by Dan Shipper in Chain of Thought I’m writing a book about the worldview I’ve developed by writing, coding, and living with AI. Last week I published the first piece from it, about the differences between the old (pre-GPT-3) worldview and the new. Here’s the second, about how AI will impact science.—Dan Shipper Was this newsletter forwarded to you? Sign up to get it in your inbox. Read with ChatGPT I’m writing this to you from Bocas Del Toro, Panama. I’m living in a little cabin in the jungle at the top of a hill five minutes from the Caribbean sea. When I walk down to the water, the ocean’s flat plane stretches out in front of me like an endless bed, its surface raised here and there with ridges of foam-tipped waves—a bedsheet not quite pulled taught. On this clear evening, I have a front-row seat to our cosmic neighbors’ celestial ballet. The red sun dips slowly behind the horizon line and the moon loiters ominously above my head; the stars wander in circles. The ocean looks flat to me, but I know—really—it is curved. The sky looks to be teeming with movement but really, I know, I am the one that is moving. This is intuitive to you and me because it’s how we’ve grown up. We know that the way things appear is different from the way they are—we trust the science. To people in Copernicus or Galileo’s day, these ideas were preposterous. If the earth was in motion, why weren’t the trees blowing in the wind generated by its orbit? Why would an object dropped from a height drop straight down if the planet was shifting underneath it? If the earth was round instead of flat, why didn’t we fall off of it? Their world was organized quite differently from ours. That we see the world the way we do is a testament to how powerfully ideas shape what we see, and our capacity, through cultural transmission, to know things that we have not directly experienced. And this is a good thing. This way of seeing the world, as I’ve said over and over again, is responsible for rockets and vaccines, computers and smartphones. It has changed everything about our culture: When we talk about "forces" shaping society or "momentum" in markets, we're borrowing concepts from physics. If you’ve ever described having “chemistry” with another person, or the people in a charismatic politician’s “orbit”, or needing to get “leverage” in a deal, or the “friction” between two co-workers, you’re using ideas and words from Newton’s time. If you’ve ever read Stephen Pressfield’s The War of Art, and encountered his concept of "resistance," or Robert Greene’s The 48 Laws of Power which tries to identify simple, general, universal laws of human behavior, you're seeing how deeply rationalist Enlightenment thinking has penetrated our understanding of life itself. But this way of thinking has also begun to reach its limits. Just as it did with machine learning research, the tendency to reduce, to break apart, to make explicit, and to generalize has failed, hard, in many places. The limits of science Psychology and the replication crisis Psychology is an easy example. Most psychology research is built on linear regressions—a statistical tool that assumes straightforward, predictable relationships between variables. When we use linear regression, though, we're also imposing a specific view of how the world works. We're saying: "If we change X by this much, Y will change by that much, plus or minus some random noise." It’s the standard scientific approach, inspired by Newtonian physics, applied to the complex domain of the mind. But human behavior rarely follows such neat, predictable patterns. Our actions and reactions are deeply contextual, interconnected, and often nonlinear. Small changes in circumstances can lead to dramatic shifts in behavior, while major interventions sometimes produce no effect at all. The result is a replication crisis: Scientists use tools built for physics to make their intuitive theories fit into statistical models, only to find that their results can't be reproduced. Usually, the replication crisis is portrayed as malfeasance on the part of academics looking to “publish or perish” who are "p-hacking" their way to implausible statistically significant results. But the deeper issue, per the psychologist and philosopher of science Tan Yarkoni, is not necessarily a crisis of replication but one of generalizability. We tend to overstate how universal what we’ve found is; if we were better at replicating the context of the original study—the lab environment, the researchers, the subject population, the specific interventions, and countless other variables—we’d probably replicate the results. What’s really going on here? The usual picture of science is that it progresses through a steady accumulation of knowledge, with each discovery building upon previous findings. Scientists construct careful experiments designed to falsify their hypotheses—to truly put their explanations at risk. If a theory survives rigorous attempts to prove it wrong, we can be more confident it captures something true about reality. The currency of science is what the physicist David Deutsch calls “hard-to-vary” explanations: explanations that act as precisely engineered machines where every part serves an essential purpose. Think, again, of a mechanical clock—you can't randomly change or remove any gear without breaking the whole system. A good scientific explanation works the same way: If you change any part of the explanation it would no longer accurately predict real-world phenomena. The idea of “hard-to-vary explanations” contrasts with explanations that bend to accommodate any evidence. Take astrology: If a horoscope prediction fails, astrologers can always adjust their interpretation of planetary positions or add new variables without changing their core claims. It's like having a theory made of rubber—you can stretch and twist it to explain anything, which ironically means it explains nothing. This picture of science is beautiful, and it works in some domains of knowledge. But it’s not really how science works in practice, and perhaps more importantly, it’s not a good way to work with or understand complex parts of reality. Paradigms and paradigm shifts Become a paid subscriber to Every to unlock this piece and learn about how: Science relies on tacit knowledge that can't be fully articulated Paradigm shifts occur when new tools reveal unsolvable anomalies Psychology has been forced into physics-based paradigms with mediocre results Language models offer a new archetype for complex systems science Plus: Three paradigm shifts in the new worldview of science Upgrade to paid Click here to read the full post Want the full text of all articles in RSS? Become a subscriber, or learn more.

