not easy to identify.”

This is the crux of the issue. For AI, we might say that the technology increases the ability of people to create misleading content. But even so, if we think about the evidence, do we really believe that an avalanche of synthetic media––of memes, and jokes, and clear fabrications––should be called misinformation? I doubt anyone would credibly call Donald Trump posting a video of him and Elon Musk dancing an example of misinformation. 

But why not? After all, the fact checkers would agree that it didn’t happen. Some researchers even described an obvious meme as an example of the misinformation monster in action. You might think these are silly examples, but––as the Indian election showed––this messiness is the rule, not the exception. 

AI is rapidly increasing the amount of content on the internet, but I don’t see any reason to believe it represents a major departure from established methods of communication. Powerful image, video, and audio generation models primarily lower barriers to entry for a party to create content that they could have done with more time and resources on their side.

Of course, it doesn’t just do that. There are genuine instances in which AI can be used in a way that represents a change to the status quo. A good example here is the use of voice generation models, which allows someone to convincingly clone the voice of anyone they like (providing they have enough examples of them speaking).   

For public figures, that poses no issue. It’s not hard to get a voice sample from anyone in a position of power or influence given that such people tend to do a lot of public speaking. Here lies the contradiction. If it’s easy to create convincing voice clones, why have they failed to influence political life by impersonating politicians, celebrities, and others? 

I’m not aware of much empirical work on this issue, but if I had to guess, I would put it down to the fact that such impersonations travel primarily via social media. Major outlets check these things before publishing, and sites like X eventually get to the truth through a combination of community notes or a simple “chat, is this real?” 

Clearly, there might be some people who get the wrong end of the stick for a moment or two. But, when verification fails to materialise (or when additional information quickly counteracts the original claim), the artefact in question doesn’t tend to stick. 

That’s the problem with fears around an AI field-misinformation apocalypse. Sooner or later, it either happens or it doesn’t. And right now, much like empirical studies that seek to understand the impact of human generated misinformation, the fear of the thing doesn’t bear much resemblance to reality.    

Thanks for reading!

Subscribe to LFE

Appendix

Does AI policy and alignment research interest you? Here are some folk I recommend following if you use LinkedIn.

Unfortunately the founder of is no longer with us. In Memoriam: Abhishek Gupta (Dec 20, 1992 – Sep 30, 2024).

💼 AI Policy people I follow on LinkedIn:

1. https://www.linkedin.com/in/ravit-dotan/ - Ravit Dotan

2. https://www.linkedin.com/in/ristouuk/ - Risto Uuk

3. https://www.linkedin.com/in/msheehan2/ - Matt Sheehan

4. https://www.linkedin.com/in/jonasschuett/ - Jonas Schuett

5. https://www.linkedin.com/in/rumman/ - Rumman Chowdhury

6. https://www.linkedin.com/in/ryan-donnelly-enzai/ - Ryan Donnelly

7. https://www.linkedin.com/in/shea-brown-26050465/ - Shea Brown

8. https://www.linkedin.com/in/reid-blackman/ - Reid Blackman

9. https://www.linkedin.com/in/kay-firth-butterfield/ - Kay Firth-Butterfield

10. https://www.linkedin.com/in/staceyhking/ - Stacey King

11. https://www.linkedin.com/in/miles-brundage-49b62a4/ - Miles Brundage

12. https://www.linkedin.com/in/acomomcilovic/ - Aco Momcilovic

13. https://www.linkedin.com/in/matthijsmaas/ - Matthijs M. Maas

14. https://www.linkedin.com/in/katharina-koerner-privacyengineering/ - Katharina Koerner

15. https://www.linkedin.com/in/buildingtrustedaiholistically/ - Pamela Gupta

16. https://www.linkedin.com/in/abhishekguptamcgill/ - Abhishek Gupta

17. https://www.linkedin.com/in/carlosig/ - Carlos Ignacio Gutierrez

18. https://www.linkedin.com/in/alextamkin/ - Alex Tamkin

19. https://www.linkedin.com/in/huw-roberts-3539b5b7/ - Huw Roberts

20. https://www.linkedin.com/in/borhane/ - Borhane Blili-Hamelin

21. https://www.linkedin.com/in/kzenner/ - Kai Zenner

22. https://www.linkedin.com/in/carlos-eduardo-torres-giraldez/ - Carlos Eduardo Torres Giraldez

