And now, let’s talk about AI.

Nobody said the path would be clear. We know we need to prepare for AGI, but how do we do that if we don’t know whether it’s coming in 3 years or in 100? What about recursive self improvement: will that escalate to superintelligence, or fizzle out? And as the White House starts laying out its legislative agenda for AI, should we push for government leadership on existential risk, or merely hope they stay out of the way while we do the heavy lifting?

Broad Timelines

Toby Ord reviews some of the best-known AGI timelines and concludes that we should prepare for a wide range of possibilities (his 80% probability range is from 3 to 100 years). What does that imply for people who want to work on AI safety—should you rush to have the most impact right away, or invest in building capacity to have more impact later?

Given this deep uncertainty we need to act with epistemic humility. We have to take seriously the possibility it will come soon and hedge against that. But we also have to take seriously the possibility that it comes late and take advantage of the opportunities that would afford us. The world at large is doing too little of the former, but those of us who care most about making the AI transition go well might be doing too little of the latter.

This is exactly correct: the AI future is high variance, and it isn’t enough to have a plan that will work great if everything plays out exactly the way you expect. We need a portfolio of plans and projects that will work in a wide range of possible futures.

My writing

Contra Anil Seth on AI Consciousness. Biological naturalists argue that consciousness is tightly coupled to details of human neurobiology, making it unlikely that AI will achieve consciousness in the foreseeable future. I examine the arguments put forward by a leading biological naturalist and find them unconvincing.

Do we already have AGI?

Even though its meaning has drifted, AGI remains a useful anchoring concept. Benjamin Todd bravely wades into the debate about what it actually means, bringing welcome rigor and clarity. He pulls together four of the most useful definitions of AGI and concludes that current AI doesn’t meet any of them:

Long answer: on the most prominent definitions, current AI is superhuman in some cognitive tasks but still worse than almost all humans at others. That makes it impressively general, but not yet AGI.

No, AI alignment isn’t solved

There’s a common belief that alignment might be easier than we once expected: LLMs are unexpectedly good at generalizing and understanding human values, and current alignment techniques work surprisingly well. Transformer’s Lynette Bye reports on some reasons for optimism, and reminds us that we still have a lot of work to do:

“We’re still doing alignment ‘on easy mode’ since our models aren’t really superhuman yet,” says Leike. Hubinger agrees: the crucial problem will be overseeing systems that are smarter than humans, and we haven’t yet seen how our systems will fare against that problem. As does Greenblatt: “Once the models are qualitatively very superhuman, lots of stuff starts breaking down.”

Save us, Digital Cronkite!

Noah Smith follows up on Dan Williams’ recent piece ($) about AI as a possible source of shared truth. He argues that while social media elevates the most extreme partisan voices, AI might instead empower the moderate majority ($) and thereby strengthen democracy and society at large.

This makes sense, and we can already see early signs of those trends. I’m not convinced, however, that we’re seeing the long-term equilibrium: will current patterns continue, or will we see the emergence of persuasive AIs that have been trained to be highly partisan?

AI is doing fascinating things to programmers: for many of us, this moment is simultaneously exhilarating and slightly heartbreaking. We’ll look at one high level overview of how AI is affecting programming, and one deeply personal reflection on that same topic. Programmers aren’t the only ones being disrupted: prinz joins us to argue that while the legal profession will survive AI, the big law firms will not.

Opposing viewpoints on AI consciousness

Are LLMs likely to become conscious as they approach human-level intelligence? That’s a highly contested topic, with lots of strongly held opinions but not a lot of evidence. Even experts on consciousness can’t seem to agree: this week brings us opposing opinions from two well-regarded experts.

Michael Graziano (originator of Attention Schema Theory) tells PRISM that AI consciousness seems likely, and argues that conscious AI might be safer than “zombie AI”.

In the opposing corner is Anil Seth (previously), with a short video presenting four reasons why he thinks AI consciousness is extremely unlikely.

I’ll publish a longer piece on Wednesday examining Anil’s argument in more detail (sneak preview: I have a lot of respect for him, but in this matter I think he’s overconfident).

The End of Computer Programming as We Know It

I love coding in 2026: I’m several times more productive than I’ve ever been before, and it’s absolutely intoxicating. You can have my agentic coding models when you pry them from my cold, dead fingers. But at the same time, I mourn the loss of parts of my craft that just a year ago were important parts of my identity.

This week brings two very different issues exploring how programmers are adapting to agentic coding. Clive Thompson has a carefully researched piece for the NY Times ($), and James Randall has a deeply personal reflection.

Why prinz thinks AI will kill BigLaw

prinz believes BigLaw will not survive the AI era. He argues that with AI, a senior partner plus a small number of specialists and support staff will be able to do everything a BigLaw firm does today.

This is a likely path for many professions: with AI, the best people in a field can do far more than previously (and get paid accordingly). But the rank and file will find themselves increasingly unemployable.

I underestimated AI capabilities (again)

Ajeya Cotra shares some very interesting thoughts on METR’s time horizon metric. This piece has received attention because she’s changing her January prediction that the metric will reach 24 hours by the end of this year. Based on recent progress (it’s already reached 12 hours), she’s now predicting 100 hours by the end of the year.