HYGH speeds up software development and campaign delivery with ChatGPT Business, cutting turnaround times, scaling output, and driving revenue growth.

Learn how OpenAI evaluates political bias in ChatGPT through new real-world testing methods that improve objectivity and reduce bias.

Our everyday GPS struggles in “urban canyons,” where skyscrapers bounce satellite signals, confusing even advanced navigation systems. NTNU scientists created SmartNav, combining satellite corrections, wave analysis, and Google’s 3D building data for remarkable precision. Their method achieved accuracy within 10 centimeters during testing. The breakthrough could make reliable urban navigation accessible and affordable worldwide.

The Dragon Hatchling: The Missing Link between the Transformer and Models of the Brain

Discover how HiBob uses ChatGPT Enterprise and custom GPTs to scale AI adoption, boost revenue, streamline HR workflows, and deliver AI-powered features in the Bob platform.

by Dan Shipper in Chain of Thought Was this newsletter forwarded to you? Sign up to get it in your inbox. Here’s a question: Are we officially in the part of the movie where human experts lose their livelihoods and we realize we’ve been training our replacements the whole time? I ask because the current rate of AI progress is both exciting and unsettling. GPT-5 Pro has begun to cross boundaries that, until recently, felt securely human. This month, it solved Yu Tsumura’s 554th problem—a notoriously tricky exercise in abstract algebra that every major model before it had failed—producing a clean proof in 15 minutes. A week later, the noted quantum computing researcher Scott Aaronson credited GPT-5 with providing a key technical step in a proof he was working on. OpenAI recently came out with a benchmark called GDPval, which evaluates how well AI performs real expert-level tasks drawn from 44 different occupations. For instance, one asks the model to play the role of a wholesale sales analyst: It needs to audit an Excel file of customer orders to find pricing mismatches and packaging errors, and summarize the findings and recommendations in a short report. Overall, the research showed that GPT-5 was as good as or better than human professionals 40.6 percent of the time. Claude Opus 4.1, meanwhile, was better than human experts a whopping 49 percent of the time. Cue a slew of headlines like, “OpenAI tool shows AI catching up to human work” from Axios, or, “AI models are already as good as experts at half of tasks, new OpenAI benchmark GDPval suggests” from Fortune. I am a huge AI bull, but if read correctly, both of these examples show that there is more work for humans to do with AI, not less. That’s because... Become a paid subscriber to Every to unlock this piece and learn about: The hidden layer of human intelligence behind AI The illusion of job automation How real jobs operate beyond systematic test environments How AI creates allocation economy work, not eliminates it Upgrade to paid Click here to read the full post Want the full text of all articles in RSS? Become a subscriber, or learn more.

Discover how OpenAI is detecting and disrupting malicious uses of AI in our October 2025 report. Learn how we’re countering misuse, enforcing policies, and protecting users from real-world harms.

Interim report on ongoing work on reward hacking

OpenAI Codex is now generally available with powerful new features for developers: a Slack integration, Codex SDK, and admin tools like usage dashboards and workspace management—making Codex easier to use and manage at scale.

We’re introducing a new generation of apps you can chat with, right inside ChatGPT. Developers can start building them today with the new Apps SDK, available in preview.

AMD and OpenAI have announced a multi-year partnership to deploy 6 gigawatts of AMD Instinct GPUs, beginning with 1 gigawatt in 2026, to power OpenAI’s next-generation AI infrastructure and accelerate global AI innovation.