23. https://www.linkedin.com/in/eugenio-v-garcia-414316157/ - Eugenio V Garcia

24. https://www.linkedin.com/in/harry-law-934a42b4/ - Harry Law

25. https://www.linkedin.com/in/kevin-klyman/ - Kevin Klyman

26. https://www.linkedin.com/in/lewis-ho-a74380178/ - Lewis Ho

27. https://www.linkedin.com/in/walter-m-pasquarelli/ - Walter Pasquarelli

28. https://www.linkedin.com/in/henryajder/ - Henry Ajder

29. https://www.linkedin.com/in/gary-marcus-b6384b4/ - Gary Marcus

30. https://www.linkedin.com/in/keith-sonderling/ - Keith Sonderling

31. https://www.linkedin.com/in/margaret-levi-67365544/ - Margaret Levi

32. https://www.linkedin.com/in/zak-rogoff-3301689/ - Zak Rogoff

33. https://www.linkedin.com/in/conor-griffin-6902bb7/ - Conor Griffin

34. https://www.linkedin.com/in/anita-ho-4804835a/ - Anita Ho

35. https://www.linkedin.com/in/dr-brenda-kubheka-7a1b1917/ - Brenda Kubheka

36. https://www.linkedin.com/in/zohaibknoorani/ - Zohaib Noorani

37. https://www.linkedin.com/in/lmiller-ethicist/ - Laura Miller

38. https://www.linkedin.com/in/lila-shroff/ - Lila Shroff

39. https://www.linkedin.com/in/jenpan/ - Jennifer Pan

40. https://www.linkedin.com/in/gillian-k-hadfield-1773987/ - Gillian K. Hadfield

41. https://www.linkedin.com/in/bryson/ - Joanna Bryson

42. https://www.linkedin.com/in/benjaminkultgenphd/ - Ben Kultgen

Image cred: Pixabay.

Hey Everyone,

What is AI literacy? Is prompt-engineering really a skill that has a bright or long-term future? How are young people using AI? As I continue to read the top AI in Education Newsletters, I keep coming back to these fundamental questions.

Upgrade to Premium

(☝—💎For a limited time get a Yearly sub for just $5 a month✨— ☝).

With a strong background in journalism (like 4 years at Reuters) and teaching, Mike Kentz is an interesting thinker. He has a very unique background and professional experiences of two career paths that might be impacted the most by AI: teaching and journalism. With AI for some of us these are crazy times. Around seven months ago he founded a Newsletter, and around three months ago he founded Zainetek Educational Advisors. See their services.

• Since the advancement of AI in 2022, AI literacy is taking on a life of its own.

• Whether you are a parent, teacher or a concerned citizen, you might want to read this.

The importance of AI literacy for all | Mike Kentz | TEDxSavannah - 12:09 🎧 ~ Ted Talk on AI Literacy

If you think AI Literacy is an important topic, consider sharing this.

Share

The Illusion of AI Literacy

By of AI EduPathways.

In the realm of modern buzzwords, few are as pervasive—and as misunderstood—as "AI Literacy." It’s the kind of term that gets tossed around in every meeting, every conference, every article that dares to touch on the future of technology. We hear it constantly: "You must learn AI," "AI won't replace you, but someone who knows how to use it will," and "We need proper AI training."

These statements, while repetitive, point to an emerging reality: the pressing need to define and teach AI Literacy. According to researchers at Georgia Tech, AI Literacy encompasses a range of competencies that enable individuals to critically evaluate AI technologies, collaborate with them effectively, and use them as tools in various aspects of life.

Yet, when you dig into the details, the concept starts to feel overwhelming. The infographic produced by these researchers attempts to capture the essence of AI Literacy, but instead of clarity, it offers a labyrinth of skills, competencies, and knowledge areas—each as intricate as the next.

Read More work by the Guest Author :

📚 Subscribe to Mike’s Newsletter.

🔵 Download the resource for teachers ““AI in Education: Transform Challenges into Opportunities” - here (scroll down).

🌍 AI, Education, and The Five Stages of Grief (Part 1)

🔴 On ‘Strawberry,’ Reasoning, and Intuition

• How should we engage with LLMs? How will frontier models change our fundamental relationship with technology?

What will be the the looking-glass relationship of how we relate to ever more powerful models trained on the sum total of all human language and data?

Upgrade to Premium

The Problem of Scale

This complexity is why I’m writing this. AI Literacy, as it stands, is too vast, too vague to be practical. There is a pressing need to break it down into digestible pieces if we’re ever going to integrate it into education and society at large.

To understand this, consider the state of education today, where AI is either heralded as the ultimate learning tool or decried as the destroyer of critical thinking. Humanities teachers, in particular, find themselves on the defensive. Traditional assessments like essays are rapidly becoming obsolete, replaced by AI-generated content that can mimic student work with a concerning level of accuracy.

In the face of this shift, a new form of digital literacy is being called for—a literacy that encompasses the safe and effective use of AI. But over the past year, educators seeking to define the term and how to teach or embed it into the next generation of students have continued to run up against a major obstacle: The concept of AI Literacy is too broad, too ambiguous, to be effectively taught or learned.

Consider the European Union's AI Act, which mandates AI Literacy training for corporate professionals. The law's requirements are maddeningly unclear, demanding that staff possess "sufficient skills, knowledge, and understanding" of AI and its risks. But what does this mean? What skills? Which risks?

