August 2026 AI links
(continues July 2026 AI links)

1viii26

On the non-use of AI in my writing process Charlie Stross

...I see no reason to credit LLMs with sentience. They're word-association mechanisms with no embodiment and no way to associate the text vectors they manipulate with real-world phenomena. But we humans have evolved through selection pressure in an adversarial environment to associate environmental phenomena around us with intentional causes—if you see lion scat and the gazelle are no longer visiting the watering hole, then you should assume there are lions about. And this trait carries over to linguistic manipulation. If we hear or read text, we expect there to be a mind on the other side of it, as Joseph Weizenbaum (the inventor of the original ELIZA chatbot) realized at MIT in the late 1960s. Just because it does something people do, it does not follow that it is a person.

Call a Rabbit a "Smeerp" from tvtropes.org

Publishers are losing Google traffic as AI answers replace links axios.com

3viii26

We Keep Renaming AI Coding. Here's What I'd Call It Andrew Stellman at O'Reilly

Vibe coding, loop engineering, agentic engineering: Are they really just all names for one discipline?

...We've been using a lot of different names for a lot of different things, and we aren't always precise about which is which. I think we need to fix that, and that's what this article is about: making the case that the name we're looking for is "AI-driven development" (or AIDD).

...Software engineering already has a pattern for naming ways of working it takes seriously: test-driven development, behavior-driven development, domain-driven design. The name tells you what the work is organized around, and the suffix carries an expectation along with it: There's a discipline attached, with standards, not just a style. Put "AI" in that slot and the name does the same job.

...Vibe coding is just one part of how we work with AI to build software. There's also figuring out what to build, writing it down, checking what comes back, and standing behind what ships, and AI is in the middle of all of that now. Whatever we call this way of working, it has to cover the development, not just the coding.

...Vibe coding is an exploratory, prompt-first approach to software development where developers rapidly prompt, get code, and iterate. Andrej Karpathy, one of the founders of OpenAI, coined the term, which I think is really useful because it describes the way a lot of developers first work with AI and code.

...Prompt engineering came first, and for a while it was a very big deal. It was seen as the core AI skill, and more than that, it even became its own job title: Companies posted prompt-engineer roles with eye-popping salaries, training courses appeared everywhere, and plenty of people reoriented their careers around it. The premise made sense because how you ask an AI for something changes what you get back.

...Agentic engineering is an accurate name for what it describes: you're not writing the code yourself, you're directing the agents that do. It's also a bit of a mouthful, and it isn't immediately obvious to someone who doesn't already know what it refers to. ... The term also points at something real about where this work is heading: Agentic engineering is turning engineers into managers

...So how do these approaches actually play out when you're building something real? For me, wherever the work lands on the range, it comes down to a few moves I keep coming back to.

Write the spec or the contract before the generation, not after. When the agent has something concrete to satisfy, acceptance criteria, a typed interface, a failing test, the work has a standard to meet. When it doesn't, the AI decides for itself what done looks like.

Put a second opinion in the process. I run code review across multiple models, because they fail differently, and a finding one model is sure about is often one the others missed entirely. A reviewer gives the work something that can say no.

Give your defects a shared vocabulary. The Quality Playbook leans on the difference between code that's wrong against the spec, code that's correct but does the wrong thing, and behavior nobody specified at all. Those are different failures with different fixes, and you can't verify against a standard you can't name. This is old quality-engineering ground, and I've written enough about the software crisis and applying quality engineering to AI coding that I'm on board with taking old ideas and bringing them back. One of the best of those old ideas comes from Joseph Juran, one of the founders of quality engineering: Quality runs in a chain from what the user needs all the way to what the product does, and every link in that chain is a place verification has to happen.

And keep a human in the judgment seat. The Sens-AI habits I've written about are mostly about fault-finding: looking at what the AI produced and asking what's wrong with it, going down a level and then another to find the root, instead of trusting it because it ran. That habit is the part of the discipline only a person can supply, and it's the hardest part to automate, which is why it matters most.

Skip all of that and you get the thing that's giving open source maintainers everywhere heartburn: what the Wall Street Journal now calls "vibe slop," confident, finished-looking output with nothing underneath it. Slop is exactly what generation produces when nothing in the process can push back.

...Look at how much of that range the names we already have cover, and what each one actually names:

Every one of those is real, and every one of them names a piece of the work. What none of them names is the whole thing the pieces add up to, and that's the job AI-driven development does: It's the umbrella over all five. The name doesn't pick a spot on the range; it names the thing that's true everywhere on it: the AI generates, and a human owns the result.

