Arvind Narayanan
Professor of Computer Science at Princeton University and director of the Center for Information Technology Policy; co-writes AI as Normal Technology with Sayash Kapoor.
Arvind Narayanan did not write this page. ppll compiled it from their own words, each linked to the source. Is this you? · Report a mistake
Latest x.com
What they believe
-
An AI agent can only have two of three properties at once: being general-purpose, being deployed in high-stakes settings, and being fully automated.
For the time being, you can have only two out of these three properties in agents: general-purpose (a language model based agent that can be instructed to do different tasks rather than purpose-built for a task like traditional software); deployed in high-stakes scenarios, and automated.
normaltech.aiSaid 13 Jul 2026This on korrents
-
Infrastructure providers rarely capture the value they create.
First, infrastructure providers rarely capture the value they create.
normaltech.aiSaid 9 Jul 2026This on korrents
The bold line is korrents’ wording; the quote is theirs. These 2: what they wrote before what they said aloud, one per source, newest first.
Feed
As its own page →Hiding
20 November
19 November
11 November
10 November
5 November
4 November
3 November
2 November
30 October
29 October
27 October
24 October
23 October
22 October
20 October
14 October
9 October
-
Recommendsrcmnd.app
Not the End of the WorldTheir words
So maybe in keeping with that, I really like Hannah Ritchie's book, Not the End of the World. She has a newer book, but this one's from 2024 and I still like it very much. The subtitle is something like how we can be the first generation to build a sustainable planet. It's an optimistic take on climate, which is of course usually full of doom and gloom stories, so I really liked it for that reason.
-
Recommendsrcmnd.app
BreakneckTheir words
I liked Dan Wang's uh, Breakneck. A lot of echoes of abundance as well. But this idea of thinking about a lawyerly society versus an engineering society was I thought a really good and succinct way to capture a lot of the macro and micro differences.
1 October
28 September
13 July
From one piece What will be left for us to work on? 3 beliefs · normaltech.ai
-
Their words
AI is still in a state where it is much better at verifiable tasks than non-verifiable tasks.
-
Their words
I think it is already the case in software engineering, and will increasingly be the case in many professions, that we can think of knowledge workers as similar to a crane operator or a forklift operator.
-
Their words
For the time being, you can have only two out of these three properties in agents: general-purpose (a language model based agent that can be instructed to do different tasks rather than purpose-built for a task like traditional software); deployed in high-stakes scenarios, and automated.
9 July
From one piece Up the Stack: How AI’s Escape From the Commodity Trap Risks Enterprise Lock-in 3 beliefs · normaltech.ai
-
Their words
We believe that AI, at least in its current hyperscaler form, risks falling into the same commodity trap that has bedeviled so many previous infrastructure builders.
-
korrents.com
AI will become an essential part of most knowledge work and will be as transformative as the industrial revolution.Their words
Over the long run, we expect that AI will be an essential part of most knowledge work, and that the technology will be as transformative as the industrial revolution.
-
Their words
First, infrastructure providers rarely capture the value they create.
11 June
From one piece Why AI hasn’t replaced software engineers, and won’t 3 beliefs · normaltech.ai
-
korrents.com
The real barrier to programming is exercising sound judgment and accountability, not learning syntax.Their words
The barrier isn't actually learning the syntax. It's having enough skilled judgment to make good decisions while maintaining accountability.
-
korrents.com
Production software cannot be built by unskilled vibe coders replacing trained software engineers.Their words
companies can't ship production software by hiring unqualified vibe coders instead of software engineers.
-
Their words
AI has compressed the middle of the sandwich, but has left the two ends largely unchanged.
29 January
From one piece Fact checking Moravec's paradox 3 beliefs · normaltech.ai
-
Their words
Since diffusion of new capabilities takes a long time, that gives us plenty of time to react - time that we often squander, and then panic!
-
korrents.com
Superhuman AI reasoning is a myth.Their words
Much of my writing is based on the idea that superhuman AI reasoning is a myth.