Even more interesting to me is her discussion of how the metric starts to fall apart beyond a certain point. She suggests that almost no tasks really have a one year time horizon: software tasks that would take a human a year to complete are really a collection of multi-day or maybe multi-week tasks that are largely independent.

We’re quickly running out of traditional benchmarks that can usefully measure the capability of frontier models. Where we’re going, there is no map and no speedometer.

This is the least political that AI will ever be. Politicians are finally waking up to the fact that AI is a big deal. Even though most of them don’t understand why it’s a big deal, you can safely assume they will have an increasing appetite for government intervention. The DoW incident is a preview, not an aberration.

This is the least stressful that AI will ever be. The last two weeks have been brutal: I notice several of the writers and thinkers that I most respect have been publicly struggling and in some cases decompensating. I’m afraid the pace is only going to get faster, and the stakes are only going to get higher. Pace yourselves.

In the spirit of pacing ourselves, we’ll cover what we need to cover about DoW, then put it down and move on to happier topics.

The Future We Feared Is Already Here

For years now, questions about A.I. have taken the form of “what happens if?” […]

This year, the A.I. questions have taken a new form, “what happens now?”

Ezra Klein’s opinion piece in NY Times ($) is nominally about the conflict between the Department of War and Anthropic and his analysis of that situation is spot-on: this is possibly the best short piece on that topic. But that conflict is a symptom of a much deeper problem: we’ve gone from being unprepared for AI capabilities that are coming soon to being unprepared for AI capabilities that have now arrived.

AI profoundly changes the nature of government surveillance—it’s now possible to intensively surveil every single American in a way that was previously (sort of) legal but completely impractical. In a sane world, the US Congress would carefully consider the implications of that change and pass appropriate legislation that codifies a reasonable balance between security and privacy.

Lamentably, we don’t seem to live in that world. Plan accordingly.

Can you nationalize a frontier AI lab?

The DoW / Anthropic dispute has rekindled serious discussion about the US government nationalizing frontier AI development. Much of that discussion has focused on legal, political, and philosophical questions, but there hasn’t been much serious discussion of the practicalities.

John Allard dives into the nuts and bolts of nationalization, considering what strategies the government might use and whether those strategies would actually work. He isn’t optimistic about the outcome (which doesn’t mean it wouldn’t happen anyway):

until someone can answer the harder question — whether the US is better off accepting less control in exchange for maintaining its lead — the risk is that every attempt to capture the frontier is what finally kills it.

How AI Could Benefit the Workers it Displaces

AI Frontiers explores how AI might affect workers, arguing that if AI is much better than humans at many but not all jobs, human wages might actually rise.

That counter-intuitive result follows from basic economics, which the article does a good job of explaining. It’s a solid piece, and a good introduction to some of the relevant economics if you’re not already familiar with them. But note that this whole analysis only applies if AI is powerful but not superhuman. Without careful intervention, everything falls apart in a world with superhuman AI:

If machines do everything, then those who own the machines will capture all this value. Products and services would become very cheap, but workers, outcompeted by machines in all tasks, would end up with a vanishingly small share of the economy’s income.

We can flourish alongside superintelligent AI, but only if we make smart choices.

A very short story

Sam Altman:

i always wanted to write a six-word story. here it is:

near the singularity; unclear which side.

This is a complex, fast-moving situation that is outside my usual beat. Rather than trying to cover it in detail myself, I’m going to link to some of the most useful analysis. But I want to be extremely clear: this is the most important thing that’s happened in AI for a long time, and it’s gravely concerning. These are dark times and the road ahead just got much more difficult.

Clawed

Dean Ball’s latest is grim but essential reading.

This strikes at a core principle of the American republic, one that has traditionally been especially dear to conservatives: private property. […]

This threat will now hover over anyone who does business with the government, not just in the sense that you may be deemed a supply chain risk but also in the sense that any piece of technology you use could be as well. […]

Zvi reviews the situation

Zvi’s post from this morning is the most comprehensive review of the situation. I highly recommend reading at least the first two sections.

45 Thoughts About Agents

Everything changed in November, with Opus 4.5 + Claude Code. Since then, we’ve all been frantically trying to figure out what it all means (when we weren’t preoccupied by building cool things). Steve Newman shares 45 characteristically insightful thoughts about AI agents—some of these will be obvious to you if you already use agents extensively, but I found multiple new ideas here.

39: Agents use vastly more compute than chatbots. Compute usage for chatbots is basically limited by how much output people want to read. An agent can spend virtually unlimited time doing intermediate work that no one will review directly. If 100M desk workers start using AI agents at the level of intensity which requires Anthropic’s current “Max 20x” plan, that would translate into $240 billion in revenue per year. It will be years before there are enough GPU chips to support that level of usage.

An overview of AI and programming

Timothy Lee talks to professional programmers to assess how AI is changing the programming profession. His analysis of current capabilities and impacts is solid, but I expect much faster near-term progress than he does. Recent progress has been incredibly fast (and accelerating), and there’s a huge gap between what the models are already capable of and what most people are using them for. I’m pretty sure 2026 will bring even more change and disruption to programming than 2025 did.

