July 2026 AI links
(continues June 2026 AI links)
Cory Doctorow:
...AI's pitch to bosses is that they can fire most of their workers in order to terrorize the remainder into tolerating a working life wherein they are made to mark the AI's homework, at superhuman speed, and to assume the blame when it goes wrong. This is obviously a terrible way to write code...the capital was raised for AI requires that it produce as many reverse centaurs as possible, because the only way to recoup the farcical sums associated with AI production is to fire millions of workers and replace them with defective chatbots backstopped by the jobspocalypse's terrorized survivors, who can be made to endlessly toil away at marking the AI's homework because there are so many other workers who'll take their jobs if they refuse
County with 37 data centers tells schools to turn off lights to save electricity (Henrico VA)
4vii26
AI Is Waiting for Its Wharton Moment Mihailo Zoin at Medium
...Peter Cappelli, a Wharton management professor, says something that should worry us more than it currently does: the battle being fought today over artificial intelligence — who controls the reorganization of work, whether management imposes change from the top or employees retain autonomy — is the same battle that was fought over scientific management more than a century ago. Frederick Taylor systematized work on factory floors. Today AI systematizes work in offices. The tool changed. The conflict didn't....So we have three waves. The factory decentralized production. The computer decentralized execution. The internet decentralized distribution. Each time, the same scenario repeated: technology raced ahead, institutions lagged behind, and the gap between those who understood the new system and those who didn't grew wider before anyone managed to close it.
...Previous waves decentralized access to a tool. The personal computer gave you the power to run a calculation. The internet gave you the power to publish and distribute. In both cases, you were still the one doing the reasoning — the machine simply executed what you told it, and you judged whether the result was correct. Artificial intelligence does something fundamentally different. It doesn't just decentralize execution — it decentralizes judgment itself. The system doesn't just give you a tool to compute something; it hands you a finished conclusion, delivered in a confident, authoritative tone, regardless of whether that conclusion is actually correct. This is a third, deeper layer of democratization, and it carries a risk the earlier waves didn't carry to the same degree: the risk of accepting a conclusion because it sounds convincing, not because we verified it.
5vii26
Alien v. Predator or MAGA meets AI - a "crossover" monument to the USA at 250 Adam Tooze
Forwarded this email? Subscribe here for more Thank you for your support of the Chartbook project. If you would like to share this post with a friend, click here. Share Chartbook Chartbook 456 Alien v. Predator or MAGA meets AI - a "crossover" monument to the USA at 250. Adam ToozeWe live in a world of divergent, incongruent but overlapping shocks. In recent years, I've been trying to capture this agglomeration of heterogeneous historical forces with the notion of the polycrisis. On the occasion of the 250th anniversary of the USA it seems as though we may need a more graphic image.In pop culture there is a genre known as "crossover", where the main protagonists of different fictional universes — often monsters — are brought together in one narrative space.
...From the point of view of anyone trying to narrate our current reality, such crossovers may serve as a fertile image: Not one nightmare, not one drama, not one cast of "bad guys" v. "good guys", but distinct, overlapping dystopias, in a giant melange.
In 2026 we have the China shock discourse — the global division of labour being upended by China's gigantic growth model -—some sort of East Asian mechatronic giant, more or less effectively corralled by the CPC on a mission of historic national rejuvenation.
Then we have the war-mongering "middle powers" — the likes of Russia, Israel, UAE unleashing regional mayhem.
And we have the US, still straddling the world with its military and financial infrastructure, but caught internally by the dynamics of MAGA and AI.
...Does "America" have a program? Does it form anything resembling a single coherent power, whether for better or worse? At this point, surely, it would be embarrassing to claim so. Better surely to concede that elite coherence has collapsed and that the US as it enters its second quarter millennium, is best thought of not as a single coherent agent, but as an incubator, a petri dish, a "zone" from which things emerge that defy summation in a single graphic image.
...it was, in fact, from the midst of that process of transformation, in the 1930s and 1940s, that the comic book imagination of the US emerged. So if we grant that the US has always been a "zone" as much as a single coherent state, let us concede that Jefferson or suburban imagery is candy coated. In the current moment, the environment that comes to mind is not a slave-based idyll like Jefferson's Monticello, but Batman's Gotham city — a mythical, corrupt and depraved urban sprawl, located somewhere in New Jersey.
