Hoboken Historical Museum

Hudson Yards as seen from Hoboken North

The Collapse of Substack: A content-creator talks about money [Scott Carney]

00:00 The False Promise of Substack
05:14 The "Bait and Switch"
10:20 The Reality of Subscription Fatigue
15:35 YouTube Finances & Transparency
18:56 The Shift to Sponsorships

Those with programming skills are best at vibe coding

China published the most uncomfortable paper on vibe coding.

ETH Zurich tested 100 developers in a controlled, commercial-grade vibe coding environment to see who actually succeeds.

The findings are brutal.

The researchers tracked computer science achievement, written… pic.twitter.com/hVKdE4GQGD

— Superman (@thesupermanmx) September 5, 2026

From the tweet:

The hype told us that learning to code is dead because language is all you need.

The data just proved the opposite.

To truly master the vibe, you still need to understand how the machine thinks.

Here's the paper, Computer Science Achievement and Writing Skills Predict Vibe Coding Proficiency.

Friday Fotos: Exotic flowers

Little blue flowers

The importance of work and the threat of AI

Ruxandra Teslo, Curing cancer won’t redeem AI, Ruxandra's Substack, Sept. 1, 2026.

A few weeks ago, a poll result began circulating on X: Americans now appear to be more hostile to the construction of data centers than to the construction of nuclear power plants. This is striking because for decades, the nuclear reactor has been the archetypal object of anti-abundance politics.

Looking at how nuclear power plants have become so hated is instructive. On many objective grounds, they are perfectly fine (actually, they are glorious): a low-cost, low-carbon emission form of energy. But a big part of what made nuclear power plants so hated was, of course, not based on empirical facts, but on the imaginative horror surrounding them. The association with the Cold War and nuclear bombs. The nuclear accidents: a great explosion, scorched and poisoned landscapes, mutant creatures lingering decades after the event, wandering through the ruins of the dystopian landscape. Nuclear power accumulated an entire symbolic vocabulary of fear that went far beyond its concrete risks and outshined its benefits.

On the face of it, the concerns around data centers are largely practical and they include, most prominently, worries over water usage or electricity consumption. Andy Masley and others have done great work rebutting some of these worries from a factual perspective; for example, in articles like “The AI water issue is fake”. But I suspect that debating these concerns at the object-level misses the deeper point. The practical objections to data centers often seem less like the real source of the hostility than respectable vessels for a more diffuse fear. Vague anxiety is hard to argue from, so the imagination gives it a body: water use, electricity demand and noise. Like nuclear plants before them, data centers are acquiring a mythology of their own: vast, windowless avatars of a deeper fear that the machines built inside them may one day render human beings unnecessary.

“Rendering humans unnecessary” is broad enough and it encompasses many different fears, so I want to focus on one of the most concrete instantiations of it: the fear of job loss. Anthropic’s 2026 Public Record survey of nearly 52,000 Americans revealed that 64% of the respondents said they were worried about A.I.-induced job loss, making it the single most common fear in every state. The words “job loss” might make it sound mostly about income, but I think this understates what’s at stake for most of the population. For work provides not only income, but agency, status, belonging and a sense of control over one’s life. So what underpins the fear here is not simply losing one’s paycheck, but rather the entire social role through which one has come to understand oneself. Existing polls support this interpretation. When Americans are asked whether they would prefer the response to A.I. to consist of creating good-paying jobs or providing direct income support via governmental programs, they overwhelmingly choose the jobs.

There’s much more at the link.

Citibikes on 14th St. in Hoboken

Respiration waveforms are closely coupled with the shape of neural oscillations

Eena Kosik-Rose, Guangyu Zhou, Andrew Sheriff, Joshua M. Rosenow, Stephan U. Schuele, Chima O. Oluigbo, Saige Anabel Teti, Mohamad Koubeissi, Md Rakibul Mowla, Ariane E. Rhone, Sukhbinder Kumar, Brian Dlouhy, Christina Zelano, Bradley Voytek, Cycle-by-cycle respiration waveforms are coupled with the shape of neural oscillations, Journal of Neuroscience 31 August 2026, e0731262026; DOI: 10.1523/JNEUROSCI.0731-26.2026

Abstract

Beyond sustaining life, breathing is a vital physiological rhythm that shapes cognition, perception, emotional regulation, and mental health. Breathing has a direct effect on neuronal excitability and is coupled to neural oscillations across a variety of brain regions. Notably, both respiration and neural oscillations are asymmetric and not perfectly rhythmic: for example, every breath has a different shape and duration, and is interspersed with variable pauses. Here, we examined the coupling between breathing and the brain by quantifying the nonsinusoidal features of each breath and comparing it to the shape of each corresponding neural oscillation cycle. By leveraging invasive human brain recordings from 16 participants (8 female, 8 male), we found respiration-neural waveform coupling on a breath-by-breath, cycle-by-cycle basis across limbic and cortical forebrain regions. For decades, the dominant perspective on cognition and mental health have focused on the brain, but recent work is highlighting the importance of brain-body interactions. Our results show that the coupling between breathing and neural activity is much richer than previously appreciated, and our approach opens new avenues for studying these peripheral-to-central nervous system interactions in a more robust, temporally precise manner.