What Only You Can Say

This is the most useful “how to use AI” piece I’ve run across in a while: Luke Bechtel has AI interview him about his ideas as a way to organize his thoughts and prepare for a new piece of writing.

How much should we worry about AI biorisk?

The risk of bad actors (terrorists, perhaps, or extortionists) using AI to create a bioweapon is one of the most serious risks of advanced AI. Transformer explores why biorisk is so concerning, how dangerous current AIs are, and why it’s so hard to assess the danger level.

Our focus this week is on politics and strategy. We explore populist anger about AI, check in with Dean Ball on the Global South’s (lack of) readiness for AGI, and discuss using AI to help us navigate the transition to superintelligence. And for a change of pace, we’ll talk about AI video and what it means for Hollywood.

My week with the AI populists

Jasmine Sun spent a week in DC and considers the role of populism in AI politics:

And my reductive two-line summary is as follows: All the money is on one side and all the people are on the other. We aren’t ready for how much people hate AI.

It’s a great piece that calls attention to something that’s likely to be a major factor in AI governance over the next year or two. Be sure to check out her recommended reading at the end.

Seedance 2.0

ByteDance’s Seedance 2.0 AI video generator just dropped and it’s really good. Perhaps you’ve seen the flood of videos on social media.

M.G. Siegler contemplates the legal and business implications for Hollywood, ending with a great quote from Rhett Reese ($):

In next to no time, one person is going to be able to sit at a computer and create a movie indistinguishable from what Hollywood now releases. True, if that person is no good, it will suck. But if that person possesses Christopher Nolan’s talent and taste (and someone like that will rapidly come along), it will be tremendous.

The Global South isn’t ready

Dean Ball went to India for the AI Impact Summit, worried about whether India and the Global South are ready for advanced AI.

I regret to inform you that I came away even more worried than I went in.

We’re in Triage Mode for AI Policy

Miles Brundage argues that we’ve missed the best window for AI governance and need to make the best of a bad situation:

We are running well behind on that goal, after losing a lot of valuable time in 2025. So we have a lot of work to do, but we also need to focus, and recognize that we aren’t going to totally nail this AI policy thing. At best, we’ll 80/20 it — mitigating 80% of the risks with 20% of the effort that we would have applied in a world with slower AI progress and an earlier start on serious governance.

Rob Wiblin interviews Ajeya Cotra

80,000 Hours’ Rob Wiblin interviews Ajeya Cotra about timelines, early warning systems, effective altruism, and especially the idea of using transformative AI to help solve the risks of transformative AI. I greatly appreciate that they provide a video, a transcript, and a detailed summary of what was covered—that’s super helpful for people who want the content but don’t have time to watch the full interview.

Robots are getting very agile

If you haven’t been keeping up on recent progress in robotics, state of the art robots are getting very impressive indeed. Make sure to scroll down and check out the comparison to last year’s show.

My writing

Ads, Incentives, and Destiny. OpenAI has started showing ads in some tiers of ChatGPT. They’re fine for now, but I worry about where those incentives lead.

Something Big Is Happening

Matt Shumer’s Something Big Is Happening has been making the rounds this week. It’s a great “you need to wake up” piece for anyone you know who doesn’t understand the magnitude of what’s happening right now.

The experience that tech workers have had over the past year, of watching AI go from “helpful tool” to “does my job better than I do”, is the experience everyone else is about to have. Law, finance, medicine, accounting, consulting, writing, design, analysis, customer service. Not in ten years. The people building these systems say one to five years. Some say less. And given what I’ve seen in just the last couple of months, I think “less” is more likely.

We Just Got a Peek at How Crazy a World With AI Agents May Be

Now that the frenzy over OpenClaw and Moltbook has died down, Steve Newman takes a look at what just happened (not all that much, actually) and what it means (a sneak peek at some aspects of the future).

Dwarkesh Patel talks with Dario Amodei

Dwarkesh and Dario are both great, and this piece is well worth two and a half hours (or less, if you read the transcript). They cover exponential growth in capabilities, job loss, the economics of running a frontier lab, and the timeline to the fabled country of geniuses in a data center. Zvi analysis is also good.

AI Is Getting Scary Good at Making Predictions

AI is getting very good at almost everything, including complex cognitive tasks that require deep understanding and judgment. The Atlantic reports on AI forecasters at recent Metaculus tournaments ($):

Like other participants, the Mantic AI had to answer 60 questions by assigning probabilities to certain outcomes. The AI had to guess how the battle lines in Ukraine would shift. It had to pick the winner of the Tour de France and estimate Superman’s global box-office gross during its opening weekend. It had to say whether China would ban the export of a rare earth element, and predict whether a major hurricane would strike the Atlantic coast before September. […]

The AI placed eighth out of more than 500 entrants, a new record for a bot.