The Danger of Semantic Satiation

This is where AI Literacy stands today.  It’s expansive and confusing nature has led to a backlash in education circles. Raise the term “AI Literacy” in a conversation with a University Professor and watch their eyes glaze over, roll, or even shut. To be honest, I am not sure I blame them.

Arvind Narayanan and Sayash Kapoor, a professor of computer science and a PhD Candidate at Princeton University, have pointed out that the term “AI” itself is overly broad, encompassing technologies with vastly different levels of competence. The same critique can be applied to AI Literacy: it’s too vast, too varied to be a single discipline.

As I see it, the blame for the current state of affairs lands in one of three areas. Either the education market has stuffed so many concepts under the umbrella of AI Literacy that it has become meaningless, it never had a clear definition to begin with, or it is a victim of semantic satiation — the phenomenon where repeated words lose their impact. In each case, the catch-all ambiguity of AI Literacy now serves to obscure more than it reveals.

The Path Forward: Narrowing the Focus

To move forward, we must break down AI Literacy into smaller, more manageable disciplines. These should align with existing educational frameworks, making them easier to teach and learn. I propose three subdivisions:

1. Machine Literacy: This is the study of the AI engine itself. It focuses on the underlying technology, employing statistical analysis and computational thinking to understand what’s happening beneath the surface.

2. AI Historical Studies: This discipline covers data ethics, history, and the cultural impact of AI. It asks the big questions: How did we get here? What will the future look like under AI’s influence?

3. LLM Literacy: The most crucial of all, LLM Literacy deals with Large Language Models—the AI tools most commonly used by the average person. Unlike Machine Literacy, which requires a deep understanding of AI's technical aspects, LLM Literacy is about practical interaction. It’s about knowing how to write, read, and critically assess the outputs of these models.

The Car Engine Analogy

To illustrate this, consider the analogy of a car engine. Do you need to understand how a car engine works to be a safe and effective driver? Not really. Sure, knowing the mechanics might make you slightly more tuned into your car's performance, but this knowledge is only applicable on the margins. Driving requires a wholly different set of skills from building, maintaining, or even understanding the engine itself.

Similarly, to use AI effectively, you don’t need to understand the intricacies of predictive analytics, machine learning, or neural networks. What you do need are the skills to engage with AI tools—especially LLMs—in a thoughtful and reflective manner.

Acknowledging this reality makes it easier to split AI skills across disciplines. To be clear, I am not arguing for an abandonment of the study of the engine. Only a realization; Studying the engine prepares you to be a mechanic, not a driver. In the context of AI Literacy, this allows for clean splits across disciplines, making the navigation and embedding of each skill – among adults and students – more navigable.

Said differently, you would not hire a mechanic to teach you how to drive. Our Driver’s Ed Instructors, rightly, do not discuss the engine with novice drivers. Instead, they focus on the rules of the road, the strategies for safe and responsible driving, and developing situational awareness in the driver-in-training. 

In the above chart, you can see that I hand off the responsibility for teaching safe and effective driving of Large Language Models – specifically – to writers, journalists, and English teachers. People who have spent their entire lives studying words, language, and communication.

The same goes for understanding the ethics or history of AI. While it’s important to know that cars contribute to environmental degradation or that the oil industry has fueled global conflict, this knowledge doesn’t impact your ability to drive safely. In the same way, understanding AI's ethical implications – such as questions around Data Ethics and its impact on our humanity – is valuable, but it’s not essential to using AI effectively. They are different skills.

Mastering the Art of Engaging with Large Language Models

With respect to the utilization of Large Language Models, “literacy” must first acknowledge that its use is an art, and not a science. It leverages the skills of creative writing and journalistic thinking - since, in many ways, an engagement with an LLM mirrors an interview. Question, Answer, Question, Answer – with a broader aim in the mind of the questioner.

In that vein, we can more easily accept that the future of safe and effective use lies in the hands of the professionals most acquainted with the art of the written and spoken word. If you are working with an engine that produces language, with whom would you rather confer – a technologist or a writer?

Within that vein, the delivery of “LLM Literacy” can be broken down into a series of understandings that will allow us to shift the lens of safe and effective use towards a more Humanities-centered approach.

Step 1: Command the Power of Words

The very first move in this strategic dance with a Large Language Model (LLM) is the craft of writing—a question, a request, or even a simple statement. Some may dress this up with the title "prompt engineering," but this will soon be a relic of STEM-centered jargon that has been applied to a Humanities skill. Consider the masters of language throughout history—did they ever claim to "engineer" a thesis statement or a question? No. They wrote them. Language is an art form, and in this realm, you are an artist, not an engineer.

Step 2: The Mirror Reflects

When the LLM responds, it mirrors back to you in text form. While some models can create images or videos, this article focuses on models that produce text. Even in the case of image generators, the engine is still manipulated by words. And when an LLM produces text, it is via close reading that a user can determine its veracity, value, or utility.

