Some recent conversations have reminded me that I have opinions about data and its use inside organisations. Especially for marketing-type stuff.

Here are three of those opinions

Gather less of it

A few years ago in a sales meeting some ad-tech person said 'I bet you wish you had more data on your customers' and the my perpetually contrary inner voice said 'oh no I don't, I wish I had far less'. I may have actually said it. I may have said 'No I don't, I wish I had less, and if you came to me promising an effective service with far less data I might have been interested.'

That's what I'd have said now.

Because:

a. Data is a risk. Every bit of data has to be managed/looked after/cared for. That costs time and money. And most of it is useless.

b. Data is distracting. Most of it is just noise. You're gathering it because you can, just in case, because it seems valuable. Then you spend ages trying to work out what to do with it. When you should be paying attention to just a couple of bits of it and actually doing something about it.

c. It becomes a job. Get enough data and you need data scientists. Then you're stuck in a self-perpetuating structure that requires more data to feed the data scientists.

The best expression for all this I've ever seen is from James Timpson, a column in The Sunday Times towards the end of the pandemic. Here's a (long) excerpt:

"I vividly remember being shown the charts room in fund manager Fidelity’s huge London office. There were graphs of everything under the sun. Was the theatre of the chart room the big sell to clients, or was it a useful tool for the analysts? No doubt Debenhams’ bosses had lots of facts at their fingertips, but data didn’t help them save the business.

Now our shops are open again and everyone is back at the office, the data is pouring forth. We have a culture where we want to produce as little information as possible, but it can feel like watching a dripping roast, with statistics flowing from every department at an alarming rate. With 2,100 shops, there’s always lots of information to consume and my eyes can quickly glaze over.I prefer to focus on a few things, as well as the basics of retailing. Are the shops open? Is everyone happy? If so, we can start taking money.

There must be a point where the costs of interpreting and using data exceed the benefits of collecting it. Can you afford a chief data officer paid £120,000 a year plus bonus? We can’t, so instead we have three simple ways of understanding what’s going on.Every night at 7pm, I get an email listing that day’s sales. This data isn’t collected by an “Epos” (electronic point of sale) till system, but by colleagues filling out a form online. They also write the sales numbers on a piece of paper and keep it on a bulldog clip. This takes five minutes a day. It sounds old-fashioned, but when people physically write things down they seem to take more notice. If you ask our colleagues what their sales are so far today, I bet they’ll know to within the nearest £50.

Over the past 25 years, we have acquired a number of (loss-making) retail chains. The first thing we do is switch off their Epos tills. All we need is a drawer to keep the cash in and a calculator to add up the sales. We have thrown away more than £8 million of kit — and it’s made life easier for us.The businesses we bought were often collecting vast amounts of data from their fancy tills, yet the managers were actually reading very little of it, and it rarely helped colleagues give better customer service. As sales plummeted, they analysed more data, and brought in more finance experts and consultants to work out where the problems were. Redundancies weren’t made from the data team — it was the people on the front line, serving customers, who lost their jobs first. These companies failed because they lost focus on what’s important: great customer service.

So our second barometer is customer service scores, which I look at every day. We ask customers to use an online form to rate their experience out of ten (our average score last week was 9.4). Every colleague sees their feedback in real time, and if we get a bad score our area managers are expected to call the unhappy customer straight away to apologise and fix the problem.

One piece of data beats everything else. A quarter of a century ago, my dad taught me the best way to measure the health of our business was to look at the cash figure every day. Each morning at 10am, I get an email from Caroline in the finance team showing the cash we have in the bank compared with the same day last year. This fact offers no hiding place."

Here's the whole thing

Keep it in your hands

Data is most useful when it's in the hands of the teams who create it or need it. The more it gets abstracted away to other teams and other softwares the more dangerous and misleading it gets.

So start off with writing it on pieces of paper or sticking it on the wall. Graduate to spreadsheets only when you have to. Move on from spreadsheets very, very reluctantly. Dashboards are dangerous. Everyone knows the stories about pilots flying into the ground while staring at their instruments. Dashboards abstract away the reality.

