What happened in 2023 and 2024 actually
Nitpicky but it's worth noting that last year's AI capabilities are not the April 2025 AI capabilities and definitely won't be the December 2025 capabilities.
It's using deprecated/replaced technology to make a statement, that is not forward projecting. I'm struggling to see the purpose. It's like announcing that the sun is still shining at 7pm, no?
And the hype was insane in 2023 already - it's useful to compare actual outcomes vs historic hype to gauge how credible the hype sellers are.
Maybe progress over the last 2-3 months is hard to see, but progress over the last 6 is very clear.
Could be data is lagging as sibling comment said but this seems wildly difficult to report on a number like this.
It also doesn't take into account the benefits to colleagues of active users of LLMs (second order savings).
My use of LLMs often means I'm saving other people time because I can work through issues without communications loops and task switching. I can ask about much more important, novel items of discussion.
This is an important omission that lowers the paper's overall value and sets it up for headlines like this.
This is because the economy is not a static thing. If one variable changes (productivity), it’s not a given that GDP will remain constant and jobs/wages will consequently be reduced. More likely is that all of the variables are always in flux, reacting and responding to changes in the market.
However, the parent comment is about an examination of what has happened so far and facts that feed into the paper and its conclusions.
I was focused on what I see as important gaps in measuring impact of AI, and its actual (if difficult to measure) impact right now.
Mostly people aren't worried about productivity itself, which would be weird. "Oh no, AI is making us way more productive, and now we're getting too much stuff done and the economy is growing too much." The major concern is that the productivity is going to impact jobs and wages, and at least so far (according to this particular paper) that seems to not be happening.
Unless twice the work is suddenly required, which I doubt.
I would also be surprised if the twice the work was "suddenly" required, but would you be surprised if people buy more of something if it costs less? In the 1800s ordinary Americans typically owned only a few outfits. Coats were often passed down several generations. Today, ordinary Americans usually own dozens of outfits. Did Americans in the 1800s simply not like owning lots of clothing? Of course not. They would have liked to own more clothing, but demand was constrained by cost. As the price of clothing has gone down, demand for clothing has increased.
With software, won't it be the same? If engineers are twice as productive as before, competitive pressure will push the price of software down. Custom software for businesses (for example) is very expensive now. If it were less expensive, maybe more businesses will purchase custom software. If my Fastmail subscription becomes cheaper, maybe I will have more money to spend on other software subscriptions. In this way, across the whole economy, it is very ordinary for productivity gains to not reduce employment or wages.
Of course demand is not infinitely elastic (i.e. there is a limit on how many outfits a person will buy, no matter how cheap), but the effects of technological disruption on the economy are complex. Even if demand for one kind of labor is reduced, demand for other kinds of labor can increase. Even if we need less weavers, maybe we need more fashion designers, more cotton farmers, more truckers, more cardboard box factory workers, more logistics workers, and so on. Even if we need less programmers, maybe we need more data center administrators?
No one knows what the future economy will look like, but so far the long term trends in economic history are very promising.
Outside of goods, there must be more schools than ever before, and more academics qualified to teach at them, but is it bringing down the cost of education?
Can you link to the evidence you’re talking about? Did companies only start being greedy in 2020? I would argue the “greedflation” after Covid 19 can mostly be explained by the fact that there was a supply shock combined with a spike in demand as people people shifted consumption to goods and away from services during the pandemic. Some economists, like Isabella Weber, argue that the pandemic served as a coordinating mechanism for sellers to raise prices, but even Weber would never argue that supply and demand are “not that closely related anymore.”
> Outside of goods, there must be more schools than ever before, and more academics qualified to teach at them, but is it bringing down the cost of education?
Well, yes supply has increased but so has demand. In 1960, 7% of adults 25+ had a college degree. Now that’s 37%.
Further the demand is less price sensitive than usual because the government freely lends money to spend on education (us government currently has $1.6 trillion in student debt) and because we mostly encourage kids to get degrees at any cost.
Finally, despite the growth in total number of universities and students, spots at elite institutions are limited and thus scarce and highly valued.
Good example for discussion because I think that it's well known that many people who have attained higher education at great expense nevertheless struggle with unemployment and underemployment, and groan about how competitive things are in academia with thousands of applications for every placement. If we're oversupplied with qualified yet underemployed academics who could be teachers, and yet teachers are unavailable to meet the demand of students who want education.. how is that not a disconnect between supply and demand?
Whenever a huge disconnect like this between supply/demand is pointed out, economists will retreat to discussion of regulation and policies, inflation, macroeconomics conditions. Sure, fine, great.. but that's just another way of saying supply/demand/cost are not closely related.
So you could work on more things with the same number of employees, make more money as a result, and either further increase the number of things you do, or if not, increase your revenue and hopefully profits per-employee.
I'm not sure why you doubt it. In a company, efficiency improvements never mean you don't have to do as much work. They always mean you just have more time in the day for extra work. It's part of the reason I really don't care at all about automation, LLMs, etc, at least from an efficiency perspective. (there's another case to be made for automated response, etc.) I hear people say that it opens them up to do more of the interesting work they care about, but in practice I've never seen this be true.