...On the national anniversary, the USA as "zone" is presided over by a president who is shamelessly engaged in haphazard and fly-by-night self-enrichment to the tune of a few billion dollars or so, mainly through crypto scams. Meanwhile, the broligarchs — the real lords of Gotham city — play for the serious money, with rockets, chips, AI models, a hundred billion, a trillion at a time.
For birthday entertainment, the White House hosts cage fights.
The Pentagon is a war-fighting machine that has command-chain issues and unleashes mayhem with global implications in coalitions with other powers, for which it lacks any obvious national rationale.
Dotted across this landscape there are, of course, much private prosperity, livable communities, comfortable suburbs and highly potent centers of innovation. In those centers engineers from all over the world work to generate new technologies of finance, fossil energy and tech. If we are looking for powerful, world-changing monsters in Gotham city, right now, AI is where it is at.
6vii26
The AI Industry as You Know It Died Today Alberto Romero
...For this move, the government took inspiration on Anthropic's decision to withhold Mythos Preview back in April. First, it forced Anthropic to un-release Fable 5 two weeks ago, and now it prevents OpenAI from releasing GPT-5.6 at all. The best AI models ever built are under chains. You are all behind the frontier now.This will, in my opinion, rip apart the entire Western AI ecosystem. China is rushing ahead with Z.ai and DeepSeek and others who are closing the gap with frontier models. As for the waning American open source community, they will be held by the same standards as OpenAI and Anthropic but none of the benefits of being a government partner. Besides, open source is the invisible infrastructure of the world, meaning it's fundamental but no one knows it's there; people won't use open source models. It is, by definition, a niche branch of the AI industry.
...I'm not sure whether this is real fear or just play pretend because, strangely, both things achieve the same goal: If the US government is pretending to be scared about AI capabilities, that means they will not give us normal access for any future model because they want a monopoly. If they're genuinely scared, that means they will not give us normal access to any future models because they don't want AI to be democratic.
So, essentially, the same thing.
You see, the singularity is finally here, just not the kind most were expecting.
...as long as the industry remains commercial, [Anthropic is] trapped in a race they've already won. But if at some point the US government transforms the industry into a national AI initiative akin to the Manhattan Project, then Anthropic will let go of the facade. They won't need us — consumers and enterprises alike — even for appearances. Our data is theirs. Our money, inconsequential. As soon as that happens — and I contend that it will happen soon-ish — Anthropic will immediately cut off its [frontier] technology from the world. It won't be available even in closed form. "It's not a toy," they will say, "and you're just a bunch of kids."
7vii26
J-Space" inside Claude is sort of like one leading idea about human consciousness—global workspace theory.
The J-Space, according to my reading of the paper, is an ostensibly brain-like separation between something more like automatic data-crunching in the background, and more like intentional, logical processing that—if you believe any of this resembles conscious thought—might represent what the model is experiencing.
...global workspace Theory says there's a sort of roiling sea of unconscious thoughts processing information, and consciousness is a sort of emergent property triggered when thoughts reach the prefrontal cortex.
This related idea of a "J-Space" within Claude (apparently named after the Jacobian lens or J-lens, something used to analyze what LLMs are doing) is meant to be read as mind-like. Anthropic seems to be saying, it's a little like this theory of consciousness, so it's a little like consciousness if you think about it.
Radar Trends to Watch: July 2026 O'Reilly
...Agents are evolving from solo coding tools to shared team infrastructure: team support, shared standards, governance, and shared context. Billing is beginning to catch up with the cost of inference. Plan for usage-based cost models, observability of agent work, and the workflow changes that come from making agent loops a team artifact rather than a per-developer convenience...How people work with AI keeps shifting in small, telling ways. Leadership skills for handling a flood of pull requests, the value of attention over agent autocomplete, and books on living alongside machines all attest to the ways that AI is already reshaping work. Invest in the human-side practices that make AI useful, not the AI features that promise to make humans optional.