The trick is to keep the data in your hands. Get it from the source yourselves, regularly, daily, weekly and copy it into your spreadsheets then get together and talk about what you're seeing. Yes, you might have transcription errors but you should catch them because you know the data directly.

You know, because you've been sticking it in a spreadsheet every week, how many subscribers you have. Or whatever. That's different to seeing it go green on a dashboard or seeing the lines on a pie chart move.

This has the additional advantage of matching the fidelity of the presentation to the quality of the data. When you don't have much data - and therefore don't know much - then keep it scratchy and on paper. It might look less whizzy but it reminds you of the uncertainty. There's a massive risk in taking the tiny amounts of data that startups have and pasting it into fancy dashboards and vibe coded analytics. You start forgetting you've got a tiny sample size.

Translate to human

I used to have regular rows with engineers who told me that various things they'd built worked for 99% of our customers and were therefore ready to go. But, I'd say, we've got two million customers, so twenty thousand people are about to be massively inconvenienced and most of them will phone us.

You have to think through the data to the people-sized reality.

I find two things help with that:

1. How many Wembley stadiums?

Numbers of people are hard to visualise. It helps if you think of things you've actually seen. Like 'that's the same number of people who can fit in Wembley stadium'. You might realise that a number is bigger, or smaller than you thought.

2. Talk through the reality

Check the data by imagining the story behind it. Say, out loud, in your data meeting, what you think might be happening. So, if you've changed something on your emails and the click-through rate is going down then talk it though 'I think this means that people aren't sure what they'll get when they click so they're reluctant to do it. Does that make sense?' It doesn't have to be right, it just has to be plausible. Because if you can't think of a plausible explanation for what's happened you need to revisit the data or check some assumptions.

Thanks to Sam Graf for introducing me to this and suggesting some toys.

Introduction

Multiplication tables can be fun. Line up your numbers, multiply, and find patterns. Like with 5x5, we can fill it out and highlight symmetry, divisibility, squares, and so much more. In this inquiry, we're going to play with a different version of these tables.

Starting with Six

Take that 5x5 multiplication table and divide each number by 6, and write down the remainder. Another way to say this is mod 6.

What do you notice?

What patterns emerge in the table?

What numbers produce zeros?

What numbers don't produce zeros?

Make a table for the numbers that don't produce zeros, and apply mod 6 as before. This is called a Cayley table

For each number in the table, look at its powers mod 6 and note any observations:

Here is 5:

Activity

Now, let's take what we did with six and repeat it with other numbers. Let's call each number we pick m.

For each m:

• Make a table that goes from 0 to (m-1)
  • Note that zero was added in for completeness, but not required.
• Then a Cayley table of the numbers that don't generate zeros
• Then look at the powers.

Here is a tool (full page on desktop is best). It is too big to embed here, so save it and read on.

Here is six using that tool:

Here is nine. The size or number of rows in the Cayley table is Euler's totient. For nine it is six, written as φ(9)=6.

For different values of m

• Which numbers make you stop and look?
• Which ones feel more structured?
• Which ones are alike? different?
• Which numbers don't change when you hide zeros?
• Which numbers lose a lot of numbers when you hide zeros?
• How many different numbers are there for each row or column in the Cayley table?

Conjecture

Form conjectures about the tables.

• Do certain numbers result in certain patterns? certain Cayley tables?
• What is the maximum number of different values a row can have in the Cayley table?
• What numbers in the Cayley table produce all of the numbers in the table with their powers mod m?
• Are there certain numbers that you can always expect to see in a Cayley table?

Activity Structure

This is a 60–90 minute activity exploring the multiplicative structure of integers mod m.

Exploration Phase 1 (10–15 minutes)

Building the first tables

Have learners build the mod 6 multiplication table by hand. The hand-work matters — patterns surface faster when learners feel the symmetry and notice the zeros or lack thereof.

