It's funny that this is missing the negative effects or the possibility of an economic crisis scenario. Many data centers are being built by companies that won't exist in a few years, as the winner will take all. This is why XAI is already renting compute to Anthropic. This goes as well without saying that in few years, compute pricing will drop heavily.

Another effect is increased inequality in the US and other Western countries as software/AI/hardware companies are able to do more with less.

The increase in manual labor wages is not systemic. It affects only the labor used to build data centers.

> As the nurse incorporates AI into her job, the nurse is able to oversee and accomplish more. She can spend more time talking with patients and helping them understand diagnoses. Productivity increases.

This feels economically naïve.. If AI lets one nurse do the work that previously required two, the default pressure in a cost-driven system is not "great, now nurses can spend twice as long with patients" It is "great, now we can run the same operation with fewer nurses"

I like how the least optimistic scenario is simply LLMs not making a difference, instead of the very real possibility of them damaging education, destroying people’s attention spans and their ability to learn, eroding trust within societies, increasing the wealth inequality and leading to class wars.

I believe the net effect of LLMs is, at this point, firmly negative, and it’s going to take years until they start making up for the mess they’ve created.

For me the ideal scenario would be us having humanoid robots with AI.

This would free up so much time for me and even in the worst case if I lost my job, I can live in village side growing my own food and being fully self sustained with the help of humanoid robots and solar energy. That would actually be a dream come true

> The result: the nurse’s job changes

> As the nurse incorporates AI into her job, the nurse is able to oversee and accomplish more. She can spend more time talking with patients and helping them understand diagnoses. Productivity increases.

Except what actually happens is that the patient contact time stays the same, but then the added productivity is used to justify fewer on-staff nurses.

Finding 4 is a must-read:

> Finding 4: Labor vs. capital share: The pie will grow, but a larger share might go to capital

> Today, of each dollar the economy produces, about 60¢ goes to workers and 40¢ go to capital.

Modest scenario: 40.6% to capital in 2030 (labor share down 0.6 points)

Substantial scenario: 43.9% to capital in 2030 (labor share down 3.9 points)

Extreme scenario: 54.8% to capital in 2030 (labor share down 14.8 points)

So all the growth goes to capital. For knowledge workers: substantial unemployment and declining wages. For other workers: wages may increase on paper, but not really for actual purchasing power. (note: all of their estimates are for 2030, not for a distant future)

Very dystopic indeed.

> Like any economic model, this one has limits. For example, we did not include scenarios where humanity develops hyper-capable robots.

I didn’t see the scenario “what if we (Anthropic) are wrong about the value of AI”. Which IMHO is the most important one to consider given it is the blind spot pretty much all the AI labs have. What if the economic impact is negative? I don’t understand how you can study such a topic without considering the risk of being wrong, not in magnitude but in kind.

Feels like we're going to read a blog soon enough that mirrors Wendell Berry's opening to The Unsettling of America:

The commission said, according to an article in the Louisville Courier-Journal, that the country’s biggest farm problem was a surplus of farmers: “. . . the technological advances in agriculture have so greatly reduced the need for manpower that too many people are trying to live on a national farm income wholly inadequate for them.” The proposed solutions were to find “better opportunities for the farm people,” “a more comprehensive national employment policy,” “retraining programs,” “improved general educational facilities,” etc. Both the commission and the writer of the article had obviously taken for granted that the lives and communities of small farmers then still on the farm—and those of the 25 million who had left the farm since 1940—were of less value than “technological advances in agriculture.” There seemed also to be no official doubt that adequate solutions were to be found in government supplied “opportunities,” facilities, and programs. Reading that article, I realized that my values were not only out of fashion, but under powerful attack. I saw that I was a member of a threatened minority. That is what set me off.

I think (hope?) the fundamental fallacy in arguments like these is assuming that individual task-level productivity improvement will ladder up to economy-wide transformation.

IMO knowledge work productivity has gone up a massive amount since 1990, and companies could in fact fire most of their knowledge workers while still getting their most critical work done. (These are the bullshit jobs/email jobs.) But companies mostly haven't done this, which implies other reasons for employing so many knowledge workers.