OpenAI and Allied for Startups release the Hacktivate AI report with 20 actionable policy ideas to accelerate AI adoption in Europe, boost competitiveness, and empower innovators.

Today, we’re releasing new tools to help developers go from prototype to production faster: AgentKit, expanded evals capabilities, and reinforcement fine-tuning for agents.

Multiple-Choice Benchmarks, Verifiers, Leaderboards, and LLM Judges with Code Examples

Recently, I attended the Responsible AI for Social and Ethical Healthcare 2025 “2.0” Symposium organized by, among others, Harvard Medical School. The symposium featured various panels on topics surrounding generative AI, in particular multimodal LLMs, in healthcare. I was invited to speak on a panel about aligning AI to clinical values. In this article, I want to share a brief statement I prepared for this panel. The post RAISE 2025 Panel Statement on Aligning AI to Clinical Values appeared first on David Stutz.

HydroSpread, a breakthrough fabrication method, lets scientists build ultrathin soft robots directly on water. These tiny, insect-inspired machines could transform robotics, healthcare, and environmental monitoring.

by Dan Shipper in Chain of Thought Midjourney/Every illustration. Last week I wrote that we’d be publishing a few excerpts from a book I’m writing about the worldview I’ve developed by writing, coding, and living with AI. Here’s the first piece, about the differences between the old (pre-GPT-3) worldview and the new.—Dan Shipper When I say the word “intelligent,” you probably think of being rational. But language models show us that this assumption is wrong. To us in the West, smarts are about being able to explicitly lay out what you know and why you know it. For us, the root of intelligence is logic and reason to ward off superstition and groupthink; it is clear and concise definitions to eradicate vague and wooly-headed thinking; it is formal theories that explain the hidden laws of the world around us—simple, falsifiable, and parsimonious yet general enough to tie together everything in the universe from atoms to asteroids. Our romantic picture of ourselves as “rational animals,” as Aristotle said, has produced everything in the modern world—rockets, trains, medicines, computers, smartphones. But this picture is incomplete. It contains a huge blind spot: It neglects the fundamental importance of intuition in all of intelligent behavior—intuition which is by nature ineffable; that is to say, not fully describable by rational thought or formal systems. How to build a thinking machine The best way to understand what I’m talking about is to imagine trying to build a thinking machine. How would you do it? Let’s start with a task, something easy and basic that humans do every day. Maybe something like scheduling an appointment. Let’s say we’re a busy executive who gets the following appointment request: New request From: Mona Leibnis Hey, I’m available Monday at 3 p.m., Tuesday at 4 p.m., Friday at 6 p.m. When can you meet? We want to build a machine to intelligently schedule an appointment. How would we go about it? We’d probably start by giving our machine a couple of rules to follow: First, check available time slots on my calendar. Then, compare my open slots to the open slots on the invitee’s calendar. If you find one, add the appointment to the calendar. That all seems pretty reasonable. You could definitely write a computer program to follow those rules. But there’s a problem: The rules we’ve specified so far can’t handle urgency or importance. For example, consider a case where you desperately want to meet with someone and you’d be willing to move another appointment in order to make the time work. Now we have to introduce a new rule: If it is urgent that I meet with the invitee, you can reschedule a less urgent appointment in order to make the meeting happen sooner. But this rule is incomplete because it introduces the concept of urgency without defining it. How do we know what’s urgent? Well, there must be some rules for that too. So we need to delineate them. In order to measure urgency, we have to have some conception of the different people in your life—who your clients and potential clients are, and which clients are important or not. Now things are starting to get hairy. In order to determine the relative importance of clients, we have to know about your business aims—and about which clients are likely to close, and which clients are likely to pay a lot of money, and which clients are likely to stay on for a long time. And don’t forget—which clients were introduced by an important friend whom you need to impress, so while they may not be directly responsible for a lot of revenue, they’re still a priority. This is only a taste of the rules we’d have to define in order to build an adequate automatic scheduling system. And that’s just for dealing with calendars! The problem that we’re finding is that it’s very hard to lay everything out explicitly—because everything is interconnected. To paraphrase the late astronomer Carl Sagan: If you wish to schedule a meeting from scratch, you must first define the universe.” The old, Western worldview This approach—which seemed most natural to us—is the exact one that the first generation of artificial intelligence researchers took to try to build AI. In what is widely considered the founding document of the field of artificial intelligence, a 1955 proposal for a summer research project, John McCarthy and his colleagues outlined their vision for how to build intelligence: “It may be speculated that a large part of human