• Ask: "Which rows or columns have zeros? Which don't?"
• Once they strike out the rows/columns with zeros, the leftover 2×2 Cayley table for {1, 5} is small enough to stare at and ask, "What is this thing?"
• Look at the powers - does 1,5,1,5,1,5 continue to repeat? Why?

Exploration Phase 2 (15–20 minutes)

Comparing several values of m

Some useful values to start with:

• A prime: m = 5 or 7
• A prime power: m = 8 or 9
• A product of distinct primes: m = 10 or 15

Here is the tool — full page on desktop is best.

Groups or learners can take values and then trade.

Conjecture Formation (10–15 minutes)

Give time to write down observations before discussing. Offer examples if learners stall.

Example Conjectures:

Example: "When m is prime, no rows or columns get hidden after the zero row and column."

Example: "The numbers left after hiding zeros are exactly the numbers that share no factors with m."

Example: "Every row of the Cayley table has the same set of numbers, just rearranged."

Example: "Sometimes one number's powers generate all the numbers in the table."

Example: "Every Cayley table has 1 and m-1."

Example: "Every Cayley table is a Latin Square."

Supporting Questions:

• "What do the m values with no zeros have in common?"
• "How could you predict how many numbers survive hiding zeros, without building the table?"
• "Why does every row in the Cayley table seem to have each number exactly once?"
• "For which m does some number's powers produce all the others?"

Discussion and Discovery (15–20 minutes)

• Share conjectures across groups.
• Introduce terminology as it becomes useful:
  • The surviving numbers are called units mod m.
  • The count of units is Euler's totient, φ(m).
  • A Cayley table whose row/column entries are a permutation of the same set is a Latin square.

Going deeper (optional)

The content below is what you might find in a textbook, and possibly too heavy for light inquiry.

Do the powers always cycle? 

• Pick a number from the Cayley table — call it n — and list n, n², n³, … mod m.
• There are only finitely many remainders possible, so the sequence eventually repeats.
• For any number in the Cayley table, it always cycles back to 1.
• The smallest power that hits 1 is called the order of n.

How long is the cycle? 

• Compare cycle lengths across numbers in the Cayley table.
• They always divide φ(m) — the number of rows in the Cayley table.
  • This is Lagrange's theorem.

When does one number's powers produce all the others? 

• When the cycle length equals φ(m), that single number's powers fill the entire Cayley table.
• It's called a generator or primitive root.
• These exist exactly when m = 1, 2, 4, pᵏ, or 2pᵏ for odd prime p — so mod 8, 12, 15 have none.
• A group where this happens is called a cyclic group.

Optional: Proof scaffolding

Powers of a Cayley table number (unit) cycle back to 1

• Consider mod 7, its Cayley table, and powers of 2 mod 7:

• The list goes 2, 4, 1, 2, 4, 1, … It cycles back to 1 every 3 steps.

Is this true for all numbers in the Cayley table?

1. There are only finitely many possible remainders.
   1. Mod 7, the possible remainders are 0, 1, 2, 3, 4, 5, 6 — seven values total.
   2. Every power 2¹, 2², 2³, … has to land on one of these seven.
2. Eventually, two powers share the same remainder.
   1. This is the pigeonhole principle.
   2. Ex: 2 holes and 3 pigeons means two pigeons have to share a hole.

3. Every Cayley table number n has an inverse — a number that, when multiplied by n and then taken mod m, equals 1.
   1. Example: 2's inverse mod 7 is 4, since (2 × 4)(mod 7) ≡ 1.
   2. Note that by its construction, there are no zeros in the row. 
   3. All numbers in a row are different. Why?
      1. If two entries matched — say (n × a)(mod m) ≡ (n × b)(mod m) with a > b — then (n × (a − b))(mod m) ≡ 0.
      2. But a − b is between 1 and m − 1, and the row has no zeros there.
   4. So n's row has m − 1 different, nonzero values filling m − 1 spots. They must cover every nonzero residue from 1 to m − 1 — including 1.
   5. The number b with n × b ≡ 1 (mod m) is n's inverse.
4. For any number n in the Cayley table mod m:
   1. The list n, n², n³, … has only m possible values, so two must repeat: nⁱ ≡ nʲ for some i < j.
   2. Multiplying both sides by n's inverse i times cancels the left down to 1.
   3. What's left: 1 ≡ n^(j − i) (mod m).