-
Their words
It is really a statement about what the AI community finds it worthwhile to work on. It doesn't have any predictive power about which problems are going to be easy or hard for AI.
26 January
22 January
21 January
9 September 2025
From one piece A guide to understanding AI as normal technology 3 beliefs · normaltech.ai
-
Their words
We predict that AI will not be able to meaningfully outperform trained humans (particularly teams of humans and especially if augmented with simple automated tools) at forecasting geopolitical events (say elections).
A guide to understanding AI as normal technologynormaltech.ai
-
AI alignmentdesignrecursive self-improvement
Their words
we don't think RSI will lead to superintelligence because the external bottlenecks to building and deploying powerful AI systems cannot be overcome by improving their technical design.
A guide to understanding AI as normal technologynormaltech.ai
-
Their words
There is a long causal chain between AI capability increases and societal impact. Benefits and risks are realized when AI is deployed, not when it is developed.
A guide to understanding AI as normal technologynormaltech.ai
15 April 2025
Nothing matches.
Beliefs
Korrents What they believe 23 beliefs — each backed by an exact quote.
Each is a — compiled by korrents.com, not by them: the one-line wordings are korrents', the quotes are theirs.
Recent
An AI agent can only have two of three properties at once: being general-purpose, being deployed in high-stakes settings, and being fully automated.
For the time being, you can have only two out of these three properties in agents: general-purpose (a language model based agent that can be instructed to do different tasks rather than purpose-built for a task like traditional software); deployed in high-stakes scenarios, and automated.
What will be left for us to work on? Said 13 Jul 2026
Knowledge workers are becoming operators of AI systems, similar to how a crane or forklift operator controls a machine doing the physical work.
I think it is already the case in software engineering, and will increasingly be the case in many professions, that we can think of knowledge workers as similar to a crane operator or a forklift operator.
What will be left for us to work on? Said 13 Jul 2026
AI systems are much better at tasks whose correctness can be verified than at tasks whose correctness cannot be checked.
AI is still in a state where it is much better at verifiable tasks than non-verifiable tasks.
What will be left for us to work on? Said 13 Jul 2026
Show 20 more
Infrastructure providers rarely capture the value they create.
First, infrastructure providers rarely capture the value they create.
Up the Stack: How AI’s Escape From the Commodity Trap Risks Enterprise Lock-in Said 9 Jul 2026
AI will become an essential part of most knowledge work and will be as transformative as the industrial revolution.
Over the long run, we expect that AI will be an essential part of most knowledge work, and that the technology will be as transformative as the industrial revolution.
Up the Stack: How AI’s Escape From the Commodity Trap Risks Enterprise Lock-in Said 9 Jul 2026
AI, in its current hyperscaler/infrastructure form, risks falling into the same commodity trap that has affected past infrastructure industries.
We believe that AI, at least in its current hyperscaler form, risks falling into the same commodity trap that has bedeviled so many previous infrastructure builders.
Up the Stack: How AI’s Escape From the Commodity Trap Risks Enterprise Lock-in Said 9 Jul 2026
AI automates the execution phase of software engineering but leaves decision-making and accountability largely to humans.
AI has compressed the middle of the sandwich, but has left the two ends largely unchanged.
Why AI hasn’t replaced software engineers, and won’t Said 11 Jun 2026
Production software cannot be built by unskilled vibe coders replacing trained software engineers.
companies can't ship production software by hiring unqualified vibe coders instead of software engineers.
Why AI hasn’t replaced software engineers, and won’t Said 11 Jun 2026
The real barrier to programming is exercising sound judgment and accountability, not learning syntax.
The barrier isn't actually learning the syntax. It's having enough skilled judgment to make good decisions while maintaining accountability.
Why AI hasn’t replaced software engineers, and won’t Said 11 Jun 2026
Moravec's paradox reflects which problems AI researchers find worth working on, not a real predictor of what is easy or hard for AI.
It is really a statement about what the AI community finds it worthwhile to work on. It doesn't have any predictive power about which problems are going to be easy or hard for AI.
Fact checking Moravec's paradox Said 29 Jan 2026
Superhuman AI reasoning is a myth.