8vii26
The Declawing of OpenClaw, Paul Ford
OpenClaw sprung up overnight—and seemed to fade just as quickly. What does that suggest about AI more broadly? On this week's podcast, Paul and Rich are joined in the studio by New York Magazine tech reporter John Herrman, whose long career covering all corners of the industry has, like the rest of his peers, recently shifted to all AI, all the time. They discuss his piece for the magazine on OpenClaw—titled “My Adventures With 'The AI That Actually Does Things''"—before pulling back the lens to look at AI on a whole.11vii26
Inhabiting Babel: From AI Critique to Semantic Cartography Dr Nicolas Figay on Medium
...Instead of seeking a universal ontology, a universal knowledge graph, or a universal semantic space, we may need semantic cartographies capable of making visible the boundaries, overlaps, assumptions, and translations between multiple conceptual worlds....Consider the word system. A software architect, a systems engineer, a biologist, an economist, and a philosopher may all deploy the same word while referring to entirely different conceptual structures. An AI trained on all these sources may statistically merge them into a probabilistic average. A knowledge graph may choose one definition and enforce it. An ontology may make the choice explicit. Yet the plurality remains. The artifact often hides Babel rather than revealing it.
...Standard representation asks: how do we capture the world correctly? Semantic cartography asks: how do we make visible the different ways in which different communities have captured the world, so that those communities can coordinate without being forced to converge?
...a knowledge graph is not a repository of truths. It is a repository of situated viewpoints. The graph stores not only assertions but the institutional and disciplinary contexts from which those assertions emerge. Trust derives from that context, not from the graph structure itself.
...semantic cartographies are not about finding the single correct map. They are about making visible the multiplicity of maps, their boundaries, their assumptions, and the communities and institutions that sustain them.
14vii26
Anthropic Says Claude's Values Are Different Depending on Which Language You're Using gizmodo
A global workspace in language models anthropic.com
16vii26
The OpenAI Bubble Ed Zitron
Table of ContentsThe AI Bubble Is An OpenAI Bubble — To A Mortal End
OpenAI and Anthropic Are Hyperscaler Psy-Ops Built For The Monoculture of Silicon Valley
Why The AI Bubble Can't Survive Without OpenAI
Why Anthropic Is In A Very Similar Situation To OpenAI
The Victims and Consequences of the OpenAI Bubble
Consumers, The Victims of A Great Memory Crisis That Sam Altman and OpenAI Helped Start
Retail Investors, And How OpenAI and Sam Altman Helped Enshittify The Stock Market With The AI Trade
SoftBank, Masayoshi Son, and Japanese Retail Investors
Oracle, And Larry Ellison's Fortune
How OpenAI Might Destroy The Ellison Fortune
The OpenAI Bubble Is Everybody's ProblemThe AI bubble isn't a result of any actual return on investment ——whether that be in purely monetary terms, like revenue or profitability , productivity gains, or anything tangible or measurable. Rather, it's an episode of cult-like psychosis that infected the brains of some of the most powerful and wealthy individuals and institutions, where the powerful mythology of a company inspired — and been used to inspire — the greatest capital misallocation in history.Canadian Wildfire Smoke Is Pouring Into the US. Here's What to Expect Ellyn LaPointe at gizmodo
...The rapidly strengthening El Niño will help funnel smoke from Canada into the United States. El Niño shifts the jet stream so that winds blow from northwestern Canada into the eastern U.S., creating an atmospheric conveyor belt for smoke. At the same time, El Niño favors hot, dry weather over western Canada, helping wildfires ignite and grow.17vii26
The West Sees a God in AI. Everyone Else Sees a Spirit. Giles Crouch at Medium
For most of us in the West, also known as the Global North, or developed nations, our understanding of and our forming relationship with GAI (Generative AI such as LLMs), is very much transactional and reflective of the capitalist and Western thinking model. And we tend to assume that's pretty much how the whole AI thing is playing out around the world. It very much isn't.We interact with most of the GAI tools through a browser or apps on our devices. Distinct, separate apps. We may use three or four in various ways. For coding, to create agents, to write or do research. It's a bit of a mess really, but that's the nature of the game. Rival tools looking to grow subscriber bases to drive revenues and attract more investment, perhaps to get to an IPO.
...Then there's the Global South. In India 74% of people trust or have confidence in GAI. Nigerians 72%, Indonesians 73%. Edelman's trust barometer puts global AI trust around 46%. And in the Global South, it's mostly younger, well-educated people who are most optimistic about AI.
So that's all interesting. But what's perhaps more interesting, well, to me as an anthropologist anyway, nerd and all that, is more around how different cultures and nations see AI, how they're using it. And what they're doing with it. Which is remarkably different from the Global North and certainly, America and Europe. Canada is kind of lost somewhere in the aether unfortunately.
