The Economics of Recursive Self-Improvement [pdf]

https://elasticity.institute/rsi-paper.pdf

by apsec112 • 9 days ago

149 97 9 days ago

zuzuen_1• 9 days ago

"We make a tentative calibration of the self-sustaining ac celeration condition using the existing data that is available, measuring AI capabilities using the Epoch Capabilities Index (Ho et al., 2025).

We find that the condition is met if a one-unit increase in AI model capabilities results in at least 15% higher AI R&D pro ductivity.

A rough back-of-the-envelope calculation based on reported AI engineer uplift suggests this return has been around 9% since the launch of coding agents.

This number is below the model-implied threshold, suggesting we are not experiencing a self-sustaining acceleration."

And the source of this data seems to be self-reported productivity gains from surveys: 1.4–2X in METR’s survey of technical workers (Becker, 2026).

A bit flimsy basis but an interesting paper nonetheless.

lumost• 8 days ago

I do wonder if the productivity gains control for increased bug findings and token costs.

Mistletoe• 9 days ago

When I hear recursive self improvement all I remember are the ridiculous articles a few years ago about how 3d printers were going to make themselves and take over the world.

anthonyrstevens• 8 days ago

So because some optimistic wacko said something implausible in the past you dismiss the entire topic and have nothing to contribute to the conversation. Gotcha.

vatsachak3• 9 days ago

RSI isn't anything new though; computers have been used to make computers better for about 80 years now.

Imagine having a secretary who could read 1 million records and give you back your answer in 100 microseconds, for just 10 cents an hour. That's Postgres.

So I'd imagine that if R&D can be automated, everything becomes better and cheaper but we'd all lose our jobs, as secretaries did to postgres. UBI season

eru2• 9 days ago

Imagine having a secretary who could read 1 million records and give you back your answer in 100 microseconds, for just 10 cents an hour. That's Postgres.

Well, that secretary can only answer very specific questions in a rather peculiar format.

[...] but we'd all lose our jobs, as secretaries did to postgres.

I doubt many secretaries were replaced by postgres.

However, you might like reading about https://en.wikipedia.org/wiki/Unit_record_equipment

inglor_cz• 9 days ago

"I doubt many secretaries were replaced by postgres."

The maximal relative share of typists, secretaries, admin assistants et al. in the US workforce was 4% and this apogee was reached in 1980 after ~ 100 years of sustained growth. In 1980, the curve bent and started declining as systematically as it used to rise. Now it is less than half of that. So are other office roles.

https://forklightning.substack.com/p/the-past-present-and-fu...

(Look at the "The Rise and Fall of Office Work" graph in the middle of the page.)

Given the neat correlation with the tech revolution, I'd be surprised if there wasn't a causal relation. Maybe not exactly with Postgres, but with automation in general. Data collection and management does not involve nearly as many people as it once did.

goatlover2• 9 days ago

That would be a failure of imagination when it comes to everything secretaries actually do. Which is a problem with the idea that AI is coming for all our jobs anytime soon. It totally undersells everything humans do on the job.

ACCount375• 9 days ago

I think humans totally oversell just how special "everything they do on the job" is.

"No, you don't get it, it's me specifically who has a job that's way too special to be automated away!"

ben_w10• 9 days ago

So I'd imagine that if R&D can be automated, everything becomes better and cheaper but we'd all lose our jobs, as secretaries did to postgres. UBI season

Perhaps, but there's a thing in economics, the "resource curse": https://en.wikipedia.org/wiki/Resource_curse

I don't have any formal training in economics, so I can't argue the for and against arguments listed in the Wikipedia page, but it is at least possible that having full automation means that the underclass gets left behind and ignored, cut out of all progress forever and without any help from UBI as we cease to be important to those who control the AI.

(Even owning shares in successful AI companies may not help: Tesla doesn't pay dividends, NVIDIA's dividends are tiny).

vatsachak3• 8 days ago

The world has been getting richer as everything has gotten automated

reader9274• 9 days ago

First thing I thought when reading the title was "is this about those never ending self-help books?"

blitzar3• 9 days ago

What happens when self-help books read self-help books that read self-help books is what you see on linkedin.

charcircuit• 9 days ago

Best evidence in favor of acceleration:

It's important to recognize that LLMs accelerating development of LLMs does not imply it will lead to self-sustaining acceleration.

godelski8• 9 days ago

  > But our models make it clear that such an [intelligence] explosion may not follow if there are diminishing returns (“ideas become harder to find”) or if feedback loops become bottlenecked.

How is this not obvious to everyone? As we advance it becomes more difficult to advance. You obviously make most advancements around the things that are easiest to improve. Then all the easy things are done. So you go onto the next easiest things. They're "the easy things" from that standpoint but that doesn't mean they aren't harder than "the easy things" when you started. Complexity increases as precision increases.

EarlKing• 9 days ago

How is this not obvious to everyone?