Note: Try 2 mod 6, where 2 isn't in the Cayley table. The powers go 2, 4, 2, 4, … The cycle never reaches 1, because 2 has no inverse mod 6. There's nothing to cancel with.

Tools and Supplies

• Grid paper for hand-built tables.
• A spreadsheet tool works well for this
• Calculator or spreadsheet for larger m.
• Units mod m tool (full page on desktop).
• Colored pencils or highlighters for marking symmetry, zeros, and cycles.

Vocabulary

• Modulo / Mod: The remainder when one number is divided by another. Example: 4 mod 3 = 1.
• Unit (mod m): A number with a multiplicative inverse mod m; equivalently, a number coprime to m.
• Cayley table: A table showing the result of a binary operation on every pair of elements in a set.
• Latin square: A square table where every row and column contains each symbol exactly once.
• Euler's totient (φ(m)): The count of integers from 1 to m that are coprime to m.
• Order of an element: The smallest positive k such that aᵏ ≡ 1 (mod m).
• Generator / Primitive root: A unit whose powers produce every unit mod m.
• Cyclic group: A group generated by a single element.
• Group: A set with an operation that has closure, associativity, identity, and invertibility.
• Lagrange's theorem: The order of any element divides the size of the group.
• Conjecture: A statement believed true but not yet proven.
• Counterexample: A specific instance that disproves a conjecture.
• Monoid: A system that has closure, associativity, and identity.
• Ring: A set with two operations (like + and ×), where + forms a commutative group, × forms a monoid, and × distributes over +. So,  a × (b + c) = (a × b) + (a × c).

Extensions and What Ifs and Resources

• Play with the concept in more dimensions: Toy for 3D is here.
• William Stein, Elementary Number Theory: Primes, Congruences, and Secrets
• Compute φ(m) for m up to 30 and look for patterns in the Cayley table sizes.
• Addition vs. multiplication - what does addition look like mod m?
• Public-key cryptography applications
• Carmichael function λ(m)
• Gauss defined primitive roots in Disquisitiones Arithmeticae (1801).
• Chords on a circle.
  • Put n evenly spaced points on a circle and connect them with rules.
  • Skip-k chords visit every point exactly when k is a unit mod n.
    • Multiplier-k chord rules send each point to a different image exactly when k is a unit mod n.
  • Both rules live in the same ring (ℤ/nℤ, +, ×) — step rules use the additive side, multiplier rules use the multiplicative side.
  • See Beautiful Chords.

Multiplicative group of integers modulo n - Wikipedia

Wikimedia Foundation, Inc.Contributors to Wikimedia projects

I like the Fits on a Floppy manifesto by Matt Sephton:

Software should be as small as it can be. Not as a gimmick, but as a discipline. The floppy disk is the measuring stick: 1.44 MB. If the software that ran entire businesses could fit in that space, then a modern, focused, single-purpose tool certainly can.

In my own work, I have mostly focused on the web side of this equation, as this is where the situation feels the most dire: tens of megabytes dedicated to heavy frameworks, unnecessary tracking scripts, and video ads have a real negative effect on experiencing websites. Progressive loading challenges also make it harder to offer a great experience.

But space considerations are starting to feel more pertinent to local software too, in an era where SSD and hard drive prices are going up, and where local LLM models start taking up more room.

Also, this passage feels very Unsung, and is exactly why the tag #history exists on this blog:

I don’t miss floppy disks. I miss the mindset they demanded—that every byte matters, that constraints breed creativity, and that software should be light on its footprint.