Much of my writing is based on the idea that superhuman AI reasoning is a myth.
Fact checking Moravec's paradox Said 29 Jan 2026
The slow diffusion of new AI capabilities gives society time to react to them, though that time is usually wasted before panic sets in.
Since diffusion of new capabilities takes a long time, that gives us plenty of time to react - time that we often squander, and then panic!
Fact checking Moravec's paradox Said 29 Jan 2026
AI's societal benefits and risks are determined by how it is deployed, not by how capable the underlying models become.
There is a long causal chain between AI capability increases and societal impact. Benefits and risks are realized when AI is deployed, not when it is developed.
A guide to understanding AI as normal technology Said 9 Sept 2025
Recursive self-improvement in AI will not produce superintelligence because bottlenecks outside the AI system's own technical design limit what it can achieve.
we don't think RSI will lead to superintelligence because the external bottlenecks to building and deploying powerful AI systems cannot be overcome by improving their technical design.
A guide to understanding AI as normal technology Said 9 Sept 2025
Teams of trained humans, especially when aided by simple automated tools, will continue to match or outperform AI at forecasting geopolitical events such as elections.
We predict that AI will not be able to meaningfully outperform trained humans (particularly teams of humans and especially if augmented with simple automated tools) at forecasting geopolitical events (say elections).
A guide to understanding AI as normal technology Said 9 Sept 2025
AI should not be understood as a humanlike or potentially superintelligent agent, now or in the foreseeable future.
We do not think that viewing AI as a humanlike intelligence is currently accurate or useful for understanding its societal impacts, nor is it likely to be in our vision of the future.
AI as Normal Technology Said 15 Apr 2025
The speed of AI's economic impact is limited by how fast individuals, organizations, and institutions can adapt, not by how fast the technology itself improves.
Thus, the speed of diffusion is inherently limited by the speed at which not only individuals, but also organizations and institutions, can adapt to technology.
AI as Normal Technology Said 15 Apr 2025
The 'bitter lesson' that general methods eventually beat human domain knowledge does not hold for building AI-based products.
In the context of AI-based product development, the bitter lesson has never been even close to true.
AI as Normal Technology Said 15 Apr 2025
Industry leaders' predictions about AI are not more reliable than outside observers' because they are shaped by vested commercial interests.
Regardless of whether model scaling will continue, industry leaders' flip flopping on this issue shows the folly of trusting their forecasts. They are not significantly better informed than the rest of us, and their narratives are heavily influenced by their vested interests.
Is AI progress slowing down? Said 18 Dec 2024
AI's real-world impact is bottlenecked by the pace of product development and the rate of adoption, not by further gains in AI capabilities.
The connection between capability improvements and AI's social or economic impacts is extremely weak. The bottlenecks for impact are the pace of product development and the rate of adoption, not AI capabilities.
Is AI progress slowing down? Said 18 Dec 2024
Whether AI capability progress has slowed down cannot be settled, since reasonable observers can interpret the same evidence differently.
It is impossible to predict whether progress in AI capabilities will slow down. In fact, forget prediction - reasonable people can have very different opinions on whether AI progress has already slowed down, because they can interpret the evidence very differently.
Is AI progress slowing down? Said 18 Dec 2024
The target variable an algorithm is trained to predict matters more for fairness than which input features it uses.
In reality, the target variable - the output - is often more important for fairness than the features.
Does the UK’s liver transplant matching algorithm systematically exclude younger patients? Said 11 Nov 2024
Utilitarian reasoning has become dominant across many domains, including medical ethics and public health, over recent decades.
For this reason, utilitarianism has been ascendant in many, many domains over the last few decades, including medical ethics and public health.
Does the UK’s liver transplant matching algorithm systematically exclude younger patients? Said 11 Nov 2024
What is a korrent?
A korrent is a belief a person has stated in their own words: one sentence stating the claim, backed by a quote and a source, kept at korrents.com.
Under a name here, the quoted block is what they actually said. The korrent beneath it is the claim those words support, in korrents' wording — tap it to see the record, its source, and who else holds it.