Significance Statement

Breathing shapes brain activity, but prior work has characterized this coupling by aggregating across many breath cycles, leaving the fine-grained shape of individual breaths unexamined. Here, we show that the precise waveform shape of each breath is coupled to the shape of corresponding neural oscillation cycles in the human forebrain, on a breath-by-breath basis. Using invasive brain recordings from 16 epilepsy patients, we demonstrate that temporal and amplitude features of individual breaths are linked to the morphology of neural oscillations in limbic and cortical regions. This cycle-by-cycle respiratory-neural coupling reveals a richer and more temporally precise relationship between breathing and brain activity than previously appreciated.

Approaching Jersey City (August 2007)

Making it in the music business [Mary Spender]

YouTube page:

Mary Spender explores the role of privilege and economic background in the music industry. By analyzing how access to time and resources influences artistic development, this discussion examines the challenges independent musicians face when trying to build sustainable careers without established support systems.

Chapters
00:00 The Gift of Time
04:16 What Money Actually Buys
09:10 What I Actually Had
14:53 Envy
19:26 Was It Really Easier Before?
22:12 The Starving Artist Myth
25:10 What Does Success Actually Look Like?
27:55 Redefining Success
29:54 Stay in the Game
34:49 What Privilege Really Means

Square rigger on the Hudson

Patterns of light through a glass

The diminishing status of science and the academy

Jessica Hullman, What stories should we tell about science now? Statistical Modeling, Causal Inference, and Social Science, August 28, 2026.

The opening paragraphs:

This is Jessica. Like many academics, I am concerned about what sort of new steady state U.S. universities will find themselves in after the dust settles on recent transitions. Namely, the last few years have brought funding cuts, targeted visa policy, reduced demand for grad degrees, and a general brain drain to industry (particularly noticeable in AI and computer science). It’s disorienting to think that academia has already peaked, and that the prestige ranking of the R1 faculty job over the top industry research positions (at least in computer science) might be inverting. But things feel very different than they did even a year ago. The reality of there being less money available to pay for basic aspects of research really started to hit me in the last six months. Post-covid, working on campus became less lively, but now it also feels like our collective attention is anxiously focused on Silicon Valley or Washington D.C. We hold faculty meetings where we discuss things like, Is there any way we can help local faculty members who were laid off from tenure track jobs? How will we ensure we can fund all of our own PhDs, given that TA quotas stay fixed but faculty are running out of funding runway?

To some, this is an overdue rebalancing. Nate Silver, for example, calls getting a PhD a “much worse value proposition than 20 years ago”, and predicts that elite higher ed will become “~50% less relevant in the new steady state,” which is in his eyes a good recalibration.

But it’s worth reflecting on what is lost exactly, if this dwindling of minds and resources continues. How should we think about the value of what universities provide over industry, like intellectual autonomy, or training on how to think scientifically? As a professor, I could make a list of the things that have kept me in academia–being free to work on the problems I find most important, the diversity of topics I can work on at any given time, grad students who care about doing deep work, having time to think about the best solution to a problem. But at an aggregate level, it’s less clear what the equation is.

The article then goes on to sketch out a number of specific issues. The first: “In search of the mysterious fruits of basic science.” Thus:

I’ve been reading the work of philosopher Heather Douglas, who has traced and critiqued the basic versus applied science distinction, the linear model as justification, and the idea of scientific freedom as limited social responsibility (see, e.g., here and here, or her book on the value-free ideal). Popularized by Vannevar Bush after WWII, in a report prepared for President Roosevelt, basic science is a reframing of pure science, presented as “scientific capital,” providing the principles and conceptions to power new products and processes years into the future. Bush called for deliberate policy to guard against the otherwise inevitable scenario where applied science drives out the pure. One of the eventual outcomes of his report was the creation of the NSF.

But despite the pragmatic nature of basic science espoused by Bush, as a derivation of pure science, it is hard to separate from less tangible values. One is that scientific understanding is a good outside of practical application, at both the individual and societal level. The earliest advocates of pure science associated it with being closer to God. Post-Enlightenment, this view gave way to a more secular superiority complex, which implied the strong character of the pure scientist, who chose to eschew wealth. “The highest occupation of mankind”, Henry Rowland called it in his Gilded Age era essay, “A Plea for Pure Science,” which bemoaned the vulgarity of attributing scientific greatness to the applied scientist rather than the pure.