Even if AI becomes capable of doing everything, I would predict there will be a period of many years/decades where companies still employ a similar number of people to sit around and babysit/watch what the AI does.

> Ultimately, what the economy looks like in 2030 depends on many factors, like what AI can do, and how companies and workers choose to adopt it.

So the answer is basically "beats me lol"

People get income from one of three places: capital income, labor income, or the welfare state. If this technology truly unlocks a holy panacea of productivity with a commensurate drop in employment then capital’s share of the national income can and should provide for a wider and deeper welfare state. Nothing new need be invented here. Anthropic’s elaborate scenario-building makes appropriate preparedness sound like a problem requiring heroic prognostication when a simple theory of distribution will suffice.

If it's anything like that web page, I'd say the future looks... slow.

Definitely a bit of positive framing. My big take aways, either stagnant wages or significant paycuts for knowledge workers and overall decreases in employment rates on all scenarios. Worst of all...

That's only forecasting out to 2030.

Those with the money and influence have always been in conflict with the masses that actually make the economy run. They would love nothing more than to do away with that problem. It allows them to control the terms of the economy.

I think we have a real problem because we know what we still need to build and as soon as it is build, it will work.

Software development for example is enhanced with running your claude locally.

But lets be honest, the right way to adjust 'project' is by writing some spec in a text field as an input, letting the ai do whatever it wants, having expert ais make sure its secure, then you do the last review and its done.

What do we need for this? A full end to end integration. Its just something we need to build, nothing we need to invent.

The nurse example is the same. We only need to build the computer end to end flow and plugin AI in the right spots and finetune for country specific requirements.

The industry is building this 'agentic layer' today. The Hyperscaler are building this layer too.

There is a golden phase were we will build all of this and were we cleanup old stuff and were we adjust with the current model capabilities and change systems and this will take a little bit of time, but the bottom line is already loosing left and right.

The Blender support from Fable and Astra? Probably kills a lot of peoples jobs on fiverr.

The small software someone needs? Got already killed by earlier models. Just a few days ago some architect (never written software) talked about how he vibecoded some tool for tracking working hours.

Man even a basic website was expesnive to build just 2 years ago.

The smaller, easier and better the code base, the easier it is to remove people from this.

The software market in China is imploding and we will also see budget adjustments to move away from consultants and move them into AI Token budgets. We probably reached peak software development.

You’d think the most powerful AI could make a website that doesn’t completely break usability by destroying scroll.

Good or bad, nothing ages as poorly as economic or market predictions.

One guy will die with "all the things" and the rest of us will just die.

> Only humans can do some tasks

> For instance, AI can’t bathe a patient.

Oh sweet, we're left with literally cleaning up feces. Thanks AI!

The biggest critisism from my side is not how AI will take over jobs but I believe we will see peak of certain things.

Creating images (menues, flyers, whatever) became so cheap, that we reached peak graphi design GDP generation. This will go down.

Software development for complex systems is enhanced but not yet solved. But a lot of small software development stuff for sure. We reached Peak Software Development and this GDP will go down, not up.

I think it looks like the labs front running everyone’s work, freeloading on the last bits of human intelligence to ensure their owners a place as our masters.

Watch Ann Droid on the Beeb for a dark but funny prediction of what it will be like wrt to nursing care for the elderly.

I'm surprised to not find the word "deflation" anywhere. It doesn't necessarily seem likely in the end given governments would presumably just print money, but intuitively AI seems like the ultimate deflationary technology. Fairly generally, not just for labor.

It's been 4 years. Nothing meaningful.

Are you guys feeling all high and mighty doing all these predictions? I think the money they are spending on trying to inform policy could be better spent on getting all major labs together with the government, and making sure that

1. At least 10% of your net spend is on safety and harm mitigation before you accelerate yourselves into doing real harm to real people tomorrow.

2. Pause all AI enhancement until a proper mechanistic study can be done by a working group funded by you all to understand how AI model Safety can be enforced by something less lame than a Constitution/Spec.

One simple thing can’t make everything better or everything worse—every, big change is the sum of time, actions, and many actors. The thing you describe as a result is only the end of a process. I hate simplicity when people talk about the future.