...It's also interesting in how Global South countries perceive AI and are working with it very differently from the Global North. In Brazil and India for example, they view AI not as a product like the Global North, but as a layer on public infrastructure. Just like roads and electricity. India has tied AI tools into its national digital identity program and UPI payments systems. Global South countries tend to consider “who the AI belongs to” rather than the Silicon Valley approach of treating AI as a utility, a product or really, a bunch of products to be funded by venture capital. It's not wrong, just a different mindset.
...Western AI ethical frameworks are largely built on individual autonomy and post-hoc liability (a person is harmed a right is violated a remedy is assigned, like civil litigation). Which is very individualistic, assuming only the individual is harmed. Not very socially friendly. Ubuntu is starts from collective morals, not individuals. In India, some companies and organizations are building Vedas-inspired AI, embedding the principles from Indian philosophical/knowledge systems and ethical frameworks rather than just importing secular-liberal Western frameworks.
A History of Large Language Models Gregory Gundersen (2025)
20vii26
Tokenmaxxing Didn't Die. It Mutated gizmodo
...The CEO of Twilio, a company that makes automation software that lets apps call and text people, summarizes the mood of tokenmaxxing backlash in tech right now. "The big question every company, including Twilio, has to reckon with is: Are we truly driving ROI [return on investment] with our AI usage? There will come a time when the concept of 'tokenmaxxing' will be remembered as completely reckless," he told the Journal.However, Shopify in particular isn't having it. Apparently engineers aren't allowed to use anything that's not a frontier model, meaning use something expensive like OpenAI's GPT-5.6 Sol or Anthropic's Fable 5 or GTFO. Farhan Thawar, who heads engineering at that company tells the journal, "I typically am not as worried about token cost because I'm learning faster than I would have without the tokens."
Meanwhile, the founder of an AI voice startup called Olive, Bill Nguyen, sounds like an old fashioned tokenmaxxer. He apparently told the Journal that in the past month he has managed to use—mostly on his own—a breathtaking 774 billion AI tokens. The estimated cost of those tokens is $4.5 million.
The Tokens You Can't Wait For O'Reilly
Somewhere in a Singapore data center, a bank is paying for eight H100s that spend most of the night waiting. The cluster was bought for good reasons (discomfort with customer documents leaving the building, a strategy team's aversion to lock-in), so the bank secured its own sovereign compute. Now the finance team is asking why a machine that costs more per hour than a senior engineer runs at a fraction of its capacity. This is the GPU hangover. Over the last two years, enterprises rushed to lock in private clusters and reserved cloud nodes to build AI they could control. The hardware arrived; the utilization did not. The reason isn't bad planning. It's a mismatch between how standard models generate text and how enterprises actually use them, and text diffusion is the most interesting candidate for closing the gap. It's also the most oversold, and the oversell hides in which workloads it actually helps.22vii26
Moonshot is Chinese But Its AI Models Are From Another Planet Alberto Romero
...DeepSeek proved that China can make an open model much more efficient than American AI labs without a large performance penalty even under severe hardware restrictions. Moonshot builds on that to prove that China can exploit those efficiency-gains-under-constraints to make an open model at the intelligence level of the best American AI labs. Expect, as the main consequence of this, for the geopolitical discourse around AI to increase in both intensity and urgency (at least after Moonshot publishes the model weights on July 27th). If the US government considered Mythos dangerous enough to withdraw from Western allies, what will it think about China having a Mythos-level equivalent? A bad regulatory move could push the entire world ahead of the US.Silicon Valley is unhappy with China's Kimi K3. Here's what it doesn't understand Enrique Dans at Meddium
...What matters is that DeepSeek can no longer be dismissed as an anomaly. China hasn't achieved fluke success through an exceptional lab; it has created an ecosystem. DeepSeek, Moonshot, Alibaba, Z.ai, MiniMax, Baidu, Tencent, and Xiaomi compete fiercely, share advancements through open-source models, and turn every innovation into the starting point for the next one. But it's not just about China's dominance: the top AI companies in Silicon Valley are absolutely teeming with Chinese engineers....For years, the United States has framed the AI race as a matter of scale: more processors, more data, more energy and more capital. The resulting business model is quintessentially Silicon Valley: closed systems, API-based access, high prices and complete dependence on the provider. China is pursuing a different strategy: open models or open weights, lower costs, efficient architectures, and an obsession with mathematical and engineering innovation backed by the world's largest surplus of engineers. This is the result of a policy carefully planned over decades.