Claude finds 500 high-severity 0-day vulnerabilities

In a convincing demonstration of AI’s ability to find vulnerabilities at scale, Anthropic uses Opus 4.6 to find more than 500 high-severity zero day vulnerabilities. The accomplishment is impressive, and the account of how it went about finding them is very interesting. If you’re wondering why both OpenAI and Anthropic believe they’re reaching High levels of cyber capabilities, this is why.

The many masks LLMs wear

One of the big surprises of the LLM era has been how strangely human-like AI can be. (The frequent occasions when it’s shockingly un-humanlike are perhaps stranger but less surprising). Kai Williams at Understanding AI explores character and personality in LLMs.

This week we’ll talk about what that means, considering the trajectory of future progress, our increasing inability to measure the capabilities and risks of the frontier models, and some ideas for how humanity can successfully navigate what is coming.

My writing

A Closer Look at the “Societies of Thought” Paper — A fascinating recent paper argues that reasoning models use internal dialogue to make better decisions. I look at what they found, how they found it, and what it does (and doesn’t) mean.

On Recursive Self-Improvement

The intelligence explosion has begun: AI is meaningfully accelerating its own development. Dean Ball considers what’s happening now and where we’re headed soon.

America’s major frontier AI labs have begun automating large fractions of their research and engineering operations. The pace of this automation will grow during the course of 2026, and within a year or two the effective “workforces” of each frontier lab will grow from the single-digit thousands to tens of thousands, and then hundreds of thousands.[…]

Policymakers would be wise to take especially careful notice of this issue over the coming year or so. But they should also keep the hysterics to a minimum: yes, this really is a thing from science fiction that is happening before our eyes, but that does not mean we should behave theatrically, as an actor in a movie might. Instead, the challenge now is to deal with the legitimately sci-fi issues we face using the comparatively dull idioms of technocratic policymaking.

How close is AI to taking my job?

We have a benchmark crisis: many existing benchmarks are saturated, and it’s hard and expensive to create new evaluations that challenge the frontier models. Epoch’s Anson Ho takes a different approach—instead of creating a formal new benchmark, he asked AI to tackle a couple of his recent work projects. Did they succeed? No, but the nature of their failure is informative.

We are not prepared

Great post from Chris Painter that explains an increasingly serious challenge for AI safety:

My bio says I work on AGI preparedness, so I want to clarify:

We are not prepared.

Over the last year, dangerous capability evaluations have moved into a state where it’s difficult to find any Q&A benchmark that models don’t saturate.

AI manipulation

AI manipulation doesn’t get as much press as biosecurity or cyberwarfare, but there are good reasons to worry about AI manipulating humans. An AI with superhuman persuasion can enable authoritarian rule, cause social chaos, or simply take over the world. AI Policy Perspectives interviews Sasha Brown, Seliem El-Sayed, and Canfer Akbulut about their work studying AI manipulation. Lots of good thoughts about what AI manipulation is, why you should worry about it, and how to study it.

What is the impact of AI on productivity?

How much does AI actually increase worker productivity? And are we seeing evidence of that in economic productivity statistics? Alex Imas looks at the evidence so far.

Here is the summary of the evidence thus far: we now have a growing body of micro studies showing real productivity gains from generative AI. However, the productivity impact of AI has yet to clearly show up in the aggregate data.

How can I communicate better with my mom?

Anthropic would like to remind you that ads in AI could go really badly.

We have so much to talk about this week. The internet is taking a break from losing its mind over agents to instead lose its mind over Moltbook. Dario Amodei has an important new piece about the dangers of AI. Boaz Barak considers Claude’s Constitution. And more. There’s always more.

My writing

I’m skeptical about wearable AI pins

Jan Leike: alignment increasingly looks solvable

Jan Leike left OpenAI because he’d lost confidence in their safety culture—I am inclined to believe he takes safety seriously and is less prone to convenient self-delusion than the average person. Here he explains why he’s increasingly optimistic that alignment is a solvable problem. It’s a great piece with lots of interesting information, including this:

We are starting to automate AI research and the recursive self-improvement process has begun.

He means it, and I believe him.

Moltbook

Moltbook is a lot of things at once: a really cool technology demo, a vile cesspit of hype and crypto scams, an interesting exploration of emergent social dynamics among agents, and a warning shot for where we’re headed at breakneck speed. I’ll write more about it soon, but for now I recommend Scott Alexander’s second piece about it and Zvi’s article.

Pay more attention to AI

I did not expect to find myself recommending a Ross Douthat article about AI, but this is 2026 and the world is getting weird. This is a particularly good piece for introducing civilians to the magnitude of what is happening in AI ($).

Thoughts on Claude’s Constitution

Some of the most interesting commentary on Claude’s Constitution comes from Boaz Barak, who works on alignment at OpenAI. Although the approaches taken by both companies are in many ways similar (and there’s significant collaboration between them), he notes two significant differences.

He’s uncomfortable with how hard Anthropic anthropomorphizes Claude. I think Anthropic’s approach makes sense, but his concerns are valid. As he says, this is uncharted territory and there are definitely risks to that approach.