If you reduce tech history to just nostalgia, it won’t be that useful. But if you look at it as inspiration, you might find some truly wonderful and meaningful stuff in there.

On that note: Bonus for a nice classic domain, and a nod toward Mac’s most famous screensaver.

#history #performance

When a commencement speaker at the University of Central Florida intoned that “AI is the next industrial revolution,” she was met with a chorus of thundering boos from the graduating students in attendance. The mass disapproval left the orator, Gloria Caulfield, Vice President of Strategic Alliances for Tavistock Development Company, flustered and unmoored. She fumbled through the rest of the portion of her speech amid more jeers and exclamations of “AI sucks,” which didn’t relent until she noted that AI wasn’t a factor in our lives just a few years ago, to which the grads cheered.

The clip went viral, of course, in the now-well-stocked genre of ‘stark and unambiguous reminders that lots of people hate AI’, a genre that stretches back to early 2024, when a SXSW audience booed a pro-AI video, through Guillermo del Toro’s 2025 “fuck AI” exhortations on the press tour for Frankenstein, and onto the darker, more violent instances of AI rejection that made headlines this year.

The commencement speaker clip is a particularly striking artifact, though. It resonates particularly deeply, I think, because it reflects the generational and economic breakdown of who AI is for, and who it harms. Here is an executive with one of the most impressively generic corporate managerial titles I have encountered, blithely repeating a line about AI being “the next industrial revolution” that she has likely uttered many times in other circumstances, to hundreds of young people who have now been hearing for years about how AI is both erasing their career prospects and is the future. (By one count, the entry level job market is the worst that it’s been in nearly four decades.)

I’ve heard this industrial revolution line or a variation of it so many times from people with titles like ‘VP of Strategic Alliances’ over the last three years—in meetings, at seminars, at conferences, in personal conversation—that beyond losing count of the utterances, it’s become like white noise, the rule rather than the exception, as predictable as the warm, slightly curdled exhalation of breath of your office colleague in a late-afternoon meeting. It’s something that is said dutifully by mid-to-late career executives and managers to signal (to investors and partners, to the public, to colleagues angling for their jobs) that they understand and tacitly endorse the changes to be effectuated by AI. Almost every person who says this, it seems, has something to sell, and usually that something is bound up in the idea that AI is inevitable, and that we must all get used to the cognitive offloading, increased surveillance, and amped-up productivity demands that accepting as much entails.

Meanwhile, the graduating students, who likely either have already begun trying to find post-collegiate work, know people who have, or have at the very least seen the headlines about AI and entry level jobs and felt the bad vibes, have an eminently superior grasp of what ‘AI is the next industrial revolution’ means in practice. It means that right now, employers have decided they can hire fewer people, and for lower wages, and that they are graduating into a notably bleak economy for people like them with fresh degrees. Thus we have our chorus of revulsion and our apt demonstration of who AI is made to serve, and who it will not.

AI is after all being blamed for deskilling or even decimating a lot of industries that college graduates may well have wanted to work in: the arts, entertainment, tech, gaming. I too would loudly boo at the prospect of this next industrial revolution if I was in my early twenties, unemployed, and had aspirations for my future greater than entering prompts into an LLM. It’s yet another reminder that enthusiasm for AI tends to break along age and class position, as recent polling has demonstrated, and brings to mind that recent NBC survey that showed that Gen Z respondents (ages 18-34) gave AI a favorability rating of negative 44, while one of the only groups that found it favorable were those earning over $200,000 a year. All of this seems pretty straightforward to me, and I noted as much in my recent post about what was motivating the increasingly violent backlash to AI.

Subscribe now

In the weeks since, there have been a number of other efforts to diagnose the AI backlash. In particular, a term ‘AI populism’, coined by the writer Jasmine Sun, has been popping up, especially at the New York Times, where it’s been quoted by Ezra Klein and showed up in the headline of a piece by David Wallace-Wells1.