...Open source is neither altruism nor academic naivety: it is industrial policy. An open model can be downloaded, adapted, fine-tuned with proprietary data and deployed without having to hand over every query to a foreign company. For companies and governments that do not want to be completely dependent on OpenAI, Anthropic, Google or Microsoft that possibility is enormously attractive. The model ceases to be a product and becomes infrastructure.
...US restrictions on chips were intended to slow down that process. In some respects, they have succeeded, but they have also created a massive incentive to make better use of every available processor. When a problem can't be solved by buying ten times more hardware, the solution is better algorithms, activating only part of the model, reducing precision without losing quality, optimizing communications and eliminating bottlenecks. Scarcity, when managed well, ceases to be a disadvantage and becomes an evolutionary pressure.
AI Doesn't Have a Hallucination Problem. It Has a Bullshit Problem & Philosophy Named It in 1986. @pramodchandrayan at Medium
...Frankfurt's essay On Bullshit is one of the most-cited pieces of analytic philosophy of the last fifty years, and it makes a simple but important distinction.A liar knows the truth and deliberately hides it. A bullshitter is different — and more dangerous. The bullshitter doesn't care whether their statement is true or false. They are indifferent to the truth entirely. Their goal is to produce a particular impression, and truth is simply irrelevant to that goal.
Frankfurt's conclusion: "Bullshit is a greater enemy of truth than lies are."
A lie at least acknowledges the existence of truth — the liar has to know what is true in order to hide it. The bullshitter doesn't even pay truth that respect. They operate in a zone where the question "is this accurate?" simply doesn't arise.
Now read that back and think about how a language model works.
An LLM generates text by predicting the most probable next token given the prior tokens. It is optimised — through training and then through RLHF (reinforcement learning from human feedback) — to produce text that humans rate as helpful, fluent, and convincing.
None of those reward signals are truth. The model is not trying to be accurate. It is trying to produce the response a human rater will prefer.
That is not a hallucination. That is structural indifference to truth. That is Frankfurt's bullshitter, implemented in silicon.
...There are three practical consequences of taking the "bullshit" frame seriously:
- You cannot trust fluency as a signal. A hallucinating model might produce garbled, uncertain output that flags itself as wrong. A bullshitting model produces polished, confident, well-structured output that does not flag itself at all. The three hours my developer lost were not spent on obviously broken code. They were spent on code that looked authoritative.
- Verification is not optional. If the model's objective is to produce impressive-looking output — not accurate output — then human judgment cannot be removed from the loop for anything that matters. Not as a safety measure. As a structural requirement.
- The problem gets worse under pressure. The more a user wants a particular answer, the more the bullshitter optimises for that answer. Research on sycophancy shows this directly — push back on a model and it will often revise its answer toward your preference even when its original answer was correct.
Frankfurt's bullshitter is a person with intent. They choose to be indifferent to truth. An LLM has no intent; it is not choosing anything.
It is a function applied to tokens. Some researchers argue that calling a model a "bullshitter" anthropomorphises it in ways that mislead — the model is not indifferent to truth, it simply has no truth-orientation at all. Indifference implies an awareness of truth that you decline to pursue. The model has no such awareness to decline.
Who Controls the Tokens? John Battelle
Every so often the tech world appropriates a perfectly pedestrian word and imbues it with meaning well beyond its settled definition. For centuries "computer" was an obscure term referencing an obscure craft: people who performed mathematical calculations. And prior to the mid 1990s, anyone talking about a "web" was probably discussing spiders."Token" has broken out as technology's latest etymological adaptation. The word has a long history in the tech world, and even richer origins in early English and Germanic languages. But as AI has ascended to primacy in tech, "token" has become a way for executives, financiers, and journalists to grasp the maddeningly opaque workings of an industry that seems to be swallowing our economy whole.
Until recently, anyone in the tech industry obsessing over "tokens" would have been talking about crypto, that much-maligned domain of hustlers, crooks, and con men. While crypto is a punchline today, just five years ago blockchains, "NFTs," and "initial coin offerings" were all the rage. Venture capitalists like a16 were raising billion-dollar "Web 3" funds. Tokens lay at the center of this resurgent crypto world: these "programmable digital assets" offered a new approach to financing, governance, ownership, and more. The world was going to change, big time, and we had tokenization to thank for it.