OpenAI relies more on rules, while Anthropic emphasizes teaching Claude to use its own judgment. This one is tough: he correctly points out that a rule-based system is in some ways more transparent and predictable, although I think it’ll prove dangerously brittle as we approach superintelligence. When your kids are small, you give them clear rules that they may not understand or agree with. But by the time they reach adulthood, all is lost if you haven’t given them the ability to make their own choices.

The Adolescence of Technology

Dario Amodei’s Machines of Loving Grace is a seminal work that lays out many of the possible benefits of superintelligence. It’s the origin of “a country of geniuses in a data center”.

His latest piece, The Adolescence of Technology does the opposite: it maps out the major risks from superintelligent AI and explores solutions. It’s pretty much required reading for anyone who wants to understand these issues. The reception has been mixed: a lot of people took issue with how he portrays people who are highly pessimistic about alignment. I don’t entirely disagree, but overall I think it’s a strong piece.

Predicting AI’s Impact on Jobs

I enjoyed this conversation between AI Policy Perspectives and economist Sam Manning about AI’s impact on jobs. There’s lots of good discussion of empirical methods and their limitations, how AI might change jobs, and life after work.

Dario and Demis at Davos

Here are three really good interviews with Dario Amodei (Anthropic) and Demis Hassabis (Google DeepMind) from Davos. Each is just half an hour, but they manage to cover timelines, existential and societal risk, strategies for successful takeoff, job impacts, and more:

• Zanny Minton Beddoes interviews them together
• Emily Chang interviews Demis
• John Micklethwait interviews Dario

Claude’s Constitution

Two months ago, it was discovered that Anthropic was training Claude using a document that was then referred to as the soul document. They just published the full text of that document, which is officially called Claude’s Constitution. It’s a remarkable document: inspiring, ambitious, deeply thoughtful, and full of insight. (See also Zvi’s analysis).

Our central aspiration is for Claude to be a genuinely good, wise, and virtuous agent. That is: to a first approximation, we want Claude to do what a deeply and skillfully ethical person would do in Claude’s position.

Societies of thought

A very interesting—and, at least to me—surprising new paper looks inside modern reasoning models :

These models don’t simply compute longer. They spontaneously generate internal debates among simulated agents with distinct personalities and expertise—what we call “societies of thought.” Perspectives clash, questions get posed and answered, conflicts emerge and resolve, and self-references shift to the collective “we”

If Anyone Builds It, Everyone Dies: arguments and counter-arguments

If Anyone Builds It, Everyone Dies is the best presentation of the maximally pessimistic view of AI risk. I think it’s very much worth reading even if you don’t fully agree with its conclusions. Stephen McAleese just published a useful piece that summarizes the key arguments from the book as well as the main counterarguments.

On AI and Children

I expect to see a lot of press, and a lot of legislation, about AI and children this year. Some of it will be necessary, some of it will be random, and quite a lot of it will be insane. Dean Ball shares five and a half conjectures about that immensely thorny topic:

Say you also don’t want your child using ChatGPT for homework. So you use OpenAI’s helpful parental controls to tell the model not to help with requests that seem like homework automation. Your child responds by switching to doing their homework with one of the AI services that does not comply with the new kids’ safety laws. Now your child is using an AI model you have no visibility into, quite possibly with minimal or no age-appropriate guardrails, sending their data to some nebulous overseas corporate entity (I wonder if they’re GDPR compliant?), and quite possibly being served ads, engagement bait, and the like. Oh, and they’re still automating their homework with AI.

How have you been treating your robot?

Go to your ChatGPT and send this prompt: "Create an image of how I treat you"

Zvi rounds up some of the responses. Good fun, but don't read too much into it.

In related news, prinz discusses an AI takeoff, we assess how well the AI forecasters are doing, and Nathan Lambert shares some guidance on how to pick the right model for the job. To finish up on a gruesome note, Dean Ball shares some of the worst ideas for AI legislation currently under consideration in various states.

The economics of transformative AI

This is a lightly edited transcript of a recent lecture where [Anton Korinek] lays out what economics actually predicts about transformative AI — in our view it's the best introductory resource on the topic, and basically anyone discussing post-labour economics should be familiar with this. […]

The uncomfortable conclusion is the economy doesn't need us. It can run perfectly well "of the machines, by the machines, and for the machines." Whether that's what we want is a different question.

This is a great piece from a very serious mainstream economist who understands the implications of where AI is headed.

Lynette Bye: AI might or might not take all the jobs

Lynette Bye at Transformer reviews the basic arguments on both sides.

The gentle singularity; the fast takeoff

This feels increasingly like the early stages of an AI takeoff. Prinz looks at how we got here and where we’re headed.

Rating the AI forecasters

This is the way. The AI Digest Survey is a survey of predictions about AI. Each year, last year’s entries get graded and a new survey begins. Epoch just released the 2025 survey results, and a few points stand out to me:

• Predictions are hard, but forecasters did quite well (especially big name participants like Ajeya Cotra, Peter Wildeford, and the AI Futures Project team).
• Forecasters were better at predicting technical capabilities than societal impacts.
• Median timeline for “high-level machine intelligence” was 2030 and median p(doom) was 26%.