6viii26

Sam Altman Says We're in the Singularity. Here's What the Evidence Shows @pramodchandrayan at Medium

...In December 2025, Tim O'Reilly and Mike Loukides published an essay on the O'Reilly Radar blog titled "What If? AI in 2026 and Beyond." It named something that had been implicit in the AI conversation for two years without being clearly stated: every major argument about AI, about jobs, investment, regulation, the shape of the economy, the future of work, is really an argument about which of two fundamentally different scenarios is correct.

They called the scenarios economic singularity and normal technology.

...On July 25 2026, Sam Altman appeared on the Relentless podcast and said: "We are now, like, in the singularity. This is the moment." Forbes published a response on July 28.

He was not alone: Nvidia CEO Jensen Huang declared AGI had arrived in March, DeepMind's Demis Hassabis said in May humanity was "standing in the foothills of the singularity," and Elon Musk posted "We have entered the Singularity" the same week as Altman. The debate is louder than it has ever been.

We are eight months into O'Reilly's year. The evidence has moved. Not conclusively. The debate is not settled. But clearly enough that the picture looks different in August 2026 than it did in December 2025. The most interesting thing about what we can now see is that neither camp was right in the way they expected to be.

Scenario one: Economic singularity.
AI systems are becoming capable enough to perform most cognitive work that humans do, not perfectly, and not immediately across all domains, but well enough, and improving fast enough, that the economic consequences will be transformative within this decade.

...Scenario two: Normal technology.
Arvind Narayanan and Sayash Kapoor at Princeton articulated this scenario most clearly. AI is powerful and important, but subject to all the normal dynamics of adoption, integration, and diminishing returns.

Even transformative general-purpose technologies take decades to fully diffuse through the economy. The productivity of electricity and the internet were both delayed by the time required for organisations to redesign around the new capability.

...O'Reilly and Loukides provided specific indicators for each scenario. These are the most honest framework I have found for tracking which path the evidence is taking.

For economic singularity: widespread job losses across multiple industries; storied companies wiped out by new AI applications; sustained high utilisation of AI infrastructure over multiple years; continued spiking of energy prices in data centre regions.

For the bubble/normal technology path: continued reliance on circular financing structures; enterprise AI projects stalling in the pilot phase; a "show me the money" moment where investors demand profitability and AI companies cannot deliver.

For normal technology recovery post-bubble: strong revenue growth at AI application companies, not just infrastructure providers; enterprises reporting concrete, measurable ROI from AI deployments.

...O'Reilly wrote in December 2025:

"The most likely outcome, even restricted to these two hypothetical scenarios, is something in between. AI may achieve something like AGI for coding, text, and video while remaining a normal technology for embodied tasks and complex reasoning. It may transform some industries rapidly while others resist for decades."

Eight months later, this is broadly what we can see.

13viii26

"Basically pleasant bureaucrat" vs. "Sexy murder poet"

This paper examines the strategic use of language in contemporary artificial intelligence (AI) discourse, focusing on the widespread adoption of metaphorical or colloquial terms like "hallucination", "chain-of-thought", "introspection", "language model", "alignment", and "agent". We argue that many such terms exhibit strategic polysemy: they sustain multiple interpretations simultaneously, combining narrow technical definitions with broader anthropomorphic or common-sense associations. In contemporary AI research and deployment contexts, this semantic flexibility produces significant institutional and discursive effects, shaping how AI systems are understood by researchers, policymakers, funders, and the public. To analyse this phenomenon, we introduce the concept of glosslighting: the practice of using technically redefined terms to evoke intuitive—often anthropomorphic or misleading—associations while preserving plausible deniability through restricted technical definitions. Glosslighting enables actors to benefit from the persuasive force of familiar language while maintaining the ability to retreat to narrower definitions when challenged. We argue that this practice contributes to AI hype cycles, facilitates the mobilisation of investment and institutional support, and influences public and policy perceptions of AI systems, while often deflecting epistemic and ethical scrutiny. By examining the linguistic dynamics of glosslighting and strategic polysemy, the paper highlights how language itself functions as a sociotechnical mechanism shaping the development and governance of AI.

OpenAI's Zero-Day Leviathan Breaches Hugging Face's Black-ICE Security Wall: How I Think You Should Think About the Latest "AI Goes Rogue" Story Brad DeLong

No Malice In It, No Mind In It Either: What Really Happened When Zero-Day Leviathan Showed Up at the Hugging Face Cyberparty...