It would be interesting to know what percentage of global inference is spent on projects/problems that would otherwise "wait"/"decay" (if generative ai wasn't a thing). Or maybe the question is: What percentage of global inference is spent on cleaning up and migrating old systems? "New" vs "Old". The decisions I make today can now incorporate what I once termed as "nice to have" when developing a sytem.

I thought for a moment "our" might refer to Anthropic's future

At least they are thinking about this. The possibilities still seem to limited though. A coder and a designer will 'retrain' to electrician or nurse, because we are busy having a construction boom. But do all of the individuals in this system get to choose the construction boom? Or are we swept up in the desires of the capital deploying owners?

A thought experiment: robot bees can do everything biological bees do --- pollination for the crops and making honey, but they are tireless and cheaper to maintain.

Would beekeepers and farm owners still want to keep biological bees?

I dont think in the future people will run these overpriced frontier models and instead have their specialized local models.

Even then, assuming the difference is that huge between frontier and open-source, a specialized „human worker“ _should_ be able to fix the gap.

I can't scroll through this site on Firefox and a recent mac, anyone else?

I just wonder when we will see 8ft tall humanoid robots walking through the streets.

It would have been useful to use the internet from 1995-2015 as a reference model for economic impact, and then ask how the trajectory of AI will resemble it.

> As the nurse incorporates AI into her job,

But she won't, will she? Her employer will do the incorporating. With very different objectives.

Prices will fall to their raw material costs?

CEOs will make AI drive into the office for culture.

Knowledge workers are safe until FSD.

I try to never be that guy who complains about the way a web page scrolls, but this page is completely unreadable.

Judging the quality of this blogpost/presentation, AI is centuries away from being able to replace humans.

> Think of a day in the life of a nurse You can think of any job as a bundle of tasks that someone does. She does rounds to check on a sick patient. She draws blood.

Why is the nurse a she?

the make your own predictions part is a nice touch. how sensitive are the outcomes to the capability assumption

The sociopaths running the show at present are probably hoping for a future that looks like the film "Elysium".

What a reversal from Amodei's doom marketing!

This is of course a slop paper with vague assumptions and equations that no single human economist will ever read. The messaging is "it will not be so bad", in line with the recent Economist piece.

We are ready for the midterm elections!

How about we instead make Earth be a planet good for humans and living beings?

We can slow down AI development and make it a companion for us, not a replacement.

We don't need to automate ourselves out of it. Or take the risk of extinction.

Just like with nukes, we decided to not blow up the planet. We can too change this.

GDP is a terrible measure. I bet we will see a huge drop on HDI with so many suicides and people going through hell because their job got automated or they no longer see meaning in life, feeling powerless.

Instead they should run an estimate on how many would suicide until 2030 if the 'extreme scenario' happens.

Big GDP growth and a lot of people no longer finding purpose in their lives, going bankrupt, losing their homes and family etc.

For Anthropic, you are just clearly a number. They only care about output.

They have long forgotten the human being.

Like others in the thread have pointed out, the nurse example is interesting.

From my point of view, the point about the nurse is especially relevant because, in the future, dealing with an aging population is going to become a major undertaking and a major burden.

Demographically, we simply don't have the number of people we need to give elders the care that they'll probably have to have near the end of their life, so we have to rely on AI to help us get through it.

Probably, we will see this happen in Japan and South Korea before it hits the United States.

Before AI, my worldview was that our monetary system was concentrating wealth in a negative-sum game; essentially; every so often, taking wealth from 100 people, destroying half the underlying value and giving the remaining value with a 2x nominal markup to 1 rich person who is already spending more than they can physically spend and so they won't realize that the 2x amount of wealth they got on paper only buys them a quarter of the stuff compared to the previous decade (for example). They won't know how frothy their wealth is until they start selling large quantities of their stocks. It's just buying nominal 'growth' at the cost of risk/fragility/frothiness.

In such system, you could hit record nominal GDP numbers every year while simultaneously destroying the real economy. This was my view of the economy before AI.

Now I think AI will deliver real efficiency gains but also, it provides a convenient narrative to explain why people might be losing jobs in a way which distracts from much bigger systemic issues which existed prior to AI.