Step 3: Decipher the Code

Our next realization should be that analyzing the outputs of an LLM is not about verifying facts alone; it is about evaluating value. When using the LLM as a brainstorming ally—a role where it thrives—it does not deal in absolutes. Ideas are neither true nor false; they are either valuable or worthless.

Our focus on hallucinations—when an AI fabricates a claim or a fact—is holding us back. In the evolving world of AI literacy, you must understand that your primary task is to evaluate the usefulness of the output – especially in the case of a collaborative dialogue – not just its accuracy. On the other hand, if your goal is research, the path is clear: verify the LLM outputs through the sources provided. But most often, your task is to discern what value the LLM's output brings to your table.

Step 4: Reflect and Adapt

Because LLMs produce language so rapidly, we are often tempted to keep up. Not only will we fail at that endeavor, but it is within the discipline of slowing down and reflecting that we can ensure a level of responsibility and safety. Ask yourself; Where has your strategy led you? Where do you wish to go next? The process is one of metacognitive decision-making, a trait shared by great writers and journalists. This continuous cycle of reflection, in conjunction with your imagination and creativity, allows you to chart the best course forward.

Step 5: Refine the Written Word

Having reflected, you return to writing. You refine your request, provide feedback, and strategically structure your language to guide the LLM toward your desired outcome. The principles of good writing—broad-to-narrow structuring, clarity of context—apply here as well. If you are already skilled in the art of communication, you will find these techniques natural extensions of your abilities.

Step 6: The Endless Cycle

At this point, working with LLMs becomes simply a question of habit and repetition. It is a cycle of writing, reading, reflecting, and writing again. To be sure, there will be some cases where the model is used as a means to an end – an efficiency-booster. But with a deeper and more purposeful approach, the user gains an awareness of its larger universe of possibilities. Understanding the parameters of what it can do – through the skills of the Humanities – will allow you to make concerted choices regarding the use of an LLM for efficiency or for depth, all the while maintaining a deeper awareness within your subconscious. 

Conclusion: A Call for Practical AI Literacy

In conclusion, AI Literacy must be redefined, subdivided, and made practical if we are to teach it effectively. An approach to LLM Literacy through the lens of the Humanities will allow us not only to teach effective use but also develop writing, reading, and reflective skills that will be vital to the future health of society. We must move away from the sprawling, undefined concept it has become and focus on specific, actionable skills. Only then can we hope to integrate AI Literacy into education and, by extension, into society as a whole.

Bio:

Mike Kentz - Founder/CEO

Mike Kentz is an award-winning educator with over 14 years of experience across journalism and teaching. He is a TEDx Speaker and his work in AI in Education has been featured in EdSurge, Edutopia, The Wall Street Journal, Canadian National Radio, and more. 

His workshops have been formulated as a product of direct classroom experience in 2023-24, when he developed a ground-breaking approach to teaching with AI in the classroom that has since been accepted for presentation at multiple national and regional education conferences, including the TCEA AI for Educators Conference, the 2024 Course Hero Education Summit, The Learning Ideas Conference and the University of Central Florida’s prestigious Teaching and Learning with AI Conference. 

These experiences now inform his unique, Humanities-based approach to developing AI Literacy among adults and students and helping schools to adapt to the advancement of artificial intelligence.  

He is a graduate of Georgetown University (B.A. American Studies), Relay Graduate School of Education (M.A.T.), and Wilkes University (M.A. Creative Writing). 

In his earlier career as a financial journalist in New York City, his reporting and writing was regularly featured in a wide variety of national news publications including Thomson Reuters, The New York Times, The Los Angeles Times, The Chicago Tribune, CNBC, Fox Business, and more. He now lives in Savannah, Georgia with his wife, dog, and cat.

Four Popular Articles:

1. Stop Grading Essays, Start Grading Chats: A New Assessment Framework for the Age of AI: 

   1. A New Assessment Design Framework for the AI Era: Reflections Part 1 (substack.com)

2. Harvard AI Pedagogy Project: Safe and Effective Use of Personality Bots in the Classroom 

   1. Harvard AI Pedagogy Project: Safe and Effective Use of Personality Bots in the Classroom (substack.com)

3. Stranger Danger: Effective AI Use via Critical Thinking and Humanities Skills

   1. Stranger Danger: Effective AI Use via Critical Thinking and Humanities Skills (substack.com)

4. AI, Education, and The Five Stages of Grief:

   1. AI, Education, and The Five Stages of Grief (Part 1) (substack.com)

Hey Everyone,

Getting back to some applied AI tool discovery: Let’s take a spin of one of the most trending products that is now expanded globally using Gemini 1.5’s multimodal capabilities.

Upgrade to Premium

(☝—💎For a limited time get a Yearly sub for just $5 a month✨— ☝).

Products like Perplexity and Copilot 365 are changing how we interact with AI-generated content with Perplexity pages (May, 2024), Copilot Pages (September, 2025), Genspark Sparkpages (June, 2024) and also in September, Google upgraded how easy it is to use NotebooksLM and create Audio Overviews.