Here’s Sun:

I define AI populism as a worldview in which AI is viewed not only as a normal technology but as an elite political project to be resisted. It regards AI as a thing manufactured by out-of-touch billionaires and pushed onto an unwilling public to achieve sinister aims like “capitalist efficiency” (layoffs) and “population management” (surveillance). AI populists don’t really care whether ChatGPT is personally useful, or if Waymos eke out some safety gains: AI’s utility as a tool is immaterial relative to the unwelcome societal change it represents.

Sun, who describes herself as an anthropologist of disruption, uses ‘AI populism’ as a means of theorizing why the AI industry is attracting ire from both x-risk doomers and anti-data center organizers. It’s a provocative coinage. But like David Karpf, who points out that such groups have very different reasons for and methods of opposing AI, and that it’s not particularly useful to lump them together, I don’t ultimately think this a great way to think about the broader animosity percolating around AI. (For one thing, the language presents the idea as “sinister” and faintly conspiratorial, and seems to patronize those who might believe it.)

Directionally, as a tech guy might put it, it’s not wrong. There is undoubtedly anger at out-of-touch billionaires helping companies execute mass layoffs, and many people don’t think ChatGPT is useful enough to warrant the social (or economic and environmental) burdens it imposes. The problem is that Sun’s coinage aims to position AI as a project that can be considered novel, or even apart, from the political economy from which it emerged. But I don’t think most people are formulating a new worldview in which AI is a boogeyman political project hatched by billionaires. I think they’re more likely to understand AI as an extension of an already inequitable system, and as an accelerant of that inequality. At a time when consumer sentiment is stuck at all-time lows, housing costs are sky-high, the price of basic goods is spiking, entry level jobs are disappearing, tech firms have concentrated enormous power and “broligarchy” was shortlisted for Dictionary.com’s 2025 word of the year, AI has become the avatar of the ills of unrestrained capitalism. “AI populism” is really just “21st century populism” or just, “populism.”

AI has after all been adopted and promoted as an instrument of efficiency, control, and leverage by just about every layer of management at every institution, from any given Fortune 500 company to a department in the federal government to your boss who makes you use Copilot, to which one might direct their populist anger. This is less the result of a specific political project, as much as it is how capitalism tends to function when there is a new instrument to discipline workers on offer. As writers and thinkers like Ted Chiang and Hagen Blix have pointed out, fear and anger at AI are often best understood as fear and anger at how AI will function within capitalism. Few are worried about the prospect of public research scientists using LLMs to discover new peptides; plenty are worried about how AI might be used as leverage against them in their workplaces, or to replace their labor, or to narrow their job opportunities. They’re worried that AI will exacerbate existing conditions in a precarious system.

Firms have used automation technologies to impose layoffs and surveillance regimes on their workforces to achieve improved efficiencies for as long as such technologies have existed; there’s nothing sinister, or at least unusually sinister, about this. But Silicon Valley has certainly raised the stakes, in pursuit of ever-greater profits and investment capital: AI has been developed, pitched, and sold by tech firms as the most powerful automation technology of all time. As OpenAI’s charter puts it, the company is building “highly autonomous systems that outperform humans at most economically valuable work.” This declared aspiration to sell one-size-fits-all, mass deskilling-as-a-service in a destabilized, post-pandemic, post-J6 world feels in hindsight like a dependable formula for generating widespread anger.

At the dawn of the AI moment in 2022/23, AI firms promised that the technology would help solve climate change and cure cancer. While there has been little notable progress on those fronts—and in fact the energy demand of data centers has thus far moved the needle in the opposite direction, towards increasing carbon emissions—AI very much has initiated a transfer of wealth from the middle and working classes to the rich, and a mass degradation of jobs once thought widely desirable. Just this week, a TV writer published a piece in WIRED about how she had turned to selling her editorial skills to AI training companies in a wildly unregulated labor market to get by:

I never intended to write about this industry. I came to it not as a journalist but as a disgruntled, broke TV writer determined to make a dent in student loans and keep paying LA rent while my industry withered in front of me. But working with and for AI had proven even more cruel than I could have ever imagined. Mercor says it employs about 300 full-time staffers. Meanwhile, each week it keeps some 30,000 independent contractors caught up in a fever dream of aimless, directionless urgency, corralled across Slack channels by achingly young adults, sending messages at 3 am to “push on” and “finish strong” and “lock in” and “Go Team GO!” All in service of the grandest purpose in history: to successfully remove a scuba diver from a picture with one click of a mouse, transport him to the moon without any glaring artifacts—and bring him back again.

