2026-08-30 10:51:35
Good morning!
What are the conditions under which AI systems could undergo recursive self-improvement (RSI) — and how long might that last? Cards on the table, I’m not wildly excited by the theory of unending accelerating recursive self-improvement for the simple theoretical issue of control and alignment. But I also think it’s not likely for practical and theoretical reasons.
Now philosopher, Toby Ord, has done the heavy lifting for me. He argues that a key gating factor to RSI is the generation time: how long it takes of an AI system to go through a single loop of improvement (where it helps design and train its successor).
Ord concludes that while it is mathematically possible for extreme RSI, intelligence rising without bound, the conditions are unbearably difficult to achieve. The key question is whether the entire research to training to development cycle can shrink to zero or not. Ord reckons unlikely, I too don’t believe it is possible.
Generation time can’t get to zero because real-life intrudes: experiments take time; training runs take time; making new chips take time… lots of things take time. It might still feel fast, but it wouldn’t race to infinity.
Eventually, physics intrudes too: the speed of light limits communication speed; the Bekenstein bound limits the information contained within a finite bit of space; and Landauer imposes an energy tax on irreversible computation.1
The Universe, it seems, agrees with me. Unbounded RSI has its limits.
You can read Ord’s paper here. Premium members can explore a plain English interactive version too.
Open-weight models are growing in popularity in the business world: their token share at Vercel hit a single-day record of 62%, up from 28% two months earlier. Some Western firms are even moving workloads to Chinese open weights — Thomson Reuters has developed its first in-house model based on Qwen to cut costs. You can tune cost-effective open weights to match, or sometimes beat, frontier performance on the tasks that matter to you. Take Bridgewater: working with Thinking Machines, it fine-tuned an open Qwen model on expert-labeled data and beat every frontier model it tested on its internal information-filtering tasks: roughly 30% fewer errors than the best closed model, at one-fourteenth of the inference cost. Trainloop, which I am an investor in, does something similar, using the tiny Qwen 3.7-27b model, and can outperform GPT 5.6 Sol on specific fine-tuned tasks at a fraction of the cost.
The results that Trainloop is getting are pretty impressive, seeing as they are based on a pocket model—a 27b model will even fit on a desktop Mac.2
Openweight models are, of course, getting better and better. Z.ai GLM 5.3, released this week, completely reshapes the cost-performance Pareto frontier. Of course, it isn’t a small model, but smaller distillations will emerge from it.
All of this speaks to a welcome competition in AI provision. Clearly, firms could move focused workloads onto the most-performant, fine-tuned small models they can. Where possible, they might choose large, generally capable open models. But the appeal of being at the frontier, which is more than just model performance—it is service guarantees, harness quality, reliability, and a host of other requirements—still drives significant business for Anthropic and OpenAI.
We don’t think it has much impact on the question of whether revenues flowing into the industry will materially change. For one thing, we don’t have a counterfactual to test against. But more importantly, every open model still involves paying inference providers. We’ll be looking at this question in more detail in the comings weeks.
Recursive self-improvement may be entering the compute realm. OpenAI’s new chip, ‘Jalapeño’, was designed with a heavy helping hand from the company’s own models, which helped write kernels and cut roughly 10% from one of the chip’s main compute blocks. In around 16 months from first hire to tape-out, OpenAI has built a chip that beats comparable Nvidia silicon by 1.5–1.9x on tokens per megawatt at peak throughput. This suggests frontier models can compress the design cycle for competitive silicon.
AI will result in far greater heterogeneity in chip architectures than we saw in prior computing markets. Personal computers battled between the x86 standard and the Motorola 68x, before today’s duopoly of Intel and Apple silicon. Different uses for AI will need compute optimised for intelligence, latency, power consumption, training, and inference. This creates lots of room for specialist firms. One example is that ChatGPT’s fast response mode is powered by Cerebras’ low-latency silicon. Another is Fractile, where I am an investor, which has a deal with Anthropic for its low-latency inferencing chips.
Compute is becoming a highly segmented market, where chips aren’t a standardized commodity. This differentiated hardware demand will expand the market even if it potentially reduces Nvidia’s relative dominance.
Good post from Chad Syverson on AI productivity. Some micro-evidence for improvements, not much at the aggregate level.
The least bad place to hide from global catastrophe? Australia.