My negative-sum economy example with 'booming GDP' is possible without AI even as net economic value is being destroyed. So basically we won't be able to tell if people are losing jobs to AI or to systematic wealth transfer mechanisms which predate AI.

I suppose if we see people losing jobs permanently and workers in other sectors aren't seeing wage increases with more openings, that may be the most reliable indicator we have that the system is losing real wealth faster than AI productivity can offset it.

The bad news is that we won't see any problem in the numbers; the problem would show up as polarization and radicalization of the masses... Which is more qualitative . Since almost everyone is being radicalized concurrently as everyone is affected by the same pressures (both from monetary system and AI automation). The question is not "how many people became radicalized this quarter?" but "how much more radicalized is everyone feeling this quarter?"

AI productivity discussion go!

The model seems defensible, but for some reason it isn’t passing the sniff test. Maybe its that the extreme scenario seems wildly optimistic, or that the observed modest productivity gains are framed as either the trend or “just the start of something amazing !”.

The issues are that

1) The median AI use case isn’t increasing productivity but expanding capability and reducing the need to communicate across teams. Ex: if someone needs a dashboard or a ppt designed, now they don’t have to ask another team. AI is substituting for intern and early career workers in someone else’s firm.

This is also borne out in the hiring data.

2) The cases that show maximal productivity gains are when AI is leveraged by experts.

This is a problem, because the evidence for automation gains is poor. AI automation seems to be going to the same place ML projects went to die.

If augmentation is the primary lever that lifts up productivity, and it is dependent on AI being matched to experts, then the limiting factor is expertise.

This makes the fact that AI is both, taking out early career roles AND one shooting education, a divine irony.

Where exactly are those future experts supposed to come from? The ones with taste and the ability to deliver us this AI driven GDP growth?

AI is also setting up a conflict between liability, insurance and productivity, and thats a whole point in itself. However it will be expressed as a drag on productivity.

I found myself interested in the papers analysis of wages, so I did some digging.

The researchers model says the following about hourly wage growth in the substantial and extreme scenarios:

- the hourly wage* in the knowledge workers group will fall (-1.1% and 11.5% respectively)

- but the hourly wage* of labor across the rest of the labor force will grow (5.9% and 33.6% respectively)

*Important to note, AFAIK this hourly wage decline/growth is a not measured against 2026 dollars, but the counterfactual of what hourly wages would be in the absence of AI.

Okay, so my first take away from reading these numbers is that declining wages in one group appears to be offset by rising wages in the other group. Not terrible all things considering.

But then I started to think about the difference in hourly wages between the knowledge workers group and the non-knowledge workers group (which I'll now refer to as *K* *N*). I had a sneaking suspicion that the inter group wages gap could be large enough where growth in the non-knowledge share of labor doesn't offset the loss in hourly wages from displaced knowledge workers who find themselves migrating occupations.

Looking through the paper, it breaks these groups down as follows:

K — Knowledge / Cognitive:

- Management, Business and Financial Operations, Computer and Mathematical, Architecture and Engineering, Life, Physical, and Social Science, Community and Social Service, Legal, Educational Instruction and Library, Arts, Design, Entertainment, Sports, and Media, Healthcare Practitioners and Technical, Sales and Related, Office and Administrative Support

N — Non-Knowledge / All Other:

-Healthcare Support, Protective Service, Food Preparation and Serving, Building and Grounds Cleaning and Maintenance, Personal Care and Service, Farming, Fishing, and Forestry, Construction and Extraction, Installation, Maintenance, and Repair, Production, Transportation and Material Moving

So I had GPT Sol 5.6 take each of these occupations and find the mean hourly wage using BLS data [1]. The findings suggest that as of 2026, the simple group average for K is $45.17 vs. $23.67 for N. But this is short sighted as this doesn't take into consideration the different levels of employment in each occupation. I asked ChatGPT to account for this using the same BLS data, which gives a weighted average of $41.50 for K and $23.40 for N.

This means that knowledge workers are currently making ~double what non-knowledge workers make in the US. The crucial question is how does this change when considering the wage decline/growth between K and N across the papers three scenarios?