• The interface/UX of NotebooksLM is extremely well thought out.

• I also use this tool to get summaries of my articles.

• Google’s NotebookLM app can now generate ‘lively’ audio discussions with two AI hosts about the documents you’ve given it. Collaborate with what Google proudly refers to as a “virtual research assistant”.

For audio learners this is a useful form of a page, but puts it into a podcast format instead if you select the Audio Overviews format.

“Google’s AI note-taking app NotebookLM can now explain complex topics to you out loud”

• Google is enabling the creation of fake or synthetic podcasts, which they call Audio Overviews, on any topic and it is super easy to use!

• Try Genspark 🔎 (Genspark is blazing fast and a decent alternative to Perplexity imho).

Google’s NotebooksLM re-launch of Audio Overviews has been trending on social media in mid September, 2024. Your personal data is not used to train NotebookLM, so any private or sensitive information you have in your sources will stay private, unless you choose to share your sources with collaborators.

NotebooksLM is not unlike the research tools that already exist in the Generative AI market. However Audio Overviews while the output can feel generic has surprisingly reusable applications.

Google brands NotebookLM as a kind of practical AI-powered research assistant that lets people in your organization interact with trusted source content to get grounded insights.

• Source grounding

• Summarization and Q&A

• Idea Generation

• Structured Formats

• Modular features

I asked who recently conducted a webinar: "AI and the Writer's Mind: Exploring the Cognitive Shifts of AI-Assisted Writing" for his take and a short guide on this tool.

is the founder of the Newsletter Educating AI. and has become fairly entreprenrurial of late at the intersection of Education and AI. Join the Pragmatic AI Solutions community for more of his insights and to meet others as well.

Watching NotebooksLM “grow up” has actually been fairly interesting as far as Google AI products go. So why am I focusing on audio? Well OpenAI’s advanced voice mode is also out this week (probably announced the day this post goes live).

With more immersive text-to-video and audio products soon available and the rise of apps like Suno AI, how we "experience” Generative AI is also changing from a chatbot of 2 years ago, to a more multi-modal educational journey. The AI tools on the research and curation side are also starting to reflect these advancements 🎉.

🕯️📖 Recent Articles you may have missed: 📚

1. Sam Altman Pitches Utopian impact of AI while Accepting UAE Oil Money Funding

2. Top Software Engineering Newsletters in 2024

3. OpenAI Keeps Releasing Prototypes & Previews of Actual Products

4. What is the Ceiling for Nvidia backed Japan based Sakana AI (さかな)?

5. AI for Non-Techies 🛠️: Top Tools for Search, Agent Building, Academic Paper Reviews & Sales Automation

🙏🏻 Thanks for reading AI Supremacy. To Sponsor go here. 👀

Subscribe now

Audio Overview is still experimental and has some known limitations, but is a fairly refreshing audio alternative to annoying chatbots.

By , September, 2024.

Hey there, AI explorers!

Today, we're diving deep into Google's NotebookLM. It's making waves in the education sector, but is it all it's cracked up to be? Let's find out!

For years, we've witnessed OpenAI create tools that, while impressive, often caused headaches for educators. It seemed like every new release was another challenge for teachers trying to maintain academic integrity. This trend began with GPT-3.5 in 2022, escalated with GPT-4 in 2023, and reached new heights with ChatGPT 4o and o1 in 2024.

But while OpenAI was stealing headlines, Google was quietly cooking up something interesting. NotebookLM didn't appear out of thin air – it's the result of years of R&D. But is it the hero education needs, or just another tech giant muscling in?

Google's AI Journey: A Trip Down Memory Lane

🗓️ 2018: Smart Compose in Gmail (Not educational, but hey, it's a start!) 🗓️ 2021: LaMDA enters the chat 🗓️ 2022: PaLM joins the party 🗓️ 2023: NotebookLM makes its debut

Google's foray into AI-powered educational tools can be traced back to 2018 with the launch of Smart Compose in Gmail. While not directly educational, it showcased Google's ability to create AI that could understand and generate human-like text. 

In 2021, Google introduced LaMDA (Language Model for Dialogue Applications), which laid the groundwork for more sophisticated conversational AI. This was followed by the release of PaLM (Pathways Language Model) in 2022, which further advanced Google's language understanding capabilities.

NotebookLM, first introduced in 2023, represented a culmination of these efforts. It's like the final boss of Google's AI education game. It's got all the moves from its predecessors, plus some new tricks up its sleeve.

To learn about another amazing Google Education product, check out my post on LearnLM!!!

NotebookLM: What's All the Fuss About?