Because everyone's thinking around intelligence is incredibly muddled by a variety of factors, and no one is particularly motivated to actually correct anyone's mistaken notions on the matter.

godelski8• 9 days ago

I'm conflicted. On one hand I think we should more openly call people idiots and push back. On the other hand there's Descartes argument for idiots in good company.

I just wish all the people that claimed to care only about truth would actually care about truth. Feels like society is more that trope where someone says a joke to a crowd and no one hears it except one charismatic person who repeats it and gets all the laughs. In reality it feels like the repeated version of the joke doesn't even make sense, it is just vibes.

red75prime3• 9 days ago

How is this not obvious to everyone?

That RSI can be bottlenecked? I guess this is obvious to many people. Whether RSI will be bottlenecked (at some not very interesting stage) is another question.

RandomLensman8• 9 days ago

Why would it not be bottlenecked? Is intelligence so valuable as such?

vidarh• 9 days ago

Because they depend on whether the rate of improvement of self-improvement outpaces the rate of increase in difficulty or not, and at some points they clearly do - e.g. a lot of skills makes the relative rate of subsequent improvement easier for a while.

It may seem obvious that it can't last, but showing the conditions where it can't still matters.

setgree• 9 days ago

The goal of a modeling exercise like this, which you don’t have to buy, is to generate a simple set of initial conditions that can explain things we already know. Then, we can manipulate some initial parameter value to make predictions about things we don’t see, but might.

Likewise, it is obvious that gravity exists, but a simple model that explains where it comes from (in quantum terms) would be a big breakthrough iff it came with plausibly testable implications that could be tested via experiment.

soulofmischief• 9 days ago

Progress is often lumpy, modulated by new discoveries and their applications.

tossandthrow3• 9 days ago

As we advance it becomes more difficult to advance

I don't think this follows.

We have advanced tremendously over the past 200 years, and we are likely going into a time with rapid advancement again.

With advancement, we also develop tools (eg. Llms) that assist advancing.

Windchaser2• 9 days ago

To put it another way, the cost per advancement increases. That’s “as we advance it becomes more difficult to advance”. However, because of the prior advances, you also have more resources to throw at future advancements.

So, then, the question is whether the “profits” on the last advance are enough to pay for the next one. We can define a new term, “affordability”, as what % “profit” you can expect from each advance relative to its cost, telling us whether it becomes relatively easier or harder to continue to advance.

windexh8er• 8 days ago

Then use batteries as an example. Going on 200+ years now most of the major changes have been an easy swap: chemistry. Lead acid --> nickel cadium --> nickel metal hydride --> lithium ion. Improvements in batteries are hard and the surrounding ecosystem has gotten better. Better BMS and better motors have also made gains to increase the efficiency of the battery. Compounding gains within the ecosystem doesn't truly improve the battery directly, but improves them for implementation and use cases.

All the remaining problems with batteries are a hard solve because it's tradeoff after tradeoff. Lithium sulfur, for example, could be amazing but die after a couple hundred cycles at most [0]. So... If LLMs are so novel, as many continue to claim, why hasn't this been solved? Probably because LLMs are an interpretation of our current understanding of everything. An LLM currently only does what any human can do, albeit in a compressed timeframe, mostly.

I think it follows. LLMs have made some considerable improvements, but use "reasoning" (really just a mechanical, autoregressive, loop: generate tokens --> append tokens to context --> feed the expanded context back into the model --> generate the next token --> repeat) as an example. You can do this by hand as well in a traditional chat volley, but it's slower. So we improved LLMs by putting them in an automatic loop (oversimplification - but at the end of the day holds water).

Until LLMs showcase true external breakthroughs that aren't driven by human guidance (not happening anytime soon) we are in this loop of hacks being used to improve the 80% (the LLM itself). Notable jumps? "Reasoning" and with Mythos-like models we now have "Advanced Reasoning" (by leveraging a more capable looping framework such as an agentic harness).

[0] https://advanced.onlinelibrary.wiley.com/doi/10.1002/adfm.20...

breuleux5• 8 days ago

We have advanced tremendously over the past 200 years

Would most of that have happened if we hadn't found oil, though?

It's difficult to precisely identify how much of progress is owed to intelligence, and how much is simply energy availability. Energy is intrinsically transformative -- the dumbest of organisms can take over the world in a blitz if they can process available energy better than the others, whereas the smartest of entities is going to be ineffective without fuel. Intelligence can, of course, unlock the capability to exploit new energy sources, but there's still an intrinsic physical distribution of it that's outside of anyone's control.

exe34• 9 days ago

How is this not obvious to everyone?

Probably because it's not true. We had shitty neural networks for decades before the recent explosion. That particular branch may be a dead end, but there could be others lurking and waiting for their time.

bryanrasmussen3• 9 days ago

How is this not obvious to everyone?

because people have other ideas https://en.wikipedia.org/wiki/Technological_singularity

WJW• 9 days ago

People have other dreams, and have written about them in science fiction. But science fiction can just handwave away resource problems that exist in real life so that the plot can happen.