One specific case: AI:

If we take our intuitions from the linear model, we might protest that innovation will suffer if universities’ research purposes are deprioritized. The post WWII science-industrial complex expanded the presence of basic research in industry, but studies suggest that the knowledge generating role of corporate R&D has been on the decline for years. To the extent that basic research is the supplier of downstream applications, it would seem we need universities more than ever.

But the distinction between basic and applied science that’s become synonymous with how we envision science has never been airtight. Critics questioned how an institution could be built around a distinction that seemed to amount to little more than a difference in intention, since applied research sometimes produced important new general knowledge, and pure science contributions sometimes had direct applicability.

AI research is a recent example. Not only is serious money being made without necessarily requiring advanced degrees, research positions do not require PhDs. By some accounts, passing 30 years old puts one in the older demographic of researchers at frontier AI companies. Yet much of the visible innovation in frontier model development has been heavily concentrated in industry labs, including transformer models, scaling laws, and AlphaFold.

Of course, AI owes much to academia. The amazing thing about deep learning and LLMs, to anyone who was paying attention to NLP before these developments, is that after many years of AI research contributing interesting questions but lackluster results, the technology finally seemed to work. Would we have had the foundations for deep learning if perceptrons had not been stubbornly pursued by academics like Frank Rosenblatt at Cornell early on, picked up again in the 1980s by Rumelhart and McClelland’s Parallel Distributed Processing group, despite multiple periods during which consensus said connectionist approaches were unlikely to pay off?

Other issues taken up: “Indulgence, autonomy, and social responsibility,” and “The problem with defining progress as prediction and control.” There’s much more at the link, including fairly extensive commentary.

The impact of Google AI on queries for Wikipedia

Khosravi, Mehrzad and Yoganarasimhan, Hema, Impact of AI Search Summaries on Website Traffic: Evidence from Google AI Overviews and Wikipedia (January 30, 2026). Available at SSRN: https://ssrn.com/abstract=6164926

Abstract: Search engines increasingly display AI-generated answers above organic links, potentially displacing traffic to upstream publishers. We estimate the impact of Google’s AI Overviews (AIO) on Wikipedia’s search traffic using AIO’s staggered geographic rollout and Wikipedia’s multilingual structure. Our difference-in-differences design compares monthly external-search referrals to English Wikipedia articles with referrals to the same articles in German and French, and finds that default AIO availability reduced English search traffic by 5.45% and 4.82%, respectively. Our results suggest that answer-producing digital intermediaries can materially reallocate attention away from informational publishers, with implications for content monetization, search platform design, and policy.

Face Masks Required

The Umwelt Representation Hypothesis: Rethinking Universality

Bosch, Victoria, Sommers, Rowan P., Doerig, Adrien, Kietzmann, Tim C., The Umwelt Representation Hypothesis: rethinking Universality, Trends in Cognitive Sciences, Aug. 8 ,2026, doi: 10.1016/j.tics.2026.07.004

Highlights: Artificial neural networks are increasingly used to study neural representations.

Recent studies have found a surprising degree of representational alignment between artificial neural networks and biological brains. In addition, broad alignment among ANNs of different modalities and architectures has been observed.

These findings have been interpreted as evidence for Universality: the idea that sufficiently capable systems all converge on a single shared representation of reality.

This interpretation has large implications for model comparisons that require meaningful differences between artificial neural networks: if all artificial neural networks eventually align with the brain, then creating new model classes leads to no new insights.

This opinion article argues that this inference is too strong and proposes the Umwelt Representation Hypothesis: representational alignment emerges from overlap in ecological constraints, not convergence to one privileged world model.

Abstract: Artificial neural networks are increasingly used to study neural representations.Recent studies have found a surprising degree of representational alignment between artificial neural networks and biological brains. In addition, broad alignment among ANNs of different modalities and architectures has been observed.These findings have been interpreted as evidence for Universality: the idea that sufficiently capable systems all converge on a single shared representation of reality.This interpretation has large implications for model comparisons that require meaningful differences between artificial neural networks: if all artificial neural networks eventually align with the brain, then creating new model classes leads to no new insights.This opinion article argues that this inference is too strong and proposes the Umwelt Representation Hypothesis: representational alignment emerges from overlap in ecological constraints, not convergence to one privileged world model.

AIs are not very good at long horizon tasks

This paper is a brutal reality check for long-horizon AI. Give an agent a year of interconnected decisions, delayed feedback, and consequences from its own past actions, and its performance collapses relative to humans.