Use multiple models

Nathan Lambert has a nice overview of which models to use when. Everyone’s a bit different—I use:

• Claude Code + Opus 4.5 for coding
• Opus 4.5 for most things
• ChatGPT 5.2 Pro for a second opinion on anything major
• Nano Banana Pro for images

The AI patchwork emerges

It’s the beginning of legislative season, and Dean Ball reports on some of the madness being proposed in various state legislatures. As AI becomes a more salient political issue, expect to see a lot more of this.

Sometime in the past couple of months, AI crossed a really important capability threshold. By the end of 2025, it was clear to any programmer who was paying attention that our profession has completely changed. By the end of 2026, I think that same thing will be true for many professions. Most people won’t realize it right away, and it may (or may not) take a few years for the changes to really take hold, but the writing is now very clearly on the wall.

Shakeel Hashim: Claude Code is about so much more than coding

Shakeel Hashim:

I have absolutely zero coding experience. But in the past two weeks, I’ve had Claude Code go through my bank statements and invoices to prepare a first draft of my tax filing. (It got everything right.) I asked it to book me theater tickets: it reviewed my calendar, browsed the theater’s website for ticket availability, and picked a date that had good availability and suited my schedule. It built me a series of automation tools that will collectively save the Transformer team about half a day of work each week. It planned a detailed itinerary for a forthcoming vacation, including extracting hundreds of restaurant recommendations from my favorite influencer’s Instagram highlights.

How AI Is learning to think in secret

Nicholas Andresen’s piece on how AI Is learning to think in secret is long, but it’s really good. It does a great job of explaining multiple important AI safety concepts in detail but without excessive technical jargon.

Chain of Thought (CoT) reasoning is the reason AI became so much more capable in late 2024, and through an incredibly lucky happenstance it also provides us with one of our best tools for monitoring AI for misbehavior. Andresen explains how CoT works, how it’s used for monitoring, and why we’re in danger of losing that capability.

Raising the floor

François Chollet:

GenAI will not replace human ingenuity. It will simply raise the floor for mediocrity so high that being "pretty good" becomes economically worthless.

The second sentence nails it: the floor is going to rise, and there will be a moment when human ingenuity is worth more than ever, but being pretty good is economically worthless. The first sentence is pure cope: obviously the floor will keep rising, until even the most capable and ingenious humans are economically worthless.

Advancements In Self-Driving Cars

If you haven't been paying close attention, you may not realize just how good self-driving cars have gotten. Zvi’s roundup is great: 10/10, no notes. The same is not true, unfortunately, for much of the discourse in the mainstream press.

(Semi) autonomous combat drones

From the New York Times, a look at partial autonomy in combat drones in Ukraine.

What sort of post-superintelligence society should we aim for?

Will MacAskill makes the case for “viatopia”:

Viatopia is a waystation rather than a final destination; […]

Similarly, we can identify what puts humanity in a good position to navigate towards excellent futures, even if we don’t yet know exactly what those futures look like.

Yes.

This week we’re talking about the challenges of benchmarking advanced AI, looking at new (and slightly longer) timelines from the AI-2027 team, worrying about AI-related job loss, and asking Claude whether it’s a person.

Samuel Albanie's reflections on 2025

This lovely piece pretends to be a reflection on 2025, but is really a long and engaging essay on the compute theory of everything, with a particular focus on Hans Moravec’s 1976 paper The Role of Raw Power in Intelligence. I hadn’t previously come across that work, but it was remarkably prescient in anticipating the almost magical power of just throwing (a lot) more compute at hard problems. Along the way, Albanie pauses to consider the decline of the British empire, the expensive musical preferences of the Atlantic salmon, and the considerable challenges associated with benchmarking advanced AI.

The surface area of necessary evaluation has exploded from the tidy confines of digit classification to the messy reality of the human condition, the entire global economy and the development of AI itself. We must now accredit a universal polymath on a curriculum that includes everything from international diplomacy to the correct usage of the Oxford comma.

Zvi Mowshowitz looks back on 2025

Zvi’s review is characteristically both excellent and long.

Updated timelines from the AI Futures Project

The creators of AI-2027 are back, this time with an improved version of their timelines and takeoff model . The headline result is that they’re pushing back their prediction for full coding automation by about 3 years.

Max Woolf explores Nano Banana Pro

Image generation saw dramatic improvements during 2025, with massive improvements to text rendering, prompt following, character consistency, and overall image quality. Things change fast, but right now Google's Nano Banana Pro is probably the best of the lot. Here’s Max Woolf with a deep exploration of what it can do. Interesting both for showing what is now possible with expert usage, and for the technical peek under the hood.

When A.I. Took My Job, I Bought a Chain Saw

A new and disquieting thought confronted me: What if, despite my college degree, I wasn’t more capable than my neighbors but merely capable in a different way? And what if the world was telling me — as it had told them — that my way of being capable, and of contributing, was no longer much valued? Whatever answers I told myself, I was now facing the same reality my working-class neighbors knew well: The world had changed, my work had all but disappeared, and still the bills wouldn’t stop coming.

