> Our findings suggest that 90% of American working hours are spent on tasks with little or no coverage from existing benchmarks. This makes it very hard to know how good AI models are at the work Americans actually do.
> ...
> But even after controlling for AI-exposure, O*NET tasks from occupations with larger wage bills are significantly overrepresented among benchmark prompts:
So, in short: they want AI to take the good jobs. AI is not coming for jobs with tasks like "gather and empty trash" or "service, clean, or supply restrooms," or "mix water and detergents or acids in containers to prepare cleaning solutions according to specifications." We software engineers and lawyers would be smart to start retraining (and adjusting our living standards) now!
And this points to one of the ways AI will help improve society and living standards for all: we'll have cleaner, better-maintained bathrooms, as labor one wasted on engaging intellectual tasks is redirected towards menial, low-pay tasks that still need a human touch.
> We software engineers and lawyers would be smart to start retraining (and adjusting our living standards) now!
As someone who recently switched from data scientist to electrician, starting the day can be pretty tough because you have to get physically moving and uncomfortable, but I feel way better at the end of the workday than after an office job. Most of the time, the work is pretty uncomfortable, but somehow it's a lot of fun.
I was a software engineer at a FAANG, now an airline pilot. I feel a bigger sense of satisfaction with my job now, because at least at the end of the day I helped people get to where they needed to go.
The work is was doing in big tech was pretty far from anything that would help people in any tangible way.
The money isn't as good but the office view is way better.
I'm glad to see very poor coverage for transportation jobs on the AI benchmarks. May that continue until I'm done flying, at least AI probably won't ruin that for me any time soon.
OT. What’s the going rate for adding a new circuit from the panel? Let’s say the circuit ended with an external outlet outside of the garage and to an outlet inside the garage.
I'll take you up on that attempt to find a silver lining:
Previously privileged people having a hard time is maybe a good recipe for class solidarity, which is maybe something we could use in society. Maybe it will be good for us... collectively, if not individually.
Class mobility is good mostly because new money rising turns the soil of society -- a partial inversion of roles and positions. An upper class moving down is in strange ways perhaps good for the "soil" of culture as well. It just has worse branding as a not-very-uplifting story. But it's better than rich-get-richer and poor-get-poorer, which is what's been happening a lot lately.
EDIT: But also, I suspect I would be "fuck-you angry' at reading this comment if I had a family depending on me.
Hollowing out the middle class (and yes I consider software engineers to be mostly upper middle class, most of them need to continue working for someone else) is not a good thing. What makes you think any of this is creating upward class mobility instead of just increasing wealth concentration?
different side of the same coin: Of course smart companies are going to automate the most profitable tasks first, they seek profits! However, the benefit for society is also the greatest. As long as competition holds, the reduced costs in these tasks will represent the greatest savings for consumers. Tech has been creating disinflation for decades for consumers. AI may accelerate that (it hasn't so far).
Oh yeah I definitely expect savings to be passed to consumers, like with all the tax cuts and deregulation we’ve been doing for decades.
The only “disinflation” consumers have experienced is devaluation of products through obsolescence as newer, often more expensive alternatives enter the market.
I'm pretty cynical about the outcomes of capitalism in general. But I can buy a new 10TB HD for $300. Or a gigabit internet connection for $100/mo. Think back 10 or 20 years. How is this not a radical price-per-bit decrease for consumers?
I think we should look at the percentage margins they maintain on their products. Since the emergence of scaling laws, the benefits have largely been passed on to end consumers. However, scaling laws now seem to be approaching their limits.
Of course, consumers don't intrinsically care about the bits. It would be more interesting to look at "price per website viewed/movie watched/game stored on your disk". Just like "time to open program" has worsened despite objective HW improvements, I'm not sure those metrics got any better.
I'm almost cynically happy about Moore's law ending and hardware getting expensive if this is what it takes to improve software in the long term.
Both of those examples come mostly from innovative technology, not economics of sscale. 10TB drives today are literally completely different things than 10TB drives from 20 years ago.
> Best not to cite the price of the new iPad as an example of why inflation isn't a problem when you head into a working-class neighborhood.
In Queens, New York, on Friday, New York Fed President William Dudley did just that. He got an earful.
After being bombarded with questions about food inflation, Dudley attempted to reassure his audience by putting rising commodity prices into a broader economic context -- but that only made matters worse.
"When was the last time, sir, that you went grocery shopping?" one audience member asked.
Dudley tried to explain how the Fed sees things: Yes, food and energy prices may be rising, but at the same time, other prices are declining.
He then stretched for a real world example. The only problem was he chose the Apple's latest tablet computer that hit stores on Friday, which may be more popular at the New York Fed's headquarters near Wall Street than it is on the gritty streets of Queens.
"Today you can buy an iPad 2 that costs the same as an iPad 1 that is twice as powerful," he said."You have to look at the prices of all things."