In fact every week seems to bring new stories of mass job loss that the companies are attributing to AI: Meta, Disney, and Cloudflare are just some of the latest. I’ve chronicled scores of stories about lost work, careers, and income security in AI Killed My Job. And a brand new study shows artists report declining income and opportunities.

And who’s winning, in the current environment? Certainly, the executives of tech firms. Over the last couple of weeks we’ve learned through the Musk vs. OpenAI trial, that co-founder Greg Brockman’s stake in the onetime nonprofit started to benefit all of humanity is worth $30 billion, and Ilya Sutskever, no longer with the company, has a cut worth $7 billion. Other winners of the AI era? Guys like Matt Gallagher, who created the first “one-person billion dollar company.” His company is an entirely automated version of Hims called Medvi, a digital middle-man business of selling health supplements online that apparently relies on fake, presumably AI-generated doctor accounts to hock GLP-1, falsified, AI-generated patient testimonials. It’s the target of an ongoing class action lawsuit, has been formally warned by the FDA for misbranding violations, and more.

Who else? Well, people who can weaponize AI to game prediction markets, leading to a situation where, at sites like Polymarket, 0.1% of the accounts capture two-thirds of all the gains, as the Wall Street Journal reported this week. (Title: “Why Everyone Loses—Except a Few Sharks—at Prediction Markets.”) This is, as analyst Paul Kedrosky notes “partly a function of their nature, but also of vibe-coding script kiddies attacking every market anomaly as quickly as it arises.”

He continues:

“The same dynamic is now spreading across retail-dominated markets. A driver is how AI lowers the cost of systematic exploitation and exploration to near zero. What used to require infrastructure, data pipelines, and bearded quants is now accessible via off-the-shelf models, APIs, and loosely stitched “agent” workflows doing ... stuff that even their users don’t fully understand.

The result isn’t democratization of returns. It is wider participation, of a sort, alongside the rapid re-concentration of profits. A small subset of users—those willing to iterate fastest, monitor continuously, and deploy capital programmatically—capture gains, with everyone else just liquidity…

…Prediction markets are simply the cleanest expression of this trend because they combine thin liquidity, discrete outcomes, and high retail participation. But the same pattern is visible in options flow, single-stock volatility events, and even online poker, which AI increasingly dominates.

As AI tools continue to scale, expect this to get worse: a small cohort running semi-automated strategies extracting semi-consistent edge, and a much larger base supplying them returns. Under the pressure of AI prevalance, markets don’t flatten, the return gradient steepens to a cliff.

So we have AI looming over our withering creative industries, a generation of young people who are angry and disillusioned by the lack of opportunities, and precarity and anxiety nearly everywhere. In exchange, we get a new batch of tech oligarchs, new shady billion-dollar businesses that employ no one at all and use AI to evade consumer protection laws—that pretty unequivocally leave the world worse off in the wake of the founder’s mad dash to personal enrichment—and new tools for the unscrupulous to accumulate wealth at the expense of those still following the rules, whether in stock trading, prediction markets, or even online poker. That and Claude Code.

That’s why the students are booing, I think.2 They’re experiencing AI in realtime as a forecloser of futures; as the cruel new face of hyper-scaling capitalism, as the prime agent moving a world that’s become a deck stacked against them.