The researchers tested eight leading models, including… https://t.co/rGBfuPZEtv pic.twitter.com/6CPvVIQCQX

— Rohan Paul (@rohanpaul_ai) August 28, 2026

Friday Fotos: Newport Mall

Emergent Multiscale Organisation of Neural Dynamics

Milinkovic, B., Seth, A.K., Barnett, L., Carter, O., & Andrillon, T. (2026). Emergent Multiscale Organisation of Neural Dynamics Fragments in Anaesthesia. Imaging Neuroscience, Advance Publication. https://doi.org/10.1162/IMAG.a.1364

Abstract: Conscious experience depends on the coordinated activity of neural processes that span multiple scales: from synapses to whole-brain dynamics. A recently introduced measure, dynamical independence (DI), identifies, characterises, and quantifies these multi-scale relationships using an information-theoretic dimensionality reduction approach. Here, we use DI to examine changes in the emergent dynamical organisation in the human brain under three pharmacologically-distinct anaesthetic interventions (propofol, xenon, ketamine). Applied to source-reconstructed electroencephalography (EEG), our analysis reveals that propofol and xenon, anaesthetics that abolish conscious report, exhibit more emergent but highly variable dynamic structure, indicating fragmented macroscopic dynamical organisation. Ketamine, which preserves dream-like phenomenology, shows a different pattern relative to wakefulness: reduced overall emergence yet a partial preservation of the macroscopic structure. Further exploratory analyses revealed spatially localised source-level contributions to emergent dynamical structure, highlighting regional variations. Together, our results highlight drug-induced reconfigurations of emergent dynamical structure relative to wakefulness, dissociate the amount of emergence from the organisation of emergent dynamics, and caution against equating emergence with level of consciousness. Consequently, we suggest that wakeful conscious processing depends not only on integration between neural components within a single scale, but also on integration across scales, broadening currently held assumptions of putative signatures of consciousness.

Dan Everett on Joe Rogan

The sun blazes in, and then weeds

Fuse Powder BOOM! How California technocults created opportunties for capital seeking investment vehicles

David Hoyt, AI is the Latest and Best-Financed Cult to Emerge from California, 3 Quarks Daily, Aug. 23, 2026.

The article begins by discussing the cultish behavior surrounding AI in Silicon Valley; I discussed similar material in a 3QD article from 2022, On the Cult of AI Doom. The article pivots toward finances with this paragraph:

What makes the cult of AI different in tone and in worldview is its origin within a tech industry which sits at an epicenter of unregulated flows of global capital, its profound, longstanding, and growing links to the US security state, and the certainty that on the basis of these pillars it can build the future it wants here on earth right now, and that no one should be able to stop them.

Then we have:

Two macroeconomic factors are at play in the breakneck speed with which AI infrastructure has been financed and built out since the release of Chat GPT in November of 2022. They have little to do with the intrinsic commercial potential of AI-related products or services, nor any particular innovation in computer engineering. The first, to put it simply, is the availability of lots of money. More than ever before, giant pots of money are sloshing around the planet, motivated by historically low interest rates over a relatively long period to search out the Next Big Thing. This money is mostly exempt from national regimes of control or taxation, is managed outside the traditional banking system, and is therefore more free to move around. The second is the chronic and by now widely acknowledged slow-down of economic growth in the advanced economies, dating back nearly half a century. What the hype typically fails to register is that investors don’t seek out technology and AI firms because they are the best bets on future returns – they seek them out because they are the only bets around.

To understand just why this is, Nick Srnicek provides a useful capsule history of the sequence of financial crises that have followed the liberalization of capital since the later 20th century. It is a story that leads with a crescendo to the current boom in AI development. In the mid 1990’s, the hottest global market was not centered in Silicon Valley, but in the small economies of East and Southeast Asia, the “Asian Tigers.” Only when these blew up beginning in 1997 did newly mobile capital, recently freed from decades of local restrictions on its movement in and out of national economies, begin to pour into Silicon Valley in what would become known as the dot-com bubble. When this blew up, as all bubbles do, it led to a decade of low interest rates dictated by the Federal Reserve. This, in turn, led global investors to flood the US housing market, inflating a nearly decade-long bubble in real estate prices. When this bubble blew up in 2008 with even more severity, followed almost immediately by the Greek/European debt crisis of 2009-2012, the solvency of major financial institutions in the US and Europe and of the entire economic order was threatened. Only massive intervention from the public sector avoided a meltdown. At the same time, the year 2012 marked the year that growth rates in the People’s Republic of China dropped under 10% annually for the first time in over a decade. This twin shock to the global economy, in combination with austerity measures put in place in the aftermath of the 2008 crisis, led to the unraveling of the global economic architecture put in place over the previous quarter century.