The harness matters as much as the model: SwarmOS pushed GPT-5.6 Sol from 13.3% to 100% on ARC-AGI-3 Public.
The University of Chicago’s Social Sciences Core is going back to paper, banning most classroom technology to deal with the AI-learning crisis.
Meta considered shrinking some teams by up to 60% to become “AI native.”
Would you take this bet? It pays out if, in any quarter up to and including Q4 2033, US real GDP per capita is at least 15% higher than its previous peak.3
With one video, you can reconstruct a moving 4D avatar of a person and render it from novel viewpoints, like a video-game character.
You can now teach an adorable “Pixar-ish” robot new tricks for only $399.
You can now generate videos in less time than it takes to watch them.
A walk down memery lane:
I run a version of Qwen 3.7-27b on one of our local machines for various tasks.
Recorded at least four quarters earlier.
2026-08-24 18:46:04
Hi all,
Here’s our Monday roundup of data signals across AI, energy & markets.
Enjoy!
Every week, we will share the latest updates on the state of the AI economy based on our own latest research and tracking.
In our latest inference token update, the share of open-weight tokens has doubled in the last twelve months. While we are approaching a 1:1 closed-to-open token ratio, the number of closed-weight tokens grew sevenfold over the same period.
See our State of the AI Economy 2026 report for more.
📧 For advisory requests and institutional inquiries, please contact [email protected]
🤝 Want to work with us? We are hiring an AI Economy Research Fellow
The canary keeps coughing. Employment of young workers (ages 22–25) in AI‑exposed occupations is now 19% below trend. Up from 15% last year.1
Agent token dominance. AI agents used more tokens than humans in February this year; agents are now using 14x that amount, while human token usage grew only 2.8x.2
2026-08-23 10:19:40
Good morning!
We are looking for an outstanding economist to join us as an AI Economy Research Fellow. If you know someone we should speak to, send them our way.
A few months ago, we (alongside ) looked at whether AI is immune to groupthink. The answer was no. Blending several models’ answers kept about a quarter of the good ideas that had come from a single model. This is called the hidden-profile problem: when groups discuss what everyone already knows and don’t get to the knowledge that only one member holds. Anthropic has now run that classic experiment on agents: four agents must arrive at a decision. The evidence they hold in common points to the wrong option, while only a few agents (or just one) have the facts that lead to a correct decision. Getting it right means a small set of agents pressing its private facts and the others trusting them over the apparent consensus. After discussion, most model families chose correctly in only 17-36% of runs, while a single agent handed the entire evidence base got it right nearly every time. Only one model (somewhat) escaped: Mythos 5, at about 85% (why, we don’t know).
I see two problems at work here. First, LLMs lack diversity (they are low-variance): set 30 agents the same coding task and 18 of them will name their git branch identically. Second, agents lack the institutions that make human groups robust: reputation, recourse and protection for the lone dissenter. These aren’t necessarily unfixable, but it’s not yet clear what the fix is. On the diversity side, I particularly like the solutions Thinking Machines puts forward: an ecosystem of AIs raised in different places, with different values and purposes, “keeping the weirdness alive.” After all, most good ideas started weird.
In our State of AI report, we found a positive but underwhelming elasticity for tokens. A 10% price cut lifts token use by 12–18%: enough to raise total spend, but not by much.
Patrick Saner made a comment that made me rethink why: “the cost per token is irrelevant. What matters is the cost of completing a useful unit of work.” Elasticity might be underwhelming because users haven’t found a way to properly price “a useful unit of work.” Firms exist exactly to avoid pricing work. Especially for knowledge work, we buy a lot of it in bundles: a salary, a retainer, an hour. Creating a priceable task from knowledge work is not easy. Some may have found a useful unit: since October 2023 the top 1% of firms raised AI spend per employee by $6,542. The median rose only $9.63. I would guess this is mostly software, where AI is both most proven and, in a sense, most measurable (commits, pull requests and releases).
2026-08-22 21:23:14
The petard was a sixteenth-century explosive charge. An attacking engineer would carry it to a castle gate, attach it, light the fuse and scramble for cover.
It was a tricky business. The charges were temperamental, their fuses particularly so, and the installer might blow himself up. In Hamlet, the phrase earns its immortal meaning:
For ’tis the sport to have the engineer Hoist with his own petard; and ’t shall go hard But I will delve one yard below their mines And blow them at the moon. O, ’tis most sweet When in one line two crafts directly meet.