Subscribe now

BLOODY MEDIA HITS

I sat down to chat with Taylor Lorenz about the AI backlash, and for a little friendly debate:

A bit back, I recorded a podcast with Thomas Dekeyser for the University of Minnesota Press for the release of his book; it’s out now. Give it a listen here:

CHART OF THE WEEK

BLOODY GOOD READS

and Saul Levin on organizing and the anti-data center movement the Guardian:

As usual, ordinary people are ahead of their leaders. The remarkable organic growth of the datacenter resistance movement across geographies, economic interests and ideology reflects the myriad harms that come with AI infrastructure and growing anger at the tech elite. The tremendous energy unleashed by these fights, and their sensible and unifying demands, have the potential to form the foundations of a new and powerful populist coalition, one poised to help define a working-class agenda that meets this moment and resonates with disaffected voters. This excellent organizing should be cultivated rather than dismissed.

Alondra Nelson has a new paper in Science, The Civic Grammar for AI Rights, that’s worth a look:

Some commentary has argued that AI companies reaching for the vocabulary of constitutional democracy are attempting to fill a vacuum that democratic institutions can no longer hold. That reading is not wrong about the vacuum. But it is incomplete about which actors have moved to fill it. Not only technology companies, but legislators, civil society organizations, and the constituents they represent have produced a civic grammar: a shared set of rights claims that publics can extend to new institutions and new harms, and that has been traveling across jurisdictions, partisan lines, and institutional contexts.

The great novelist and playwright Ishmael Reed is working on a new play, called, wait for it, “King Ludd’s Revenge”.

From the NYT:

Mr. Reed, a novelist, playwright and provocateur who has been upsetting opinions across the political spectrum for at least six decades, is aiming high with a new drama. “King Ludd’s Revenge” is a rare attempt to take on the tech moguls with something more than mere journalism.

“Instead of a straight narrative, I improvise,” the 88-year-old writer said. “It’s like Louis Armstrong singing ‘Stardust.’ He doesn’t do it the way it’s written.”

Oakland is poorer, Blacker and more maligned than San Francisco and Silicon Valley, both of which are just across the bridges that span the Bay. Having the trial here happened at random — Mr. Musk’s lawsuit against Mr. Altman and the company they founded together, OpenAI, was filed in San Francisco and assigned to the federal court in Oakland — but feels a little like one of those episodes where the Greek gods descend to mundane Earth to settle a dispute.

Mr. Reed, an Oakland resident who has celebrated and defended the city for decades, may be the only one in town noticing who’s here. “Everybody’s focused on the N.B.A. playoffs,” he explained.

“King Ludd’s Revenge” takes its title from the legendary leader of the workers’ revolt in England in the early 19th century. With the ascent of A.I., the Luddites have come back into fashion. The play begins with Mr. Musk receiving a pedicure from a robot. Peter Thiel, the tech billionaire who backed President Trump in 2016, bursts into the room. “I think I’ve identified the leader of the Anti-Christ Syndicate,” he says.

Mr. Musk: “Who might that be?”

Mr. Thiel: “Greta Thunberg.”

Karen Hao on the Elon Musk vs Sam Altman courtroom drama:

…[F]ixating on questions of whether Altman is untrustworthy, or whether Musk is even less so distracts from a far deeper problem. If OpenAI lost its footing as the AI industry frontrunner, another barely distinguishable competitor – Musk’s xAI or other – would simply replace it. That includes companies like Anthropic, who enjoy a better reputation yet engage in many similar behaviors like compromising careful decision-making for speed, disregarding intellectual property, and aggressively scaling their computing infrastructure to the detriment of communities.

Nothing about this trial or OpenAI’s financial structure will change the imperial drive of these companies to consolidate ever-more data and capital, terraform the Earth, exhaust and displace labor, and embed themselves deep within the state to gain leverage over its apparatuses of violence. We would still exist in a world in which a tiny few have the profound power to cast it in their image and dictate how billions of people live.

Alright alright, that’s it for now. Have a good weekend everyone, and apologies if that was all a bit punchy, I’ve been fighting a cold all week. Until next time—hammers up.

Subscribe now

This is the most efficient way to pack 10 unit squares into a surrounding square.

It looks like something I would turn in after running out of time.