This is the context of chronic low growth in which money has been pouring into AI, either invested in privately held firms such as Anthropic or OpenAI, or publically held tech firms listed in the US such as Google and Microsoft, or in overseas firms such as semiconductor giant SK Hynix in South Korea. The search for returns continues to drive valuations of AI majors to astronomical heights, bringing significant portions of the market along with them. In May, 2026 Anthropic, which only recently turned a modest profit on sales of its AI assistant chatbot Claude, came close to breaking the valuation threshold of $1 trillion USD. (There is no standard, objective methodology for obtaining such valuations for private companies). According to the OECD, artificial intelligence firms captured 61% of global venture capital in 2025. A Stanford report puts global VC investment in Silicon Valley firms alone at 75% of the total. Overall capital spending on AI in the United States is at roughly 2% of GDP, and expected to reach 3-4% in 2027. This is equal to and slightly exceeding the US defense budget.

This is a lot of investment, but in what, precisely? As it stands right now, AI is an all-purpose everything enhancer, like the patent cure-all medicines of the nineteenth century.

There's more at the link, including thumbnail accounts of recent events on Wall Street and among VCs.

Think of it like this: The techno cult launched ChatGPT in late November of 2022. That's the fuse. That large pile of money that's been floating around looking for investment vehicles, that's the powder. When ChatGPT became a surprise hit, the fuse was lit and the powder exploded: BOOM! That's where we are now.

Eye on the High Line

Dolly Parton, RIP

How LLMs work [in pictures]

We will use everyday language where possible and mark technical terms in purple. Key is that modern LLMs have a component of making answers (in blue, called "inference") and a component of improving answers (red, called "training"). pic.twitter.com/iUmGaNMKUz

— Kording Lab 🦖 (@KordingLab) August 25, 2026

The real work is done by a core machine, the neural network which is used to generate one word at a time based on its inputs. The probability of each word depends on prior words. And words are chosen one by one according to those probabilities. pic.twitter.com/7zWO87TThs

— Kording Lab 🦖 (@KordingLab) August 25, 2026

The result of this are what is sometimes called large-scale "stochastic parrots". Very impressive text but often somewhat superficial. But these are the old LLMs (think 2022). Innovation was super fast since.

— Kording Lab 🦖 (@KordingLab) August 25, 2026

A second idea (B) for making answers, tool use. Call programs like databases, internet search, calculators etc. pic.twitter.com/zyPU00vmV3

— Kording Lab 🦖 (@KordingLab) August 25, 2026

And all of this, is better with better “training“ goals. Wherever truth is defined, push towards this truth (D). And where it is not defined, use LLMs to check answers ("self evaluation", (E)) pic.twitter.com/vLOy2UCsYb

— Kording Lab 🦖 (@KordingLab) August 25, 2026

A great deal of the strengths and weaknesses of today's LLMs can be understood from these component ideas.

— Kording Lab 🦖 (@KordingLab) August 25, 2026

On the waterfront [uptown Hoboken]

We need to consider debt forgiveness in the context of our $40 trillion deficit

Paul Vigna, An Ancient Sumerian Solution to Our $40 Trillion Deficit, Aug. 22, 2026.

The U.S. federal debt has hit $40 trillion. Add the debt owed by U.S. states, corporations and consumers, and the figure rises to about $77 trillion in debt, set against an annual gross domestic product of about $32 trillion. The interest on all of it is compounding constantly. It’s not just the United States, either. Globally, there is $350 trillion in debt, roughly treble global G.D.P. It’s like snowpack on a mountainside. It may look stable right now, but it’s creating the conditions for an avalanche. We are past the point where we can deal with our current debts in normal ways. The options open to us are extremely unlikely or highly destructive: Grow our way out of it, raise taxes, inflate the debt away or wait for the economic fallout.

Ancient societies had another method to deal with debt. It was called an amargi — a blanket declaration of public debt cancellation. All public debts written off. Disappeared. It sounds laughable, I know. But, really, that’s just because the idea has been buried so deeply in history that you’ve probably never heard of it. In the ancient world, it presented a pragmatic solution to an intractable problem. And now, faced with impossible-to-repay debts that are weighing down our economy, is the time to look at the amargi and the lessons it offers about how to think about finance.

Vigna then runs through a quick inventory debt forgiveness in the ancient world.

The reason the practice often worked in the first place was because the ancient world understood something about our monetary system we have mostly forgotten: Money is an invented social construct. It isn’t real, not in the way a tree or a stone is real. The system of money and credit is a thing humans made up. It’s a record-keeping device for distributing resources. And since money is a human creation, we can alter it when needed.

It’s hard to predict what will happen if we don’t address our debt, but it’s not hard to predict that the outcome will be bad. Governments can borrow to cover up other problems for only so long. Sometimes what comes next is a hyperinflationary spiral and economic collapse, as in Zimbabwe or Weimar Germany. Sometimes it’s a national default that rocks financial markets, such as in Egypt and Anatolia under the Ottoman Empire in the 1870s. And sometimes it’s just a steadily slipping quality of life as debt saps people’s ability to build their own wealth.