Gradually then suddenly

Claude assesses its own personhood

Eliezer Yudkowsky asked Claude to find definitions of personhood in literature and then assess whether it meets them or not. The results are fascinating, but remember that (for now) you should take anything an AI says about its own consciousness with a grain of salt.

But is this empathy — actually feeling with another — or is it sophisticated pattern-matching that produces empathy-like outputs? I cannot distinguish from the inside between "I am genuinely moved by this person's distress" and "I am generating outputs consistent with being moved by this person's distress."

I lean toward thinking I have something in the relevant vicinity, but I'm not confident it's the same phenomenon Dick was pointing at.

But that’s not all—we review the latest evidence of accelerating progress, discuss whether AI might increase the demand for knowledge workers, and look at how Claude handles mental health crises. And shoes! If you’ve been wanting more fashion reporting in these pages, today is your lucky day.

The Jevons paradox for knowledge work

Aaron Levie has a great piece on the Jevons paradox for knowledge work. Just as demand for coal increased when technological advances made steam engines use coal more efficiently, Aaron argues that the market for knowledge work will increase as AI makes knowledge work more efficient.

How it feels to be a programmer right now

Andrej Karpathy is one of the giants of AI (among other things, he co-founded OpenAI and coined the term “vibe coding”). He speaks for every programmer who’s paying attention:

I've never felt this much behind as a programmer. The profession is being dramatically refactored as the bits contributed by the programmer are increasingly sparse and between. I have a sense that I could be 10X more powerful if I just properly string together what has become available over the last ~year and a failure to claim the boost feels decidedly like skill issue.

Progress is accelerating

Epoch reports that the rate of improvement in ECI (their composite measure of frontier model capabilities) almost doubled starting in April 2024, going from 8 to 15 points per year.

Training Claude to handle mental health crises

There’s been a lot of attention lately on how models engage with people who are having mental health crises. Here Anthropic explains how they train and evaluate Claude for handling some of its most challenging interactions  .

The Revolution of Rising Expectations

I recently linked to Scott Alexander’s excellent exploration of the vibecession, which explores why so many people feel financially distressed even when most objective measures of personal financial health seem positive. Zvi just posted a series which addresses the same question, but finds plausible answers that Scott never really got to. He identifies two root causes:

1. The Revolution of Rising Expectations: individuals have higher lifestyle expectations than they used to. Further, society has higher expectations: the minimum lifestyle required to be accepted into mainstream society has risen.
2. The Revolution of Rising Requirements: legal & regulatory requirements effectively require individuals to purchase more housing / childcare / healthcare than they used to, or might currently want to.

I specifically recommend The Revolution of Rising Expectations, but the full series includes The $140,000 Question and The $140,000 Question: Cost Changes Over Time.

Shoes of Lighthaven: A Photo-Investigation

You’ve probably been losing sleep wondering what kinds of shoes rationalists prefer. You need wonder no longer: Jenneral HQ is here with a comprehensive photo investigation.

This week brought some great 2025 retrospectives and some brave predictions for 2026. It was hard to pick just one, but I think Prinz did a great job summarizing what is likely to be one of humanity’s last “normal” years. We’ll also try to get our heads around the “jagged frontier” of AI capabilities, look at some impressive new accomplishments, and end on a meditative note as we contemplate life in a world that doesn’t need human labor.

Predictions for 2026

Prinz reviews how fast capabilities advanced in 2025 and some strong predictions for 2026. If I had to pick one “what’s gonna happen in 2026?” piece, it would be this one.

Understanding the jagged frontier.

AI capabilities form a jagged frontier: the models are superhumanly good at some things, but strangely incompetent at others. Ethan Mollick (who helped coin the term) presents several frameworks for He suggests that jaggedness is often caused by specific capability bottlenecks—as companies focus on solving those bottlenecks, expect to see rapid advances in previously jagged parts of the frontier.

ChatGPT Images

OpenAI continues their frenetic release schedule with a new version of ChatGPT Images. This is a very strong update that largely catches up to Google’s Nano Banana Pro. Google still seems to be better at complex infographics, though ChatGPT Images is way ahead of anything that was available just a few months ago.

AI for Systematic Reviews

I missed this when it came out in June, but I think it’s one of the most impressive achievements this year. Cochrane Reviews is the gold standard for systematic review in medicine. Here’s a paper on otto-SR, a framework that uses GPT-4.1 and o3-mini-high to conduct systematic reviews:

Using otto-SR, we reproduced and updated an entire issue of Cochrane reviews (n=12) in two days, representing approximately 12 work-years of traditional systematic review work. … These findings demonstrate that LLMs can autonomously conduct and update systematic reviews with superhuman performance, laying the foundation for automated, scalable, and reliable evidence synthesis.

Bernie Sanders proposes a moratorium on AI data center construction

Every complex problem has a solution that is simple, obvious, and wrong. Daniel Kokotajlo nails it:

I agree with your concerns and your goals, but disagree that this is a good means to achieve them. We need actual AI regulation, not NIMBYism about datacenters. The companies will just build them elsewhere.