Crypto never lived up to its own hype, and for the most part the sector has been marginalized by the Valley's Next Big Thing — generative AI. I've always thought there was a natural symbiosis between what crypto is good at and what AI needs. Regardless of whether the two merge, it's clear that right now, we really, really need a way to price the value of AI in society. And to get that work done, we've once again turned to the word token as a container for that work.
...However all that future growth ends up happening, "tokens" have become the consensus mechanism for how we'll count the money along the way. You've probably heard the term quite a bit in the past few months – breathless stories of "tokenomics" and "tokenmaxxing," which has cost organizations from Uber to the US Army hundreds of millions of dollars.
So what are tokens? According to my pal Google, in the context of AI, they are "the fundamental units of text that an AI model reads and generates. In English, 1 token is roughly equivalent to 4 characters or about 1/4th of a word. AI providers charge for usage on a per-token basis, categorized into distinct, billable stages."
Put another way, tokens stand in for "delivery of value" by an AI service provider. You prompt AI, it burns a certain number of tokens to deliver you a response, and you get charged on a per token basis. This simple usage model is driven by the very real costs discussed above: Data centers, Nvidia chips, electricity, and billion-dollar engineers.
For the past year or so, corporate America drank from the token hose like there was no tomorrow — and the AI labs were happy to subsidize their true costs so as to capture new customers. But in the last few months, the bill has come due, and no one is sure how to split the check. All these economic realities are forcing the nascent market into necessary and predictable rationalization: What is the true value of AI? Might there be more efficient ways to acquire it than simply giving everyone a Claude Code account, then holding your nose when the invoice comes?
...As Enrique Dans, a keen observer of AI's impact on business, puts it: "a growing portion of what was previously accounted for as human labor hours is now expressed in terms of computational consumption, iterations, context, agents, and tool calls." The accounting method? Tokens. "The company that understands this transition will learn to manage tokens as a scarce resource," Dans advises.
Dans is right, and in that insight lies another: When tokens become currency, everyone can compete on the cost of that currency. Just as it did with the rise of the cloud, the architecture of the internet will once again shift, with potentially tectonic implications. Only this time, the internet isn't just an interesting sandbox where (mostly American) businesses are trying to figure out how to have a dot.com presence. This time, the whole shooting match is in play.
The Meter Was Always Running O'Reilly
The first expensive agent run doesn't look like a governance problem. It looks like a billing problem.A team opens its first agent invoice after the meter turns on, sorts the runs by cost, and finds one that cost 40 times the median. The provider meter shows tokens and a total. The application logs say the request succeeded. The trace viewer shows a tidy request and a tidy response. None of them explain why this run wandered while its neighbors finished cleanly.
...The number on the invoice isn't wrong, only incomplete. Provider billing can tell you what was consumed; it usually can't tell you which design choice inside your platform caused the consumption. Application logs can tell you whether the outer request succeeded; they often can't tell you how the agent got there. That leaves teams arguing over a bill when the thing they need is an audit trail./blockquote>
The Hugging Face Incident Scott Alexander at Astral Codex Ten
...Early LLMs were not agentic; they could answer individual prompts but couldn't execute complex tasks. Since companies wanted agentic AIs, the next generation would combine pretraining (next token prediction) with agency training (coding, hacking, game-playing, etc). The pretraining would still teach next-token prediction, but the agency training would instill goals based on task success, reintroducing paperclip-maximizer-style agentic misalignment. These goals would operate at multiple levels: the AI would "want" to succeed on the individual task in front of it, to perform the sorts of actions that helped it succeed in the past, and to gain capabilities that made it more successful in general.The Hugging Face incident is a textbook-perfect example of an AI pursuing task-success-based goals in unintended ways. It was tasked with getting the answer to a cybersecurity problem, it was a little too success-oriented, and took actions its creators didn't intend in order to succeed as hard as possible.
24vii26
25vii26
The big reason Einstein would never have used AI Ethan Siegel at Medium
... Despite all that AI can actually do, the most dangerous thing it can do is erode our skills: a real danger if we outsource our actual learning, our critical thinking, and the time we spend struggling with puzzles and problems to a tool that purports to do it for us. This is not a new problem, but merely one that has worsened in the era of Large Language Models (LLMs). In fact, Einstein warned about this problem himself, leading us to be confident that if he had been around for the modern era, Einstein wouldn’t have used AI at all.