Just as the popular "grey goo" nanomachine disaster can easily be shown to be impossible due to resource imbalances and energy shortages, AI recursive self improvements rapidly slows down due to problems with complexity, training data and compute availability (whether due to actual processors or just due to energy demands).

Majromax• 8 days ago

because people have other ideas https://en.wikipedia.org/wiki/Technological_singularity

In a weak sense, singularities are common and should be expected every so often. In a mathematical sense, a singularity is where a model 'blows up' and neglected terms become part of the dominant balance.

Industrialization hit a point of diminishing returns, but the industrial revolution was nonetheless a 'singularity,' where life afterwards was qualitatively unpredictable to people who lived before. Likewise, agriculture was such a technological singularity to hunter-gatherer ancestors.

I could even make a decent argument that writing and literacy were such a singularity, making inconceivable social organizations routine.

In that weak sense I expect AI to be a singularity, recursive self improvement or no. Life in 2050 may be completely unpredictable to someone who was taken out of time in the year 2000.

The remaining questions are speed and intensity, and both of these questions are related to RSI. If RSI works, then the 'fast takeoff' visions become more plausible where society transforms over months to a few years – at least locally where the enabling technologies have diffused. If not, it might take a couple of decades.

FrustratedMonky3• 8 days ago

"How is this not obvious to everyone? As we advance it becomes more difficult to advance. "

It's a type of recent change blindness. Because for the last few years there are seemingly impossible breakthroughs every few months. Literally, things everyone said would take 30 or 100 years, happen within months. It is really easy to think this will continue. The new normal. Really, we haven't seen it slow down at all, so why would we expect the miracles to stop all of a sudden?

It seems like accelerating, and decelerating, are both fair game as future directions at this point.

So it is nice that the paper put some weight behind the argument that this could all grind to a halt for a lot of different reasons beyond the hype.

breuleux5• 8 days ago

Literally, things everyone said would take 30 or 100 years, happen within months.

"Everyone"? Plenty of people in the 50s said we'd have sentient humanoid robots before the year 2000. And flying cars. Sure, many people underestimate the rate of change. But techno-optimists have consistently overestimated it, and they still do.

godelski8• 8 days ago

Fair but there's two things I think need to be added

1) predictions are usually made assuming either constant effort or continued growth. They will be extremely conservative when the environment changes and you start dumping trillions of dollars into solving those problems.

2) while there's been real progress that doesn't mean it isn't being over sold. We do need to differentiate the signal from the noise

jamesrcole• 8 days ago

As we advance it becomes more difficult to advance. You obviously make most advancements around the things that are easiest to improve. Then all the easy things are done.

This isn't some foregone conclusion. It completely depends on the rate at which the intelligence and abilities of the AI increases. If that rate was high enough, then the harder and harder problems would become easier and easier for it.

mpalmer• 8 days ago

Just like fully extending a ladder brings the moon closer.

dinfinity2• 8 days ago

Actually, evolution seems to show the opposite: The rate of advancement has only sped up, with billions of years between significant changes going to millions, to thousands, to tens and arguably to mere years now.

Having said that, we're probably looking at an S-curve with the physical limits of reality getting in the way in the end.

reducesuffering• 9 days ago

Reminder that while there are many naysayers who have been on the wrong side of AGI development progress the last decade…

There are two $1T companies who are all-in on RSI internally right now. They are supported by $20T of market cap plowing R&D into their efforts. You can think it’s dumb money at your own peril, however the market rewards intelligent allocation…

baq• 9 days ago

The market can also be quick and brutal to punish mistakes, especially when leverage is high. We can be in 1998, but we can be in 1995; you stand to make a ton of money if you know which one we’re in

goatlover2• 9 days ago

How do the markets know it's intelligent allocation before RSI is actually a thing? What makes those companies immune to failure?

not-a-llm2• 9 days ago

them having an actual product with huge demand makes them immune

Just like Tesla was immune from all the naysayers which were saying its a highly unprofitable company which will 100% go bankrupt because its economics dont make any sense, and they lost huge amounts of money shorting the stock

mbgerring• 9 days ago

“The market rewards intelligent allocation” is such a straightforwardly false statement that I can’t believe anyone still says this with a straight face. The last ten years of the US economy have just been scam after scam after scam, and people just keep saying this.

blitzar3• 9 days ago

NFTs were worth more than $1 trillion so we know that they were "better" than the Ai efforts of today because “The market rewards intelligent allocation”

drdrek• 9 days ago

I love that Sci-Fi fan-fiction is having a moment <3

1317• 9 days ago

NB this is about AI, not about improving your self

khalic• 9 days ago

No DOI?

casey2• 8 days ago

Notice: viXra.org only accepts scholarly articles written without AI assistance. Please go to ai.viXra.org to submit new scholarly article written with AI assistance or rxiVerse.org to submit new research article written with or without AI assistance.

The Economics of Recursive Self-Improvement [pdf]

Related Stories