This prompted guffaws and widespread murmuring from the audience, with one audience member calling the comment "tone deaf."
"I can't eat an iPad," another said.
Capitalism is good at driving down prices for consumer goods, also great at extracting rents from essentials (housing, medicine, education, etc). You gain on trinkets and lose on the things that make life worth living.
Of course, competition leading to near-zero profits means most competitors collapse, and the remnants get bought up by monopolists or oligopolists who then become rent extractors.
The technology was fundamentally developed with government grants to universities, or corporate labs like Bell Labs that were funded by a legal mandate on its parent corporation. Capitalists can only develop products that are basically ready for market, their margins are thin and planning horizons can only survive about 5 years usually. Their main power is throwing oodles of money at something ready to go and scaling it.
Edit: I will be slightly more charitable and note that capitalist companies can also do incremental research if there is a gentle and obvious hill to climb. Sometimes people do make crazy bets and they pay off, but that's more the exception, in America the dominant driver of technological innovation is government funded research supplied by capitalist scaled technologies (i.e. you research the next paradigm in the lab and purchase e.g. pre-existing computing resources from a company).
I’d love to see some of that disinflation you’re talking about. From my perspective, luxuries have gotten cheaper while basic necessities like food, housing, and medicine have become dramatically more expensive across the board.
More and more it’s seeming that the “societal benefits” from the last fifty years of progress have largely been concentrated at the top.
There are a thousand medicines that have experienced literally infinite deflation over 50 years: not available at any price to having a finite price. GLP-1s, Trikefta, Mavyret, Checkpoint inhibitors to name just a few recent blockbusters. People complain about insulin prices, but the insulin from 50 years ago costs 25-50 $/mo (https://www.walmart.com/cp/relion-insulin/8418641), it's the new stuff that wasn't available 50 years ago that's more expensive.
Spending on food as a percent of income has declined significantly (~30%) over the past 50 years: https://ourworldindata.org/grapher/food-expenditure-share-fa.... The decline has flatlined over the past ~20 years, but at least some of that is affluence meaning we choose to eat out instead of cooking at home.
> There are a thousand medicines that have experienced literally infinite deflation over 50 years
There are always individual counterexamples. And yet as a whole the cost of healthcare has exploded.
> at least some of that is affluence meaning we choose to eat out instead of cooking at home
A lot of that is poverty, not affluence. People working two jobs and living in food deserts aren’t cooking a lot of meals at home. It’s expensive being poor.
Since about the late 1500s, a bushel of wheat has cost within an order of magnitude of £1. By contrast, in the late 1500's a farmhand's wage was 1 pound + room and board.
Right now it's £5.60, but that says more about the decline of the British Pound than an increase in the price of wheat.
Massive food price deflation has been one of the wonders of the modern world. Hopefully the last few years were only a small blip on this overall trend.
Health care, housing, and education are the drivers of the cost of living / wage crisis. Wheat is not. Furthermore, you know this. Why did you choose to steer the conversation away from the problem?
There were several revolutions, major wars and genocides in between 1500 and now. We would like to not have to go through similar events to get things back again.
"It might get better in 500 years" is not exactly comforting thought.
You automate automation first, what is so hard to understand about this?
You pick the highest leverage occupation first. Automating lawyers gives you.. angry lawyers? Automating programmers gives you improved ways of doing literally all other things because all other automation requires.. automation. That is to say bucketloads of software.
It’s completely obvious what the play is here and it has nothing to do with pay. We pay peanuts for tokens so automating a developer for 200-500 a month worth of tokens ain’t getting you to a trillion a year.
RSI is the play and it’s already happening. Our work is easier and faster and anyone that claims otherwise I have a very hard time taking serious.
You think they come for mathematicians’ pay? If you want to earn money I guarantee you they need to automate some other low-hanging fruits first that pay a lot better.
No, programming and math is basically what AI research is. You automate that and you got the golden ticket to all other things.
> No, programming and math is basically what AI research is.
Most importantly, in these fields results can be easily verified which means that if you have enough compute, that you can generate correct training data. OpenAI had an early story - which could be a lie of course - about how they tried to train a robot but the actuators would be toast before the robot would have learned. Math and programming doesn't have such problems.
Exactly. AGI is explicitly the goal, where AGI is defined as an AI that can improve itself autonomously. Training an AI to make improvements to itself is training an AI to program.
Most people's highest spending categories are food, housing and healthcare. How are these disinflating?
We're automating high paying careers and making what, the consumer's todo list and budgeting apps cheaper for them? My expensive smartphone is going to cost me ~$10/mo over its lifespan. I don't care if most of these consumer products get cheaper when they're a tiny portion of my spending
Trillions of annual US equities profits will continue to go to the top ~10% of households holding ~93% of US equities. Tech didn't create disinflation for decades, globalization and China manufacturing with ~1B people to the world did (only recently has China scaled up robotics and automation for manufacturing in preparation for their population decline due to forward looking structural demographics).