Nobody here wrote their own korrents. They are compiled from public statements, and a person can change their mind, which is recorded too.
Recommends rcmnd
rcmnd What they recommend 6 things — each backed by an exact quote.
Compiled by rcmnd.app, not by them: the quotes are theirs, and each links to the post it was copied from.
Recent
Recommends Breakneck
I liked Dan Wang's uh, Breakneck. A lot of echoes of abundance as well. But this idea of thinking about a lawyerly society versus an engineering society was I thought a really good and succinct way to capture a lot of the macro and micro differences.
youtube.com 9 Oct 2026
Recommends Not the End of the World
So maybe in keeping with that, I really like Hannah Ritchie's book, Not the End of the World. She has a newer book, but this one's from 2024 and I still like it very much. The subtitle is something like how we can be the first generation to build a sustainable planet. It's an optimistic take on climate, which is of course usually full of doom and gloom stories, so I really liked it for that reason.
youtube.com 9 Oct 2026
Uses iPad
My wife and I have a pro-screen time approach to parenting, and our 3-year old has long had her own iPad.
normaltech.ai 14 Apr 2023
Show 3 more
Disliked Apple text-to-speech
But for voice output, Apple’s native text-to-speech feature is unfortunately atrocious, so I had to plug it into Google’s service instead.
normaltech.ai 14 Apr 2023
Uses Apple Dictation
For voice input, I found that Apple dictation is good enough, so I went with that.
normaltech.ai 14 Apr 2023
Liked ChatGPT
ChatGPT was extremely helpful for doing the programming, especially dealing with unfamiliar APIs, and cut the development time a lot.
normaltech.ai 14 Apr 2023
Where they publish
Site Arvind Narayanan — Princeton His own personal/faculty page, not the department's.
Link verified 25 Sept 2026.
Newsletter AI as Normal Technology Co-written with Sayash Kapoor; formerly AI Snake Oil.
Recent
- Steering AI Beyond Automation 14 Oct 2026 A Research Agenda and Plan for Action
- A big-tent or small-tent AI safety movement? 1 Oct 2026 The unstated disagreement that underpins safety debates
- AI existential risk probabilities are (still) too unreliable to inform policy 28 Sept 2026 How speculation gets laundered through pseudo-quantification
Show 6 more
- What will be left for us to work on? 13 Jul 2026 My keynote at ICML 2026
- Up the Stack: How AI’s Escape From the Commodity Trap Risks Enterprise Lock-in 9 Jul 2026 Critics and boosters are both looking in the wrong place
- Why AI hasn’t replaced software engineers, and won’t 11 Jun 2026 Coding agents as normal technology
- Fact checking Moravec's paradox 29 Jan 2026 This famous aphorism is neither true nor useful
- A guide to understanding AI as normal technology 9 Sept 2025 And a big change for this newsletter
- AI as Normal Technology 15 Apr 2025 A new paper that we will expand into our next book
Link verified 25 Sept 2026. Recent items update automatically from the channel.
x.com @random_walker Posts on AI research and policy.