NotebookLM just got a glow-up, and people are losing their minds. Here's why:

1. Document Buffet: Feed it ALL the documents. It's like that friend who remembers everything you've ever said.

2. RAG Processing: Fancy tech speak for "really good at understanding stuff."

3. Educational Swiss Army Knife: Unlike some AI tools that make teachers want to quit, this one might actually help.

But the real showstopper? It turns your docs into a podcast. 🎙️ Talk about bringing your essays to life!

My NotebookLM Adventure: A Play-by-Play

Buckle up, folks. I took NotebookLM for a spin, and boy, was it a ride! 

Here is a link to NotebookLM so you can explore for yourself!

Try Notebooks LM

Step 1: Feeding the Beast

I started by throwing everything I had at NotebookLM. My dusty college essays? Check. That half-finished research paper on AI ethics? You bet. Even my embarrassingly ambitious sci-fi novel outline made the cut. NotebookLM devoured them all with the appetite of a grad student during finals week.

The upload process was simpler than my morning coffee ritual. Drag, drop, done. It felt like I was building my own personalized AI brain, one document at a time.

Step 2: AI Mind Reader

With my digital library loaded, it was time to play detective. Here is what the loaded library looks like:

Here is what the interface looks like when you click into a source. 

Note that there is an embedded generative AI tool that you can use to make inquiries about content on a specific source or the total collection of sources. You can even prompt the AI app to generate new documents based on the uploaded materials.

So–back to playing detective–I typed in, "What's the common thread between my AI ethics paper and my sci-fi novel?"

NotebookLM paused for a millisecond (probably judging my writing style) before spitting out an answer that made me question if I'd accidentally uploaded my diary too. It pointed out themes I hadn't consciously realized were there, drawing connections between academic arguments and fictional plotlines. Mind. Blown.

Step 3: Podcast Magic

Here's where things got wild. I clicked the "Audio Overview" button, half expecting a robotic monotone. Instead, I was greeted by two AI voices that sounded more engaging than some of my old professors (sorry, Dr. Johnson).

I initially thought the podcast would only focus on the highlighted source, but I was wrong. They bantered back and forth about ALL of my documents, taking a seriously deep dive into my ethical and philosophical content. 

Be forewarned: this is serious uncanny valley territory. While folks are already critiquing NotebookLM's repetitive joke patterns, they may be missing the point. The conversations are engaging and persuasive, and this product is synthesizing text at a higher level than any other generative AI technology on the market.

Check out my podcast here: (You can easily download Audio Overviews about anything)

The Verdict: Mind-Bending Potential

As I sat back, my brain buzzing with new connections and ideas, I couldn't help but marvel at the potential of NotebookLM. It's not just a tool; it is something uncannily bordering on a super-sassy, fun, engaging research assistant available to students and teachers alike 24/7.

For students, it could transform the way they synthesize information from multiple sources. For researchers, it's a brainstorming partner that never runs out of coffee. And for sci-fi writers? Well, let's just say AI might be both our muse and our competitor. Folks, we are living in very strange times!!! 

The Elephant in the Room: Data Privacy

Now, I know what some of you might be thinking: "Isn't this just another way for Google to get our data?" You're not wrong to be cautious. NotebookLM's ability to process so much information is amazing, but it also means we're sharing more of our stuff with a big company.

But for now, let's take a moment to appreciate how cool this technology is. It's like having a super-smart assistant who can read all your notes, summarize them, and even talk about them!

What the Experts Are Saying

Dr. Philippa Hardman (Learning Tech Guru)

Dr. Philippa Hardman, an expert in learning tech, tried out NotebookLM and was blown away. She uploaded her conference presentation slides and, in just a couple of minutes, got a summary, study guide, FAQs, and even a podcast!

Marc Watkins (AI Sage)

Marc Watkins, one of our wisest AI sages, points out how NotebookLM could change the game for students. He outlines several exciting applications:

1. Interactive syllabi

2. Deep dive into presentations

3. Digitizing analog brainstorming

4. Multi-source exploration

5. Portfolio building for writing courses

These applications showcase the versatility of NotebookLM and its potential to transform various aspects of the educational experience.

The Million-Dollar Question

Is NotebookLM the future of education or just another shiny tech toy?

• Pro: Could revolutionize learning

• Con: Privacy concerns (as always with big tech)

Wrapping Up: To NotebookLM or Not to NotebookLM?

Google's NotebookLM is shaking things up in the world of AI and education. Its development represents a significant shift in how tech giants approach educational tools. While we should always be mindful of how we share our data, it's exciting to see tools that could actually make learning easier and more engaging.

As we look to the future, it's clear that AI tools like NotebookLM will play an increasingly important role in education. The key will be leveraging these tools to enhance, rather than replace, traditional learning methods.

So, what's your take? Ready to give NotebookLM a whirl, or sticking to good old-fashioned brain power? Drop your thoughts in the comments!

Remember, the future of learning is here – with a Google twist. But it's up to us to decide if that's a good thing. 🤔

_____________

I’m Nick Potkalitsky, a dedicated researcher and educator focused on revolutionizing education through AI-driven methods. With a doctorate in narrative and rhetorical studies, I’ve worked to improve research, instruction, and assessment systems across multiple disciplines. At The Miami Valley School, I lead the development of an AI-enabled curriculum and advocate for equitable grading practices.