Here's the breakdown using the simple mean for hourly wage:

Scenario Knowledge Non-knowledge Gap K -> N penalty

-------------------------------------------------------------------

Current $45.17 $23.67 $21.51 47.6%

Modest $45.35 $23.93 $21.42 47.2%

Substantial $45.03 $25.07 $19.97 44.3%

Extreme $39.98 $31.62 $ 8.35 20.9%

Here's the breakdown using the occupation weighted mean for hourly wage:

Scenario Knowledge Non-knowledge Gap K -> N penalty

-------------------------------------------------------------------

Current $41.50 $23.40 $18.10 43.6%

Modest $41.67 $23.66 $18.01 43.2%

Substantial $41.38 $24.78 $16.59 40.1%

Extreme $36.73 $31.26 $ 5.47 14.9%

Looking at these tables, both in the current moment and across all three of the papers scenario, there is a large wage gap between K and N. Outside of the extreme scenario, this gap is >40%!

My immediate takeaways having done this bit of research is that if Anthropic is right and we're on track for either the substantial or extreme scenarios, *then many of us in this thread could see a large loss in hourly wages in the short term future.*

NOTE: I'm not an economist. I push pixels for a living. But if you are an economist and you find gaps in the following analysis, please inform me.

[1] https://www.bls.gov/news.release/ocwage.t01.htm*

When children behind a battleship with child like mentality try to predict the future.

This could be the most cringe main-character-syndrome thing I've ever seen produced by a company in the history of silicon valley.

And the people treating this as serious economic forecasting rather than a marketing puff post are even more cringe.

If Anthropic truly believed LLMs were going to raise GDP growth to 8% permanently (or even just a few percent), instead of going public to raise capital they could just go massively leveraged long on the S&P 500 and get funding for eternity.

They are not doing that, because AI is largely only good at coding, their models barely have a 6-12 month lead over free ones, and software is only 1% of GDP.

And by making software cheaper and easier to produce, they might increase the amount of software in the world by 10X or even 100X, but it will become a smaller percentage of GDP as AI models become autonomously good at creating it. Things that are abundant and cheap don't raise GDP.

Hey Anthropic: before we worry about raising GDP growth to 20%, maybe we focus on making blog posts that don't crash your browser window if you try to resize it?

Man I wanted to dunk on this, but it’s honestly not terrible. Their predictions - while still firmly rooting for the extreme results - feel more grounded. I also appreciate the little simulator where I could plugin my own predictions (echoing many here, that AI won’t amplify productivity/lift standards so much as replace workers by shitty bosses), and that it seems to caution that, yep, most of these gains will trickle up given current incentives.

Where I remain resoundingly annoyed is how useless this is. It’s pure ass-covering by Anthropic ahead of an IPO, victim-blaming the populace for the outcomes of decisions made by Anthropic leadership and the investor class. Forgive the grisly analogy, but it’s akin to a mugger telling you “only you can stop this” mid-violence: Anthropic could have chosen to allocate more of its capital to reforms that benefit the populace, but they’ve chosen to focus on personal wealth instead via this IPO and for-profit rented access schemes.

It’s nauseating. Their actions still paint a picture of greed and exploitation rather than building a better tomorrow for everyone with AI, and the fact their PR puff pieces keep getting wallpapered up here is gross.

EDIT: I guess another way to look at this is Anthropic making it unequivocally clear that they will not stop of their own accord, and thus it falls to others to stop them. Which if that’s their position, then at least they’re honest about the challenge.

>We study how AI is reshaping the economy because we’re committed to ensuring that this transition is beneficial for society, including workers.

Is it just me, or does their choice of identifying "workers" specifically seem a little peculiar? For me, it reads like "We're committed to making segregation beneficial for everyone, including minorities!"

So far LLMs seem to be a dead end technology. The work product is so poor that it can't effectively replace workers as advertised. If it was priced to margin it'd be more expensive than humans as well. It's objectively a failure.

> [we don’t know]

I haven’t forgot capitalist euphoria from 2023 about cutting worker power off at the root. That’s all these AI Cretins want. Well, that and an unnerving enthusiasm for personifying their AI Golems (Anthr. in particular here).

Sorry to poo poo a post about AI company prompting their underlings to prompt out visualizations.

Hopefully free from the burden of LLM companies.