Recent
- Reminder: racing to self-improve the verifiable subset of capabilities, which is what AI companies are currently doing, _decreases_ the window of time that they have to convert into sustainable businesses and do the boring work of turning value creation into value capture. (Of course, their bet is that progress in verifiable tasks will somehow yield progress on unverifiable tasks. We'll see.) 20 Nov 2026
- In most of the recent cases of website outages, when AI was involved at all, it was a matter of AI improving cybersecurity and safety by helping defenders find and fix long-standing bugs, even if it meant briefly taking down systems to roll out updates. For two decades, automated vulnerability detection has helped defenders over attackers, and AI's impact so far is mostly consistent with that trend. This is no reason for complacency. There is a long way to go, with a huge amount of accumulated security debt to fix, and tail risks remain possible. But the "AI is causing outages" narrative is t… 19 Nov 2026
- There have been awful cases of teen mental health harms and suicides from chatbots like GPT-4o and Character AI that were sycophantic and misbehaving. But these were all text-based, and extremely rudimentary as “companions”. I’ve talked to many people trying to hold the industry to account for teen safety and almost none are thinking about the coming wave. Completely realistic video avatars are already here, but expensive today and limited to enterprise applications. Within a year or two they will be cheap enough for always-on, ad-supported video companions. What happens then? Uptake of AI co… 11 Nov 2026
Show 16 more
- 📢 We need more orgs that take AI’s transformative potential seriously *and* take downstream bottlenecks seriously. We have just launched STABL, and hope to lead by example. Read our launch essay below and check us out at https://t.co/P55nL7Mk6d. The three of us have long felt that society’s responses to advanced AI are unproductively polarized. So we’ve decided to do something about it. Our work will have four pillars: Recursive Self Improvement, risks and resilience, steering AI’s economic impacts, and sectoral and professional adaptation. We will bring our distinctive methods and approach… 10 Nov 2026
- Most days I can just about manage this balancing act, but only because thinking about how we should adapt to AI is my actual job. Otherwise I think I’d find it too hard. I suspect that’s one reason people sort themselves into camps or tribes. It’s psychologically easier than constantly revisiting one’s beliefs. Being able to empathize with and collaborate with people who have different AI boundaries and attitudes than our own is becoming an important skill. 5 Nov 2026
- Working with AI requires perpetually operating at the edge of sanity. You have to recalibrate your mental model of AI capabilities and your own job description every few months, without getting delusional about seemingly infinite productivity gains, or feeling threatened by losing what you thought was the uniquely human part of your job, or obsessing over the perfect setup, or losing your mind over the clumsiness and jaggedness of agents, or ... Maybe AI literacy needs to be less about prompting and more about improving emotional regulation. 4 Nov 2026
- AI revenues right now are in a state of extreme disequilibrium and are produced by multiple counteracting exponentials. Naive trend extrapolation is not insightful. For forecasts to be useful we have to tease apart the underlying trends and how they interact, and pay close attention to the distinction between value creation and value capture. Many widely divergent futures are possible, including some pretty wild ones. 3 Nov 2026
- As AI labs race toward Recursive Self Improvement, there’s a possibility that what they are actually racing toward is Recursive Self Extinction — of their business model. Here’s my reasoning. 1. Progress on verifiable tasks continues. In particular, improving the cost efficiency of existing capabilities is a highly verifiable task. For the past few years, cost per equivalent capability has dropped by at least 10x per year. I think this progress will be sustained for a few more years, and even accelerate as labs use AI more and more heavily for AI research in the pursuit of RSI. There is of co… 2 Nov 2026
- Resharing our paper on the science of AI agent evaluation because the PHP vulnerability happened right as we announced the paper last week and our server was taken down all day to upgrade PHP 🫠 PDF: https://t.co/8m13p9d70F Webpage: https://t.co/xGAvBCGW4a Blog: https://t.co/qxO21qWZfm There are growing calls for independent and embedded AI evaluation, especially of AI agents. But much remains unclear: what constitutes a “good” evaluation, what needs to be evaluated in the first place, and how best to coordinate and support the influx of people entering the burgeoning evaluation ecosystem. Ou… 30 Oct 2026
- What would the writing of superintelligent AI sound like? You might expect it to have the characteristics of good human writing — figurative language, high information density, coining new terms, and punchiness — but dialed up to 11. Well, that’s what Claude did, until Anthropic dialed it back recently because of user complaints. It didn’t sound superintelligent to most of us. Instead, the neuralese writing style sounds like a “stupid person’s idea of a smart person”. Our response to information density and metaphor-laden writing looks something like the cartoon graph I've drawn. Beyond a poi… 29 Oct 2026