As the founder of Pragmatic AI Solutions, I upskill educators and professionals for the AI age, while hosting the Pragmatic AI Community, which offers free resources, webinars, and AI literacy courses. Join me as we explore the future of education through my Substack, Educating AI, featuring curriculum briefs, research, and interviews.

LinkedIn

My Website

Pragmatic AI Community

Hey Everyone,

Is Generative AI going to lead to a new economy? This is one of the biggest questions we’ve had since late 2022. IDC, the market-forecasting giant, projects that AI — including but not limited to generative AI — will boost the global economy by $4.9 trillion in 2030, up from just below $1.2 trillion this year.

Ollie is based in Bangkok Thailand and I am currently in Taiwan. No matter where you live there is a sense that the world is changing and AI may be accelerating the pace of change.

Generative AI is poised to significantly impact the global economy, potentially leading to what could be considered a 'new economy.' This article will focus on the landscape of AI unicorns poised to usher in a new economy on the back of new capabilities, an emerging agentic AI and text-to-action based economy.

Upgrade to Premium

(☝—💎For a limited time get a Yearly sub for just $5 a month✨— ☝).

Companies like Klarna have famously created more efficiency and that has led to layoffs due to their adoption of Gen AI. Klarna: AI lets us cut thousands of jobs - but pay more. So is Klarna an exception or the new rule? It might be too soon to say for sure.

Will the AI startups and tools of today lead to many more such case studies like Klarna in how companies adapt?

Ollie Forsyth is the founder of New Economies, a newsletter discovering and sharing the latest technology trends shaping our world. Previous editions include: The Ageing Population, The Workplace: AI Tools, AI Agents and, The Transition from Operators to Angel Investors.

I am obsessed with AI startups and thinking about them in terms of what they will accomplish in the years ahead. But who are they? Where do they come from? What is their trajectory?

What will their impact on the real world be? Will new kinds of AI reasoning reshape society? What will be the state of the new AI economy?

Venture capital investment in generative AI reached approximately $21.3 billion in 2023. The US and Canada accounted for nearly 45% of private equity-backed investments in 2023. Many pundits and AI evangelists say Generative AI is poised to usher in a new economy by significantly boosting productivity and economic growth.

The New Economies newsletter is really strong in AI coverage and so I asked if we could share some of his work. With innovation advancing rapidly on a regular basis, you may not always know about the latest trends taking shape so he’s providing these to you for free. Read about.

• Who are the AI unicorn startups of this era?

• Data on funding, top colleges and locations of HQ operations

• Median valuations, top VCs and AI investors

• Access to Ollie’s list of Investors

By , August, 2024.

In this edition, we discover the latest AI unicorns and the characteristics behind them. Additionally, we uncover and spotlight over 500 early-stage AI startups that should be on your radar.

Let’s get started!

Over the past year, AI has really hit its stride. What was once a buzzword is now a crucial part of business strategies across various industries. AI is not just another tech trend; it's transforming healthcare, finance, education, and shaping the future of work and daily life.

The Startup Landscape

The number of AI startups is surging as more founders seize one of the most promising opportunities of our time. If we look at the startups featured in our recent market maps covering AI Agents & Generative AI, most of them did not even exist a few years ago.

Explore the AI startup ecosystem

The Startup Fundraising Landscape (Q2 2024)

Funding to companies in AI more than doubled quarter over quarter to $24 billion — representing 30% of all dollars invested, the largest quarter for AI funding in recent years.

Five out of six billion-dollar funding rounds went to AI companies. Elon Musk’s xAI raised $6 billion, and AI infrastructure provider CoreWeave raised $1.1 billion. Automated driving company Wayve, data preparation company Scale AI and AI biotech company Xaira Therapeutics each raised billion-dollar rounds.

Aside from AI, global startup funding picked up in the second quarter, reaching $79 billion — rising 16% quarter over quarter and 12% from the $71 billion invested in Q2 2023. Mega-rounds — fundings $100 million and above — accounted for much of the increase this past quarter.

Source: Crunchbase

Be the first to know about the latest technology trends shaping our world before anyone else does. 52,000+ others already do.

Meet the 140+ AI unicorns

Meet the 144 unicorn companies working in AI.

❤️ Like this post at the top of the email to receive free access to our AI unicorns data, delivered directly to you shortly! For immediate access, consider upgrading to become a premium subscriber.

When were these AI unicorns founded?

Since 2000, our data reveals that over 140 unicorns have emerged in the AI ecosystem. Notably, 28 of the AI startups founded in 2016 have already reached unicorn status.

As the startup ecosystem has matured, the number of unicorns has significantly increased in recent years.

Which universities did some of these unicorn founders attend?