Today’s petardiers are not Rosencrantz and Guildenstern conspiring against the Prince of Denmark.
They are the titans of AI, the bosses of the labs, the investors behind them. For nearly a decade, these software coders had made promises: of reigniting economic growth, of making daily life easier and less risky, perhaps even of eliminating disease. To do this, they would need capital: to write their software and to build 21st-century infrastructure to run it. The gains will be so huge that they’d need to go quickly, very quickly.
But they warned that this was no ordinary software. It was tricky and hard to understand, so much so that only a few should steward it. After all, this was a technology that would possibly take your job, maybe kill you, perhaps get out of control and kill all of us. Still, they needed to build it, all the while warning that they would eventually need to take control of it before things got out of hand. In the meantime, let them be.
This was the petard: AI is too important for America not to let us get on with it. They placed this claim on the gates of society, itching to get on with it. And it just exploded in their face.
This is a fiendishly complicated issue, and I’m not going to pretend to understand the whys and wherefores of American political decisions. My rough take, though, is that this is a Gordian knot. It cannot be disconnected from the hapless messaging from AI firms over the past decade, which now comes to life at the county level. Nor can we separate the idea that these concrete blocks, which might abstractly benefit the economy or healthcare or whatever, tangibly serve an out-group they don't much like.
The rest of the essay has further analysis on attitudes towards datacenters; whether they actually benefit local communities; and the commentary I have been reading to understand this.
2026-08-19 20:05:40
Is AI a bubble? Not yet. Our updated dashboard tracking the investment wave currently has no gauges in the red, two in amber, and the rest in healthy green (just).
Since our last update, AI revenues have continued to rise, reaching $126 billion over the last twelve months as of July. We also experienced a jumpy market, which led to a severe correction in semiconductor stocks, somewhat cooling public valuations. On our side, we have improved the methodology for counting AI capex (we show both the published and restated series below).
That demand has smacked headlong into a tight supply of compute capacity, which is being met by increasing investment in infrastructure. And with that comes more risk. While the hyperscalers are still using a large share of their cash reserves, they are increasingly scouring the globe for capital, both straight-up debt and increasingly intricate financing vehicles. As argues, this “gaming of the system” is not only rational; it is necessary, as long as revenue is compounding. But these structures can become brittle if it slows.
Funding quality has deteriorated since Sep 2025. In our base case, we expect it and economic strain to turn red during 2027.
The full analysis shows where the tension is building.
For members, we:
Update all five boom-or-bubble gauges with new data
Show why AI revenues are outrunning even higher infrastructure spending.
Discuss the web of debt, leases and guarantees now underpinning the buildout.
Explain our outlook through 2027 and the signals that would change our minds.
The verdict remains boom, not bubble.
2026-08-18 16:53:30
Exponential View is appointing its first Research Fellow.
We are looking for an economist who can connect frontier economic thinking to messy, real-world evidence and reach useful judgments with the foresight Exponential View is renowned for.
The Fellow will investigate how AI is changing economic value, work, firms and markets. The questions may include but are not limited to:
How should AI companies, infrastructure and capabilities be valued from first principles?
Where in the AI value chain is the economic surplus created, and who is capturing it?
What is AI’s impact on wages, employment, productivity and worker bargaining power long-term?
What are the microeconomic effects inside firms?
What counts as transformative AI and which leading indicators would reveal the state of the transition?
We don’t expect the Fellow to arrive with settled answers. But we do expect the Fellow to know how to turn our questions into testable economic mechanisms, assumptions and back-of-the-envelope estimates that build towards further empirical work.
The position is based in London, UK.
Exponential View is an independent research organisation founded by Azeem Azhar. We study how AI and other general-purpose technologies change economic value, institutions and the distribution of power.
Our AI Economy programme tracks the physical, financial and organisational build-out of AI. We research semiconductors, energy and data centres, model capabilities and economics; corporate investment and adoption; firm-level performance; labour markets; and the distribution of value across the stack. We combine original datasets, company and sector analysis, economic reasoning, and direct engagement with stakeholders to produce evidence-based analyses that stand the test of time.
Our work is written for people making consequential decisions in business, investment, technology and public policy. It is rigorous, empirical and explicit about uncertainty. We don’t wait for consensus, and we rarely care for it.