It feels lot like when people in IT talk about how they are "always learning new things." Yes, that's true, but what you're learning is worthless: An updated UI for Defender, a new proprietary query language, branding hierarchies for tech company products, etc. It's all worthless knowledge.
I like this sentence because it is grammatically and syntactically valid but has the same relationship to reality as say, the muttering of an incantation or spell has, in that it seeks to make the words come true by speaking them.
Aside from simply hoping that, if somebody says it it could be true, “If everyone’s hours got cut in half, employers would simply keep everyone and double wages” is up there with “It is very possible that if my car broke down I’d just fly a Pegasus to work”
But more generally, my comment is not absurd; it's a pattern that has played itself out in economic history dozens of times.
Despite the fact that modern textile and clothing machinery are easily 1000x more efficient than weaving cloth and sewing shirts by hand, the modern garment industry employs more people today than that of middle age Europe.
Will AI be the same? I don't know, but it wouldn't be unusual if it was.
This makes sense. If everyone’s current workloads were suddenly cut in half tomorrow, there would simply be enough demand to double their workloads. This makes sense across the board because much like clothing and textiles, demand for every product and service scales linearly with population.
I was mistaken, you did not suggest that employers would gift workers money commensurate with productivity, you simply posit that demand is conceptually infinite and Jevons paradox means that no jobs ever get eliminated.
In the the past 200 years we've seen colossal productivity gains from technology across every area of the economy. Over the same period, wages have increased and unemployment has remained stable. That's where my priors come from. I'll update them if we get data to the contrary, but the data we have so far (like this paper) mostly confirm them.
More people are also available since the fields are producing by themselves, comparatively. Not to mention less of us die to epidemies, famines and swords.
My company just redid our landing page. It would probably have taken a decent developer two weeks to build it out. Using AI to create the initial drafts, it took two days.
It's not miraculous but I feel like it saves me a couple hours a week from not going on wild goose chases. So maybe 5% of my time.
I don't think any engineering org is going to notice 5% more output and layoff 1/20th of their engineers. I think for now most of the time saved is going back to the engineers.
I would (similarly insultingly) suggest that if you think this is true, you're spending time doing things more slowly that you could be doing more productively by using contemporary tools.
But here's the thing - there is already plenty of documented proof of individuals losing their job to ChatGPT. This is an article from 2 years ago: https://www.washingtonpost.com/technology/2023/06/02/ai-taki...
Early on in a paradigm shift, when you have small moves, or people are still trying to figure out the tech, it's likely that individual moves are hard to distinguish from noise. So I'd argue that a broad-based, "just look at the averages" approach is simply the wrong approach to use at this point in the tech lifecycle.
FWIW, I'd have to search for it, but there were economic analyses done that said it took decades for the PC to have a positive impact on productivity. IMO, this is just another article about "economists using tools they don't really understand". For decades they told us globalization would be good for all countries, they just kinda forgot about the massive political instability it could cause.
> In contrast, this article, i.e., the paper it discusses, is based on what has happened so far.
Not true. The article specifically calls into question whether the massive spending on AI is worth it. AI is obviously an investment, so determine whether it's "worth it", you need to consider future outcomes.
I honestly think computers have a net negative productivity impact in many organizations. Maybe even "most".
https://usafacts.org/articles/what-is-labor-productivity-and...
Even more surprising for me is that productivity growth declined during the ZIRP era. How did we take all that free money and product less?
This is not correct. Your link is only referring to manufacturing productivity, not overall productivity, which continued to rise. This Economist article has better information: https://archive.vn/6asPb. One hypothesis in the article that seems the most logical to me is that return on investment became so much better in other economic sectors that it siphoned talent from manufacturing.
Could you say a few more words on this please? Are you referring to the rise of China?
https://www.theguardian.com/world/2017/jul/14/globalisation-...
https://business.columbia.edu/sites/default/files-efs/imce-u...
But in my own words, while I think economists have always acknowledged there would be some "winners" and "losers" from globalization, I think there is widespread belief now that early globalization boosters misjudged how bad it would be for the "losers", they underestimated how much it would contribute to inequality, and they underestimated the destabilizing effect this would have on social and political systems.
At the end of the day, it doesn't really matter what the reality is for globalization's benefits if so many people are turned off by it that they want to "burn the whole thing down", as that's basically what's happening now.
But I think it's pretty clear what the parent poster means - that it's too early to imply that GenAI won't impact jobs.
Obviously you understood that that was the point parent wanted to make - because you argue against it.
What call. Maybe some readers miss the (perhaps subtle) difference between "Generative AI is not ..." and "Generative Ai is not going to ..."
Then first can be based on fact, e.g., what has happened so far. The second is based on pure speculation. No one knows what will happen in the future. HN is continually being flooded with speculation, marketing, hype.
In contrast, this article, i.e., the paper it discusses, is is based on what has happened so far. There is no "call" being made. Only an examination of what has heppened so far. Facts not opinions.