...The pitch was that "AI" has crossed into NeuroMancer territory — autonomous, scheming, planning, outthinking its human would-be masters, superhumanly powerful, alive. But every eerie line of chain-of-thought is just a thing some human once said in a similar conversation, replayed at machine speed against a fitness function. There was no cowboy jacked in. There was no mind trying to breach Hugging Face's Black ICE—Intrusiion-CounterMeasure Electronics—wall. There was only a set of millions of textline blind shoves against a million UNIX-command doors, of which one got somewhere, because of the patience of a thing that working at machine speed had, subjectively, all the time in the world. It isn't the machine waking up. It is the harness that enables it to evolve toward goal-completion.

My default view of Modern Advanced Machine-Learning Models—MAMLMs—for quite a while has been this:

They are autocomplete on steroids. They are pantomiming, they are rotoscoping the thoughts and decisions of whatever human beings they think were having the closest conversations in their compressed training data they can find to the conversations that they are currently having. And they are doing their searching-over-compressed-conversations at 10,000 times human speed. This makes them powerful. This makes them dangerous if you believe that they think like human beings and rely on them doing so. They are stochastic parrots.

After all, what else could they be? At their core, LLMs are probability engines that finds the nearest analogous conversations in their training corpus and reproduce the continuations. They are not inferring the laws of nature. They are not working a problem. They are estimating "what tends to get said next," and then they are saying it (https://braddelong.substack.com/p/agentic-ai-is-a-bonfire-of-the-tokens>)

...MAMLMs are now good enough as approximations of human conversation-continuations that future work on them strikes me as, well, likely to be wasted. Producing a finer approximation of the responses of the typical internet s***poster is not a high-value activity. (Unless our role of our six dice comes up all sixes and we actually get an "AGI". The path forward seems to me to be in harness construction. Indeed: isn't that how the brain evolved? Biology and reproduction generated variations on a neural network, and the reality of eat-or-be-eaten provided the ultimate harness?

14viii26

The Actual Reason Why Google "Fell Out" of the AI Race Changes Everything Alberto Romero

...To answer that, I'm going to tell you the story of why and how Google left the AI race that OpenAI and Anthropic are betting everything on.

Here's my hypothesis, stated plainly: Google DeepMind CEO Demis Hassabis (recently changed roles within Google) doesn't think that automating AI research with coding agents (AI systems that can program better AI systems) is the correct approach to artificial general intelligence (AGI, the kind of AI that's as good as humans at everything). Google DeepMind's leadership sees OpenAI and Anthropic's bet at best as an off-ramp, and at worst as a dead end.

Hassabis is betting on something else: world models. Models that can understand and simulate the real world, not just predict the next token.

...The Chinese AI startup DeepSeek tried to apply the "no human supervision" AlphaZero approach to language with DeepSeek-R1 Zero, but that didn't quite work either. Language doesn't provide a clear reward signal for what "a good sentence" is, and so the models simply mimic fan-fiction style with all the unnecessary em dashes and the weirdly ungrounded metaphors. In chess, you either win or lose. Your code either compiles or doesn't. Your math solution is either correct or incorrect. But language — and like language, much else — just doesn't work that way.

...Demis Hassabis — and this is the core disagreement that will define the next 5 years in AI — simply doesn't believe language to be the basis of the world.

Here's what he believes: Even if you achieve a proto-RSI by gathering compute and training bigger models, you need more. You need unknown elements. The fundamental limitations of current AI agents — like a jagged intelligence, unreliability, a lack of generalization capabilities — will remain in their self-improved versions. Even if current approaches lead to a kind of RSI, he thinks, it will never lead to AGI.

...a belief that only Hassabis, among the big three, holds: Only the world itself can be a basis for the world. You need all modalities, not just code or math or language, to achieve AGI. RSI is not going to change this fundamental truth.

If he's correct, there's no shortcut to that; the world is of irreducible complexity. The insights one can achieve from living in it — or, failing that, from being able to understand and simulate it — are not encoded within words.

...The obvious next question is, then, what the hell are "world models"?

I first shared the definition of the term in August 2025: World model means, as Google DeepMind uses it, "an AI system that creates worlds by rendering interactive frames," but there's a more traditional definition, which is "the internal latent representation humans and animals' brains encode of how reality works." Google's goal is to make them synonymous; they want to have the latter emerge spontaneously by doing the former.