Explore our AI Economy work here: intelligence.exponentialview.co.
This is a paid six- or twelve-month applied economics Fellowship with substantive responsibility across Exponential View’s research and publications.
Our proposition is as ambitious as it is simple. Inter alia, the Fellow will translate economic theory and frontier research into useful insights that might inform decision-making. They will maintain the standards of serious academic work while learning to operate at the speed of a live technological and economic transition. They will develop scenarios, measures, and leading indicators for transformative AI, connecting the work of leading economists to our own methods and models.
The Fellow will work closely with Azeem Azhar and EV’s research team, contribute to the core AI Economy programme, and develop one substantial research output of their own. Where the work warrants it, contributions will be publicly credited or bylined.
Academic projects can run for months or years. At Exponential View, we often need a good answer within hours or days. That requires tightly framed questions, intelligent use of imperfect data, visible assumptions, distinguishing between evidence and inference, and the willingness to revise or abandon a view quickly.
The Fellow will be augmented by as much AI as they need and is expected to actively build their AI research tools and skill set to become an even better researcher.
At minimum, you will have an outstanding master’s-level qualification in Economics, including an MPhil. You may be a current doctoral student in economics seeking a period of intensive applied work.
Relevant specialisms may include applied microeconomics, labour economics, industrial organisation, productivity and growth, innovation economics, financial economics or the economics of technological change.
The Fellow must demonstrate:
Serious prior engagement with AI, automation, technological change or a closely related economic question. A generic interest in AI is not enough.
Strong applied economics and empirical judgement. We are less interested in theory for its own sake than in the ability to use theory to structure an answerable question.
Experience working directly with difficult real-world data, including knowing when it is incomplete, endogenous, inconsistently defined or simply wrong.
The ability to translate frontier models and academic arguments into observables, datasets, tests and an intelligible view of the world.
Comfort with ambiguity, compressed deadlines and changing priorities.
The ability to write clearly for an economically literate audience without hiding behind academic language.
You work well independently, can make rapid progress on your own, and identify the few decisions that genuinely require senior input.
You are intellectually honest and can distinguish what is observed, estimated, inferred, and unknown; you change your mind when the evidence changes.
This role will fit an economist who wants to become faster, more empirical, and more effective without surrendering rigour.
Right to work in the UK is required.
Who this is not for:
This is not a role for a general quantitative researcher from an unrelated discipline or for someone who wants to learn economics on the job.
It is also unlikely to suit someone seeking a conventional academic postdoctoral rhythm or whose overriding objective is the next journal publication.
The Fellow will leave with:
A body of applied, publicly visible work on the AI economy.
Experience moving from an open-ended question to a defensible empirical judgement on a compressed timescale.
Feedback on research design, analytical judgement, visual explanation and writing.
One substantial, independently owned research output.
Exposure to EV’s network of economists, technologists, business leaders, investors and policymakers, including relevant academic collaborations.
Duration: Six or twelve months
Commitment: Full-time preferred; a substantial part-time arrangement may be possible for an exceptional candidate
Location: London, UK / hybrid, with regular in-person work
Application deadline: 6 September 2026
This is a paid Fellowship.
Please submit:
Your CV.
A note of no more than 200 words explaining why you want this Fellowship and what you want to improve during the appointment.
One or two substantial examples of empirical economic work.
For collaborative work, a precise description of your own contribution.
Please answer the following questions in no more than 500 words each:
Choose one important claim about AI’s effect on fundamental value, wages, productivity, or market structure. What is the economic mechanism? How would you test it with real-world data, and what result would cause you to revise or reject the claim?
Take one proposition from the economics of transformative AI – for example, from the work of Anton Korinek, Daron Acemoglu or another serious economist – and translate it into observable indicators over the next two years. Which existing data would you use, what is missing, and what new measure could you create?
Please confirm you have the right to work in the UK. Evidence will be required at a later stage.
Send your application to [email protected] by 23:59 BST on 6 September 2026.
We will invite shortlisted candidates to a first interview. Owing to the volume of applications, we may be unable to respond to every applicant individually.
Following the interviews, selected candidates will complete a paid, tightly time-boxed research exercise using public or synthetic data. It will test the ability to frame an economic question, work quickly with imperfect evidence, produce a defensible analysis and communicate it clearly. It will not be used as unpaid production work.
Reasonable adjustments will be available throughout the application process.