- Such a great post! Insights I particularly appreciated: - "Why hasn’t a huge discontinuity in damages happened since the dawn of scaled AI? This non-responsiveness of damages to a rising tide of AI capabilities is the major conceptual crux, and stumbling block, of all AI security policy today." - "cybercriminals are unskilled and get arrested a lot ... Criminal organizations need to base themselves in middle and lower income countries, where they can pay less and hire people who have few formal-economy options. ... I think folks in the AI safety and cyber world sometimes imagine adversaries a… 23 Oct 2026 joshuasaxe181906.substack.com
- A good example of why full automation is never just a matter of AI hitting some capability threshold — how do robotaxi companies deal with riders passed out, having medical emergencies, giving birth (!), vomiting, pooping, or just leaving without closing the car door? Some combination of first responders, remote workers, ground staff, and gig workers, apparently! 22 Oct 2026 jalopnik.com
- Is it "weird" or does he maybe know what he's talking about? Anyway, for those who are open to the idea that there's a well-thought-out AI-as-normal-technology perspective on AI safety and the recent loss-of-control incidents, see my essay with @sayashk, or my interview with Ezra Klein, or the writing of many others with similar views including @joshua_saxe and @natolambert. 20 Oct 2026 normaltech.ai
- This is one of the most important questions about the future of AI. But steering AI development and deployment away from the "automate and compensate" bargain that the public hates is easier said than done, and there are few concrete proposals for how to do it. We just published one: 20 Oct 2026 normaltech.ai
- As LLMs evolved into agents, earlier best practices no longer serve us well. This new research field needs a field guide, so we wrote one. If you're new to agent evals, please take a look. Seasoned researchers and those interested in steering the community will also find many useful observations. 20 Oct 2026 sage.cs.princeton.edu
- My posts on research careers, professional practice beyond academia, and AI's impact on work seem to resonate here, so I've compiled about a hundred entries: https://t.co/JtbIy5WQs2 I hope it is helpful to read it as a collection, because it reflects a coherent underlying philosophy and approach to careers that I've adopted in my own work. There's a lot of career advice online. Two things are distinct about this collection: - Grounded optimism. I'm candid about the structural challenges, twisted incentives, and other limitations of the institutions we inhabit. I share ideas on how we can adap… 19 Oct 2026
- A similar mindset shift — from "what can AI do?" to "what should I let AI do?" — is also needed for individuals using AI for productivity, not just companies. I’ve seen it in my own work. For the last few years, my workflow continually kept changing as AI enabled me to automate more and more types of grunt work. But for the last six months, once I hit my stride with coding agents, my workflow has been surprisingly stable, because capability is no longer the bottleneck to delegation for me. To be clear, capabilities have continued advancing. I could probably have AI do my research and write my… 18 Oct 2026
- I'm often asked what the cruxes are that separate the AI as Normal Technology worldview from that of our critics. I think this is one of them. We think this mindset shift is hard but necessary and achievable. Many others think that as AI capabilities continue to advance, we will inevitably give up control. Creating and compiling indicators of disempowerment and analyzing which direction they're moving remains one of the most important but underappreciated and underfunded areas of AI safety. 16 Oct 2026
- We need a mindset shift in how we deploy AI. In the past, new capabilities unlocked new use cases. Safety, control, ethics, and skill erosion were important but usually secondary considerations; risks to be managed rather than gating criteria. But now and in the future, capability won't be the main bottleneck to deployment — considerations beyond productivity should be the decisive factor. The default question used to be "What can AI do?" But we need to start asking "Of the things AI can do, what should we actually delegate to it?" 16 Oct 2026
Link verified 25 Sept 2026. Recent items update automatically from the channel.
YouTube Arvind on AI Explainers and commentary on AI research.
Recent
Link verified 25 Sept 2026. Recent items update automatically from the channel.
Interviews As a guest Long-form interviews on other people's channels, chosen, not exhaustive.
The conversations
Link verified 8 Nov 2026.
About the English under a post
Some people here publish in a language other than English. Where they do, this site shows a machine translation beneath the post, in this typeface — the site's own, not theirs.
The post itself is never changed, moved or hidden: what is set in the serif above is exactly what the person published, and it is what to quote them on. A translation can be wrong in ways that matter, especially about tone.
Only the post's own words are translated. A quoted post, a linked article and a belief on korrents.com are left in their original language.