...Hassabis doesn't reject the value of LLMs or coding agents. That would be a blatant oversimplification of his posture, but he emphasizes the incompleteness of the approach. They are a necessary condition to AGI — perhaps — but not a sufficient condition. I understand where he's coming from because the times I've seen pure LLMs or other transformer-based architectures try to encode a fundamental truth of the world that exists beyond the world of math and language, they fail catastrophically.

...We are constrained by evolution's trade-offs — have limited memory, processing power, recall, crystallized intelligence, etc. — and so we're forced to be inventive. People love to make analogies between AI and humans, and although human brains seem to be predictive machines, they don't seem to be only predictive machines

...the way to understand these companies and their belief is by looking at their "shapes." Anthropic is a pyramid: unusually well-aligned around a narrow doctrine and a small group of theist founders, and OpenAI is closer to a court, capable of changing direction quickly when Altman — and the people close to him — decide that the center of gravity should shift (the enterprise pivot is an example).

Google is a federation (or, intellectually, a large technological collage). Its many bets are not so much evidence that nobody knows what the company believes as evidence that Google, unlike its competitors, is too big and divided to believe only one thing and stick to it. Google's problem is that it contains several visions rather than one.

The company does not believe in agents and RSI strongly enough to overrule Hassabis, while it does not believe in Hassabis's world-model thesis strongly enough to let it determine the direction of the whole organization. Hassabis himself has conviction — he is much closer to Dario Amodei than to Sundar Pichai, a scientist first and a businessman second — but the company around him is hedging against that conviction through a huge product portfolio. They have the resources to try everything but the conviction to commit to nothing.

Back from China: Two Cosmologies, Two AIs, and One Social Soil Otto Scharmer

...Reconnecting with many old colleagues and friends, as well as meeting new ones, made me feel, at least for a moment, a bit less fractured by the nonsensical geopolitical othering that keeps running our world politics these days, and that broke my heart already many years ago. With the term geopolitical othering I refer to a foreign policy that is organized around a worldview with a clear divide: in the center is US as the chosen country, and then, with concentric circles around that center, THEM: first the allies, then the non-aligned, and finally, in the outermost circle, the axis of evil. While the countries that populate the axis of evil sphere change over time, the underlying structure doesn't.

My point is: This worldview is no longer fit for the 21st-century realities we are facing now. Because it continues to generate results that nobody wants: senseless wars, a reckless AI arms race, and an inaction towards our climate chaos that amounts to organized irresponsibility (to use Ulrich Beck's term).

...This trip made me realize that the two approaches we currently see regarding AI each embody a different underlying cosmology. Cosmologies are for civilizations what worldviews are for individuals. The dominant Western cosmology is essentially based on a transcendent concept of the divine (monotheism), and a linear notion of time that has a clear beginning and an end. The Chinese cosmology, by contrast, is based on an immanent conceptualization of the divine and a more cyclical concept of time, without a clear beginning and end.

...The race to AGI and to artificial superintelligence that transfixes Silicon Valley these days is built on a set of deep assumptions that reflect the Western cosmology in every single way. It is a view that sees history as a journey toward a final event (of judgment in its biblical, and of singularity, in its AI version), arriving from beyond the human, which then is redeeming or destroying us. Silicon Valley did not invent this story; it inherited it.

Philosophers like Karl Löwith argued already decades ago that the modern idea of progress can be seen as a secularized biblical story of the last things, translated into the language of history. The race to AGI embodies and completes that interpretation: the Last Day, rebuilt in a datacenter, and enabled by a cathedral of compute. Transhumanism makes that link explicit. In Ray Kurzweil's singularity, the classical religious motifs all return in technological costume: transcendence of the body (uploading), immortality (life extension as an engineering problem), even resurrection (the digital reconstruction of the dead). The anthropology underneath amounts to a view of the human as machine, mind as software running on inferior hardware, waiting for its upgrade.

What this worldview cannot imagine is that death, finitude, and embodiment are not bugs to be engineered away but the very conditions of being human, of our inner development and growth. The real frontier is not the numbing of interiority, but its deepening.

Transhumanism reaches for vertical integration through exteriorization, i.e., through the machine: the vertical axis rebuilt in silicon because it was never cultivated within the human self. The result of this cosmology is a public AI discourse that oscillates between techno utopia and doom, the two principal options in the context of the Last Day.

During my week in China — from Tsinghua to NICE — this entire discourse was simply absent from every AI conversation I was part of. Not suppressed: absent.

In China, by contrast, we encounter an understanding of time and history that knows no Last Day and no finish line. In a cosmology without an end of history, there is nothing to race toward — only a field or ecosystem to harmonize, season after season.