• Stanford University (USA): A leader in tech entrepreneurship, with founders from companies like OpenAI, Databricks, Inflection AI, PonyAI, and Anthropic.

• Massachusetts Institute of Technology - MIT (USA): Known for innovation, with connections to founders of Scale AI.

• University of California, Berkeley (USA): A hub for tech startups, including those like Databricks.

• Harvard University (USA): Renowned for leadership and innovation, with alumni involved in companies like DeepMind.

• University of Oxford (UK): A launchpad for tech founders, including alumni from Cohere and, ElevenLabs.

• Carnegie Mellon University (USA): Specialises in AI and robotics, contributing to founders in the AI industry like those from Shield AI.

• University of Cambridge (UK): Has a history of innovation, with connections to AI startups and companies like DeepMind and, ElevenLabs.

Where are these unicorns headquartered?

Based on the 144 unicorns featured in the market map, 64% are based in the U.S, followed by 14% in China and 4% in the U.K.

Insights:

Subscribe now

🔑 New Economies Most Popular Editions:

• AI Agents: The Ultimate Guide (30k reads within 48 hours)

• The Startup Tech Stack

• The Loneliness Landscape

• Read more about AI at the New Economies Newsletter

Read more

Hey Everyone,

With OpenAI raising even more funds and Nvidia’s profits growing at an as astounding rate, it’s never been more important to understand and evaluate AI risks. That’s why I think the AI …

Read more

Hey Everyone,

Recently there’s been a great debate about if BigTech’s investment in Generative AI is warranted. I asked the very talented about his take on the product space at the intersection of Generative AI. How can builders and entrepreneurs think this through.

• There are so many interesting Generative AI use cases and products.

• Builders are scrambling to build Generative AI tools, including many people who work in product at technology companies.

Emergent Product

Emergent Product Newsletter really stands out to me. Here’s is how Adam introduces it:

• A place to learn about product challenges and solutions from the best emerging tech product leaders. Listen to below:

Emergent - Product Newsletter and Podcast

A place to learn about product challenges and solutions from the best emerging tech product leaders. Includes a biweekly roundup of the best product thinking on the Internet.

By Adam Judelson

Subscribe to EMP

• This post unlocks on Saturday, July 27th, at 7AM EST to all readers and remains free for a few days.

Why this article might be valuable? Adam has a fairly unique corporate background. Adam spent seven years at Palantir, where he led product for their flagship Gotham data platform. After Palantir, Adam went on to lead product at Slingshot Aerospace, built a startup for Deloitte, and then served as president at mePrism, where he grew their data privacy marketplace from zero to 100 million data points protected.

• Are you interested in product innovation? 🛠️

Best Performing Posts Written By Adam

1. The unconventional Palantir principles that catalyzed a generation of startups

2. Your demo needs a prestige

3. How valuable is your product, and is it worth it?

4. Fixing a bad feature deprecation email

5. Developing Strong Product Judgment

Adam’s perspective is founded on experience. Adam is also a two-time founder and advisor to businesses ranging from scrappy drone startups to space and satellite companies to large private-equity-backed firms and public companies. These days, he runs First Principles, a product agency focused on helping companies achieve their world-changing visions.

• Product in the era of Generative AI has become far more interesting.

• Whether you are an investor, builder or enthusiast, these insights are fairly specific and special.

• 🎙️ Audio version can be found below.

Listen to Emergent Product Podcast

Read more

Age of the Sovereign AI Startup? ☀️

Hey Everyone,

Do I have a treat for you today? I’m obsessed with this idea that perhaps AI can create new kinds of entrepreneurs and scalable companies. I asked Marius to help me explore this topic of how startups themselves might be evolving in an age of Generative AI. , born in 1993 in Düsseldorf, Germany, is an inquisitive entrepreneur, investment analyst and technology enthusiast with a penchant for asking thought-provoking questions.

If you enjoy the topics and deep dives we cover in this Newsletter, consider becoming a paid supporter to get access to the exclusive content.

Subscribe now

I absolutely love what Marius has come up with for this guest contribution. I really admire young writers building new ways of looking at the world and asking important questions while doing the necessary deep dives around important questions.

If you were a young person today, what important questions would you be asking with regards to AI?

Around three months ago he re-launched/started this fascinating Newsletter called Skywert Analysis.

Skywert Analysis

The Multidisciplinary Lens for Navigating Exponential Change

By Marius Schober

Introduction to Skywert and Marius 🔥

• The Looming Collapse: America's $34 Trillion Debt Spiral

• The Future Trend of Longevity

• The Impending Workforce Disruption

🌟 In partnership with Gretel📈

The Definitive Guide to Synthetic Data

Dive into the key benefits and technologies behind synthetic data generation with our comprehensive guide.

Download

Introduction to his Newsletter: 🎙️

Subscribe now

• This post will be made available to free subscribers for a limited time on Saturday July 20th. The audio version available below.

Read more