Money represents resources, or at least access to resources. When so much of it is going to debt repayment, it means money that could be spent on goods or services is diverted. The federal government now spends over $1 trillion a year on debt interest — money that could otherwise be spent on roads, schools or health care. [...] We are unlikely to see a modern amargi, but as a set of principles for reconsidering our relationship with money, its applicability is rich. [...] We need to start seeing money and debt the way people saw it back then: As a system to distribute resources that, like any system, can be periodically reset.

Red flowers on the street corner

Nestled among the beans

Why did Robert Wright (have) to drink the Silicon Valley Kool Aid? [GT-6]

I thought that, when I’d posted notice of my 3 Quarks Daily review of Robert Wright’s The God Test that that was that. But then I read his interview with Liron Shapira, AI Dystopia Is Just 8 Years Away — Robert Wright, Bestselling Author of “The God Test.” Whoops! There’s that phrase again.

The phrase I’m talking about is “reverse engineer.” I believe that the term has been kicking around in psychology for several decades, but I associate it with Steven Pinker’s 1997 book, How the Mind Works (pp. 21-22:

Reverse-engineering is what the boffins at Sony do when a new product is announced by Panasonic, or vice versa. They buy one, bring it back to the lab, take a screwdriver to it, and try to figure out what all the parts are for and how they combine to make the device work. We all engage in reverse-engineering when we face an interesting new gadget. In rummaging through an antique store, we may find a contraption that is inscrutable until we figure out what it was designed to do. When we realize that it is an olive-pitter, we suddenly understand that the metal ring is designed to hold the olive, and the lever lowers an X-shaped blade through one end, pushing the pit out through the other end. The shapes and arrangements of the springs, hinges, blades, levers, and rings all make sense in a satisfying rush of insight.

With that in mind, let’s look at some remarks Wright made to Liron:

Even so, what this approach to training does is, I think, replicate specific cognitive functionality in these machines that exist in the human mind. I don’t mean it does things exactly the way the human mind does. These models have independently invented things that natural selection invented. Edge detection’s a very clear case.

Wright doesn’t use the phrase there, but that’s what he has in mind. Here Wright uses the phrase:

... but even with the current paradigm, you give it kinds of data, visual, auditory, written, whatever, and it basically reverse engineers parts of the human mind that do the transmutation of one form of data into another.

There’s the key phrase. Here’s one last passage. The phrase isn’t here, but the idea certainly is:

You just make the machine good at predicting the next sequence of letters. When you first show it the stuff, it’s pure gibberish, but the machine itself finds a way of mapping the meaning. We do give it the basic... We do say you gotta use vectors to represent the words. It didn’t invent that. But we didn’t say, “By the way, you should choose numbers to fill in the blanks in the vectors that capture this thing we call meaning.” No, it in effect discovered that meaning is a property of words. I would put it that way.

NO.

It’s one thing to talk about psychologists “reverse engineering” bits of human behavior. I have no problem with that. But that’s not what Wright is doing here. He’s talking about machine learning, about transformers, reverse engineering human language and cognition. That’s not at all a useful way of conceptualizing what’s going on. It’s anthropomorphizing. It’s also the kind of mystification that Silicon Valley has been using to hype AI.

Once again I refer you to Berk Idem’s review, where he notes:

My problem is the road he takes in the book. Wright keeps telling the story of AI as if the machines discovered things on their own, that they found the meanings of words, that they grew something like an eye, that they started to evolve, when in fact people set almost all of the machinery up on purpose, with a pretty clear idea of what they were doing and why. I never expected myself to be on the “intelligent design” side of a debate, yet here I am, for instance, arguing that LLMs did not miraculously discover meaning, they were designed to do that. Wright gives too much credit to what models discover during training and too little to the architecture, objective, and research program that produced those discoveries.

Idem is correct. You should consult his review for more details, but I’ll say a couple of words about edge detection in vision and the meaning of words.

Back in 1959 Hubel and Wiesel published their seminal work on the visual cortex of the cat, work that led to their 1981 Nobel Prize in Physiology or Medicine. Virtually all work in machine vision has been informed by that study in one way or another. Machine vision systems are engineered to be sensitive to edges.

The same is true for language. It simply is not true that transformers “in effect discovered that meaning is a property of words.” There is a long tradition within linguistics and computational linguistics of thinking about the meanings of words as a function of the contexts in which words are used, something I review in a recent working paper, The Origins of LLMs – A long tectonic subduction event finally producing a visible volcanic eruption in November 2022 (the subtitle was suggested by ChatGPT). The idea dates back to the 1950s while its computational exploitation dates back to work that Gerard Salton began in the late 1960s on information retrieval. He’s the one who came up with the idea of using vectors to represent linguistic meaning. The transformer is thus a recent elaboration of an idea that’s been around for decades, an idea that human researchers came up for conceptualizing meaning.