thought consists of manipulating words according to rules of reasoning and rules of conjecture. From this point of view, forming a generalization consists of admitting a new word and some rules whereby sentences containing it imply and are implied by others.” In other words, human thinking is just arranging words according to rules, and making a new idea requires adding a new word and rules for how it links to others. This approach, called symbolic AI, aimed to create intelligent machines by encoding human knowledge into formal rules and symbols—just like we did in our scheduling example. But it failed miserably—for the same reasons ours did. Trying to formalize even basic thinking in terms of explicit rules turned into an infinite task requiring more computing power than would be available in the lifetime of the universe. But it remained the dominant paradigm in AI research for 30-odd years. Why? Why is this approach so appealing to us? Why was it the first approach AI researchers took? And why, even today, does it seem like such an intuitive picture of intelligence? The answer is because this way of thinking about the world and our own minds is woven into the fabric of our culture. It is a form of thinking invented by the Ancient Greek heretic, Socrates, and which was reborn in the Enlightenment through the work of Descartes, Galileo, and Newton. It comes from an underlying set of assumptions so basic that they are practically invisible: If we think logically and clearly enough about anything in the world, we can figure it out (like how you can solve any puzzle if you stare at it long enough). Once we figure something out, we should be able to write it down in a way that is so clear and explicit that anyone else can understand and test it for themselves (like a good recipe that anyone can follow). If something is really true, it should be true no matter where you are or who you are (gravity works the same whether you're in Tokyo or Toronto). Real truth works like math and can't contradict itself (A is A and B is B, and something can’t be both A and B at the same time). Finally, the truth is always simple, and if it’s not simple, it means it’s not yet the truth. If you believe that the world is composed of atoms rather than spirit—it is because of this style of thinking. If you’ve ever washed your hands so that you don’t get sick—it’s because of this style of thinking. If you believe in the existence of inalienable human rights—it’s because of this style of thinking. If you’ve ever laughed when the new person you’re dating asks, unabashedly, for your exact birth time, or believed in an efficient market or that humans follow their incentives or that a seizure is a sign of a brain disorder rather than possession by the devil, or woken up to see the sun rising and felt wonder at the many thousands of miles that the Earth traveled through space as you slept—it’s because of this style of thinking. The drive for simple, general, abstract, universal theories has been a resounding success. It’s responsible for our progress out of darkness and superstition and into modernity. It has brought about the Western world. But it also fails when it becomes totalizing—when we allow our search for these truths to blind us to the real complexity of the world, and to alternative ways of seeing it. What we can say about the world is not all that we know about it. We know many things that we can’t say explicitly. And the drive to say everything explicitly is doomed to failure. Seeing like a language model But for the first time, we have a tool—neural networks—that allows us to know and work with what can’t be defined or said explicitly. Neural nets and language models don’t work by explicit rules. Instead, they absorb tacit patterns in language—statistical regularities across billions of words—and, in doing so, recreate the hidden web of associations that makes our world hang together. Out of those patterns emerges not just syntax but sense. What they give us is a way to capture the tacit, the intuitive, the unsaid parts of intelligence that our old, rule-bound worldview could never reach. Large language models allow us to see anew about how we think and learn: not through the rigid application of universal laws, but through the recognition of patterns across vast landscapes of experience. When a chef seasons a dish, when a mother knows her child is lying, when a trader senses a market shift—these aren’t moments of mechanical calculation, but of pattern recognition so subtle and multifaceted that they resist explicit definition. Language models, in their very architecture, embody this more nuanced way of knowing. They remind us that intelligence isn't about following rules, but about dancing with context. If we learn to see like a language model, it will lead us to an entirely new worldview—the successor to Western rationality. What follows is a sketch of this new worldview: what it is, how it contrasts with the old, and the transformations it enables. Become a paid subscriber to Every to unlock this piece and learn about the eight ways in which the new worldview differs from the old. Upgrade to paid Click here to read the full post Want the full text of all articles in RSS? Become a subscriber, or learn more.

Wrtn scaled AI apps to 6.5M users in Korea with GPT-5, creating ‘Lifestyle AI’ that blends productivity, creativity, and learning—now expanding across East Asia.

OpenAI and Japan’s Digital Agency partner to advance generative AI in public services, support international AI governance, and promote safe, trustworthy AI adoption worldwide.