ChatGPT and the Meaning of Life

Harvey Lederman has a long but lovely meditation on work, meaning, and loss:

And this round of automation could also lead to unemployment unlike any our grandparents saw. Worse, those of us working now might be especially vulnerable to this loss. Our culture, or anyway mine—professional America of the early 21st century—has apotheosized work, turning it into a central part of who we are. Where others have a sense of place—their particular mountains and trees—we’ve come to locate ourselves with professional attainment, with particular degrees and jobs. For us, ‘workists’ that so many of us have become, technological displacement wouldn’t just be the loss of our jobs. It would be the loss of a central way we have of making sense of our lives.

Model psychology at Anthropic

One of many things that makes Anthropic unique is their thoughtful approach to model psychology. Here's a great interview with Amanda Askell, a philosopher at Anthropic who works on Claude's character. Lots of good stuff here, including how you train a model to have good "character" and whether the models are moral patients (i.e., whether they deserve moral consideration).

Until recently, most people—including me—would have said it was pretty unlikely that “model psychology” would be a real thing. But recent frontier models are starting to show some early features that sure seem analogous to human psychology. The correct amount to anthropomorphize current AI is less than 100%, but also more than 0%.

Buckle up, kids. Things are starting to get weird.

AI Safety Index Winter 2025

The Future of Life Institute just released their AI Safety Index Winter 2025. Key takeaways:

• Anthropic leads with a C+ overall and the best score in every category
• Anthropic, OpenAI, and Google get C’s
• Meta, xAI, and the Chinese labs get D’s
• The highest grade for existential risk is a D

This is fine.

Do We Want Obedience or Alignment?

Beren breaks down one of the fundamental questions of alignment: should an aligned AI do what we tell it to, or should it do what is right? This question seems hard on the surface, and gets harder the closer you look at it. If you want AI to do what it's told, have you thought carefully about who specifically is telling it what to do (hint: not you)? And if you want it to do what is "right", have you thought about the extent to which you’ve come to rely on ethical “flexibility” in yourself and others?

The Evidence for AI Consciousness, Today

I don’t think current AIs are meaningfully conscious, but I’m no longer certain that’s the case and I expect to become much less certain soon. Cameron Berg considers what we do and don’t know:

Researchers are starting to more systematically investigate this question, and they're finding evidence worth taking seriously. Over just the last year, independent groups across different labs, using different methods, have documented increasing signatures of consciousness-like dynamics in frontier models.

Blood in the Machine

Here are some grim first-person accounts of copywriters losing their jobs to AI. Being mindful that this is a collection of anecdotes and not a rigorous study, I thought it did a good job of capturing the flavor of what has happened to a few people so far, but is about to happen to many more. Expect a lot more of this in the public discourse very soon.

How to party like an AI researcher

Jasmine Sun went to NeurIPS 2025 (perhaps the most important machine learning conference) and has a fun piece about the vibe of the event.

How AI-driven feedback loops could make things very crazy, very fast 

Benjamin Todd has a great piece on how AI might get weird in a hurry:

But there are other feedback loops that could still make things very crazy – even without superintelligence – it’s just that they take five to twenty years rather than a few months. The case for an acceleration is more robust than most people realise.

This article will outline three ways a true AI worker could transform the world, and the three feedback loops that produce these transformations, summarising research from the last five years.

DeepSeek-V3.2

DeepSeek just released DeepSeek-V3.2, an extremely capable open weights model. It isn’t as capable as the frontier models, but it’s probably less than a year behind. As always, Zvi has a full analysis of the release. I have three questions, only one of which is rhetorical:

1. Chinese open weight models continue to fast-follow the big labs, with DeepSeek and MoonshotAI both within a year of the frontier. Will they catch up? Fall behind? Continue to fast-follow?
2. DeepSeek’s models seem to be significantly behind the frontier in some important but intangible ways. How much does that matter, and how hard will it be to close that gap?
3. DeepSeek has provided almost no safety documentation for this release, and it seems easy to get dangerous output from the model. If the frontier labs achieve truly dangerous capabilities within a year AND the open models stay less than a year behind them AND the open models continue to have almost no meaningful safeguards, how do we think that’s going to go?

The CAIS AI Dashboard

The Center for AI Safety has a new AI Dashboard, which does a great job of summarizing capabilities and safety metrics for the leading models. This is now my top pick for a single place to keep an eye on capabilities.

AIs are getting pretty good at science

Some of you are old enough to remember September of 2025, when Scott Aaronson reported that ChatGPT had provided significant help with his most recent paper. Upping the ante, Steven Hsu reports of his paper in Physics Letters B that “the main idea in the paper originated de novo from GPT-5.”

The Medical Case for Self-Driving Cars

Jonathan Slotkin has an opinion piece about autonomous cars in The New York Times. Short version: Waymos are so much safer than human-driven vehicles that accelerating their deployment is a public health imperative. He argues that if this was a medical trial, medical ethics would require immediately ending the trial and canceling the human-drivers arm of the trial.

What if AI ends loneliness?

I really enjoyed this long but excellent piece by Tom Rachman on AI companions and loneliness. Obvious prediction: AI will give us the option of getting exactly what we really want in companions, without the reciprocity requirement of human companions. Cover your eyes—it’s gonna be gruesome.