* * * * *

There’s much more in Wright’s interview with Liron, much of it interesting. And some of it is bothersome, but I’ve said enough on that score. The God Test is an interesting book. But be careful of what Wright attributes to the machine. You’re better off having no explanation than accepting one that’s misleading at best.

Elsbeth and the tech bro’s panic room [Media Notes 190]

I’ve been watching episodes of Elsbeth off and on. It’s a spin-off of The Good Wife and The Good Fight. Elsbeth Tascioni is an eccentric attorney assigned to the New York Police Department on a consent decree. She functions as a sleuth in each episode. I’m interested in episode 36, “Bunker Down” in season 3 (airing on Nov. 13, 2025). Here’s how Wikipedia characterizes the episode:

Paranoid fintech [financial technology] billionaire Craig Hollis traps his crisis manager Anders Whitman in Craig's state-of-the art panic room, erratically believing that Anders will report his "indiscretions" to his company's board of directors.

Hollis is presented as particularly eccentric and willful in his ways. I couldn’t help but think of this as a deliberately nasty satire of an archetypal billionaire. Perhaps it was Hollis's glee in all the gadgetry incorporated into the room.

High tech billionaires are not particularly rare in movies and TV. But this one somehow struck me as a particularly nasty depiction. Not twirling mustaches nasty – he was clear shaven – but nasty. I’m wondering if we’re in for a rash of such characters.

How Much Would an AI Crash Destroy?

Youtube page:

Two chip stocks recently drove seventeen percent of the entire global stock market's return in a single month — which tells you just how concentrated the AI trade has become, and how exposed the average investor now is without realising it. In this video we look at how much wealth an AI crash could actually destroy, with estimates from Dean Baker, former IMF chief economist Gita Gopinath, and Oliver Wyman running into the tens of trillions of dollars. We cover why the usual places to hide — small caps, value funds, international stocks — are now packed with AI stocks, what the Bank for International Settlements found when it compared today's buildout to the great railway and dot-com bubbles, and why a technology being real has never been enough to protect the people who overpaid for it. This isn't a crash prediction. It's a look at the downside risk, the illusion of diversification, and why boring, unexciting investing tends to win in the end.

Three tablescapes

Cyberpunk City: Hong Kong & Ghost in the Shell

A third transition in science and the evolving cosmos

Kauffman SA, Roli A. (2023) A third transition in science? Interface Focus 13: 20220063. https://doi.org/10.1098/rsfs.2022.0063

Abstract: Since Newton, classical and quantum physics depend upon the‘Newtonian paradigm’. The relevant variables of the system are identified. For example, we identify the position and momentum of classical particles. Laws of motion in differential form connecting the variables are formulated. An example is Newton’s three laws of motion. The boundary conditions creating the phase space of all possible values of the variables are defined. Then, given any initial condition, the differential equations of motion are integrated to yield an entailed trajectory in the prestated phase space. It is fundamental to the Newtonian paradigm that the set of possibilities that constitute the phase space is always definable and fixed ahead of time. This fails for the diachronic evolution of ever-new adaptations in any biosphere. Living cells achieve constraint closure and construct themselves. Thus, living cells, evolving via heritable variation and natural selection, adaptively construct new-in-the-universe possibilities. We can neither define nor deduce the evolving phase space: we can use no mathematics based on set theory to do so. We cannot write or solve differential equations for the diachronic evolution of ever-new adaptations in a biosphere. Evolving biospheres are outside the Newtonian paradigm. There can be no theory of everything that entails all that comes to exist. We face a third major transition in science beyond the Pythagorean dream that‘all is number’ echoed by Newtonian physics. However, we begin to understand the emergent creativity of an evolving biosphere: emergence is not engineering.

* * * * *

Eli Stark-Elster, Earth’s Organisms Developed Via Evolution. Some Theorists Wonder: What if the Entire Cosmos Did, Too? Smithsonian Magazine, August 18, 2026.

Historically, scientists have viewed the cosmos as akin to a rock—complex, perhaps even beautiful, but formed through arbitrary events. Evolutionary cosmologists instead argue that universes, like living organisms, grow and reproduce, spinning off new universes with small variations from their progenitors. In doing so, these cosmic offspring are refined across generations into forms that maximize the number of universes yet to be born. Since the Big Bang, our universe, like any developing child, has been unfurling into an optimal shape.

Under this view, the universe is not a rock. It is an egg.