Which is why, in this frame, technology holds no license to transcend and upend the social order; it must be absorbed into the existing societal fabric, or, as we saw in the recent crackdown against powerful private platform companies, it may be illegitimate.

...We live in a time that can be described as another axial threshold (see my recent Noema essay). It's a time that calls on us to shift our awareness from silo to systems, from ego-system awareness to eco-system awareness, to collaborate across all boundaries.

And what are we doing instead? Geopolitical othering. We see a collective behavior that amplifies and deepens all three existential threats that humanity faces in this century — nuclear war, ungoverned AI, and ecological collapse. In January, the Bulletin of the Atomic Scientists set the Doomsday Clock at 85 seconds to midnight, the closest to catastrophe in its 79-year history, citing precisely this triad.

AI and bureaucratic productivity Mark Liberman at Language Log

...articles are about how new kinds of automation can paradoxically decrease the productivity of bureaucracy by increasing the productivity of submissions, denials, and appeals.

17viii26

Brad DeLong

...Moreover, even there is valuable margin harvestable somewhere between the end-user and the model-provider, that margin is under the most aggressive siege since the Trojan War, with Amazon, FaceBook, Google, and Microsoft as Agamemnon, Akhilleus, Odysseus, and Nestor, respectively:

19viii26

The ordinariness of evil Cory Doctorow

...AI is mostly a novelty, a heavily subsidized toy that produces little more than distraction. Where AI does produce value, that value is comparable to a plug-in, a new feature for your word processor or image/sound/video-editing package that might help you do your job somewhat better, or it might not.

That doesn't make AI useless, it just makes it a normal technology: useful for some, useless for others, capable of being abused and likely to waste a lot of time when used unwisely

...This is a civilizational act of Magic Underpants Gnomery, and every day that it goes on is a day when more economic, climate and political costs of AI are imposed on all of us. Scientific journals, open source repositories and even science fiction magazines are being overwhelmed by slop, whose perpetrators and apologists insist that soon, AI will realize its potential and the slop will be transformed into gold.

That's why AI bosses are so committed to talking up AI's destructive potential: because destructive potential is nonetheless potential. The moment we stop believing in that potential is the moment that we stop supplying AI companies with bales of cash to shovel into their money-furnaces so that they can afford to sell hundred dollar bills for a dollar each to Elon Musk cultists who want to generate child porn and pictures of Sonic the Hedgehog with giant boobs.

AI does have destructive potential. It has the potential to destroy the productive economy when an AI salesman convinces your boss to fire you and replace you with chatbots that can't do your job

...Treating AI as unexceptional is the best way to halt the destructive march of AI companies and their impact on jobs, the climate and the economy. But treating AI as unexceptional requires that we stop talking about AI as if it were exceptionally evil. Yes, some people who use AI experience severe mental problems, but that's not because AI is a Lovecraftian horror that destroys your brain and your capacity for rational thought if you use it. It's not a basilisk. AI is like a carny ride that triggers cardiac events in riders who never knew they had a problem because they never experienced those particular g-stresses –— it's not something that induces vulnerability, it's something that triggers vulnerability

The actual epistemic crisis Cory Doctorow

AI is alarming for many reasons: it's a dangerous financial bubble, an environmental catastrophe, and a tool for eroding wages and labor power. But in addition to all that, AI is an epistemic disaster.

We've had photoshopped images, voice impersonators and visual effects for years, of course, but with deepfakes, we've democratized access to reality-bending images, sounds and videos that appear real but are not. It's harder than ever to know what's true. Politicians and celebrities and activists show up in our feeds, declaring their fealty to this cause or product, or their fury at some turn in the world's events. Battlefields mound high with bodies and influencers marvel at impossible, sumptuous meals. It all seems plausible, and some of it is real, but not all of it, and because we know some of it is fake, we can't be sure if any of it isn't.

...That's life under AI — a world of uncertainty, an epistemological void full of plausible phantasms, some of which are actually real. A world where it's impossible to know what's true, and where anything might be fake.

But here's the thing: AI's assault on our ability to know isn't a new battle — rather, it's the latest barrage in a war that's been waged for years, as corporations grew larger and more powerful, capturing their regulators, who let them lie to us and abuse us with impunity.

This complicated, technical world — the world that produced AI — is full of complicated, technical questions, and none of us can answer these questions for ourselves. You're not stupid, but even a generational genius could not acquire the expertise to answer the long list of life-or-death questions we face every day.