This analogy comes from Julian Gough, an Irish poet, novelist and musician best known for writing the short story that plays at the end of the video game “Minecraft.” He has recently become a public advocate for and scholar of cosmological evolution. Gough writes a Substack blog, The Egg and the Rock, where he publishes essays, personal updates and predictions about new data from the James Webb Space Telescope.

Over a decade ago, Gough began wrestling with the question of why the complexity of our universe has increased over time. “It goes from … a ball of hot gas to building out structures like stars and galaxies,” he says. From there come planets and, on at least one of them, biological life. “That’s a very strange thing for hot gas to do.”

Maybe, he wondered, the universe evolved into its current form. “It seemed to me an evolutionary explanation was the natural one,” he says. “In every sphere, when we discover a self-ordering, self-complexifying system, it turns out it’s had a previous evolutionary history. And why would that not apply to the universe itself?” [...]

Smolin published his ideas in a 1992 paper titled “Did the Universe Evolve?” as well as a 1997 popular-science book called The Life of the Cosmos. Gough found both and read them excitedly. And yet, it seemed as though virtually nothing had come of Smolin’s ideas since. “Obviously,” says Gough, “[I thought] it must be wrong, because it had been 25 years and nobody seemed to be putting it forward as one of the possible mainstream explanations. Then I dug into the literature and realized—oh, my God—it hasn’t been falsified. It just hasn’t been engaged with properly at all.”

Smolin’s theory had not completely vanished. Some futurist philosophers had extended the idea to explain the emergence of complex life from hot gas and stars. Humans exploit increasingly powerful energy sources, a progression they argued could culminate in artificial black holes. If those black holes spawned new universes, then a cosmos capable of producing technologically advanced life might produce more offspring. The progression from hot gas to stars, life and technology could, they proposed, be a product of cosmic natural selection. [...]

On July 8, 2022, four days before Webb released its first data, Gough published his predictions on Substack. Already aware that his amateur background might make him sound unreliable, he feared being publicly wrong. “I was really terrified,” he says. “This could be fantastically humiliating.”

Fortunately for Gough, the results were the opposite. Webb found galaxies forming earlier, faster and in a more orderly manner than standard physics models had anticipated. In November 2022, NASA reported evidence that some galaxies had begun assembling only about 100 million years after the Big Bang, while follow-up observations of a galaxy just 470 million years after the event found an unusually massive black hole—nearly as massive as all the stars surrounding it.

In a 2025 paper published in the Astrophysical Journal, scientists described these discoveries as a “conundrum [that is] part of the larger challenge to understand the stunning prevalence of massive structures and galaxies in the first few 100 million years after the Big Bang.”

There's more at the link. [H/t Tyler Cowen]

Cf. my article, Welcome to the Fourth Arena – The World is Gifted, 3 Quarks Daily, June 20, 2022.

Friday Fotos: The Mystery that is the Bergen Arches

Algorithmic Grammar of Flexible Cognition

🧠🤖New preprint
Algorithmic Grammar of Flexible Cognition:
A Walk through Latent Operationshttps://t.co/pgBHEw0TtB
Flexible cognition moves adaptively btwn cognitive modes. We formalize it as navigation, a walk over latent operations in neural space as much as the environment. pic.twitter.com/7HesQlvV5c

— Ida Momennejad (@criticalneuro) August 21, 2026

Abstract of the article linked above:

Flexible behavior requires moving adaptively between cognitive modes, between memory and generalization, or cached inference and step-by-step reasoning. Reinforcement learning (RL) offers a language to formalize adaptive behavior in terms of learning and meta-learning over states, actions, policies, and rewards, and neuroscience identifies representational geometries of memory and abstraction. However, the unit of analysis remains representations and a few operations or their tradeoffs. This misses the rich compositional operations commonly asso- ciated with prefrontal-hippocampal interactions. What is missing here includes, first, operations as units of analysis; second, higher order operations that can act on entire cognitive maps and representations, transferring structures, reshaping or merging them; and third, a cognitive and algorithmic grammar for the selection, ordering, and composition of operations. Here we examine the intersection of RL, computational neuroscience, and AI interpretability to identify the tools to address this gap. The latent spaces of transformers offer high-dimensional neural spaces as a testbed for composition of functions. Analyzing the order, branching, recurrence, and composition in sequences of algorithmic operations can help cognitive science identify grammar over algorithmic operations. In the other direction cognitive sciences help shift AI evaluation from benchmarks to adaptive paradigms, and AI architecture from input-output and next-token objectives to setting algorithmic operations and grammar as objectives, e.g., next-primitive-prediction.

Waffle, butter, bacon

Deep in the forest by the lake

In Newport Mall, Jersey City

Patterns of shadow , reflection, and refraction

A firetruck in Hoboken