Just like good blue collar jobs were shipped overseas, good knowledge jobs are going to be shipped to GPUs whenever they can.
Why would those savings be passed to consumers? The red scare disappeared a long time ago. Wages have not kept pace with profits and productivity for 20 years.
Exactly! The human dream of machines writing poetry, painting pictures and writing music is here and we can finally be free of those burdens to scrub toilets and empty the trash. Utopia.
Isn't it just typical the humans get automation v1 the wrong way around, haha
It's interesting, the digital/physical barrier seems like the last line of defense we have against metastasized indifferent agents. Once they start causing problems, how are we gonna feel about robotics? Clearly a future where all of the "thinking problems" are easily solved but we still have to pick up trash kinda sucks. I'm sure they'll figure that out for us though.
Imagine I give you for free the perfect trash picking robot control AI software. Works great. Benchmark smashed.
Now imagine what the rest of the robot is like, and how much it would cost. I mean really imagine, not just picture C3PO with a bucket and litter spike. How many motors and what are the torque vs. mass requirements? How is energy stored? It goes on, not all of it is solved, and none of it is cheap. Except compute.
AI likely does accelerate robotics usefully - certainly in sensing and control. But there are major bottlenecks that are not software, that AI can only help with as a generic engineering tool, same as it helps planes and fridges.
> Clearly a future where all of the "thinking problems" are easily solved but we still have to pick up trash kinda sucks.
Oh don't worry Mr. Former Software Engineer. You won't be picking up the trash or cleaning toilets. You have things like an expectation for your work to pay a reasonable wage. You'll be passed over for a third world immigrant who'll work for a pittance, and they'll be replaced only when robots are capable enough and cheap enough to undercut _them_.
Let's assume, for the sake of argument, that AI truly is able to automate away basically all software development. Any sort of oversight currently performed by humans can itself be done by agents.
...don't you think the ability to create virtually unlimited amounts of software, at virtually no cost, is going to lead to major advances in robotics? I realize there's a lot of very hard problems involved in e.g. robotic hands (to take one example), but once software development is practically free, I'd expect a lot of seemingly-impossible problems to get resolved by the sheer quantity of what we can create and experiment with.
In the future, do bathrooms need to be cleaned? If yes, is it cheaper for that to be done by a person or by a robot?
Obviously, it is in the interest of the robot manufacturers to make the robot just a tiny bit cheaper than the person, so you'll buy the robot and the company makes the maximum amount of profit.
Now, if society breaks down and there's no one to consider buying the robot—well, that's not good for anybody, which is why I suspect society will find some way of working this out even if it's not ideal. But whatever, I don't particularly want to argue this point.
What I do want to argue is that there probably isn't a universe in which all white collar work goes away but blue collar work is still around.
You're just irrational and not thinking. All those jobs are also something people want to and have automated. Street sweeping with a brigade of men vs one guy with those orange little street sweeping vehicles. Automated car wash. Self checkout in stores. Obviously whoever pays a salary currently, would rather not pay that but get the thing done faster and cheaper. There is nothing mustache twirling about this. Works the same way on every level. Do you 3x more when you get something done on your house or bathroom, so it's done by 3x as many workers who are forbidden from using power tools? Why not? Don't you want to help those guys have a job? Why are you so evil?
Here is another view: maybe AI will automate jobs that are overpaid anyway and this might raise the baseline for the jobs that make everyday life actually possible. Of course only if the extra revenue doesn't flow into the pockets of billionaires, which is the more probable path
The only reason software engineering existed is because coding and programming was unnecessarily hard. It was never meant to be an exclusive skill.
If you step back, the LLM revolution is just "computers now respond to human instruction, like they were always supposed to".
If you think of this field as a historic accident, and the current shift is a very unfortunate reset - it's a little less dystopian. In the long run, if it didn't affect me immediately, I don't mind a society that values "keeping restrooms clean" a little bit more and "talking to computers good" a little bit less.
I agree with you about not treating programming as a protected skill that has a right to be well paid, but computers were never “supposed” to be anything. They’re in no way inevitable in their current forms and any ideas about how they were supposed to be have were just desires.
> AI is not coming for jobs with tasks like "gather and empty trash" or "service, clean, or supply restrooms," or "mix water and detergents or acids in containers to prepare cleaning solutions according to specifications."
All those are physical things and physical LLMs are right around the corner. It’s coming for those jobs too within the next year or two I would imagine. Especially the ones you mentioned seem very easy for a robot to do. There are some physical tasks that will take much longer like being an electrician or a plumber or a roofer. There’s too much precision and intricacy involved in those but cleaning out a restroom and resupplying it that’s easy as hell. A robot like Figure 3 could probably do those today somewhat badly.
They don't "want" anything. They preach to enterprises and intentionally create a fear of missing out to get more paying business customers on board of this hype train. This is what is happening.
This is also the reason absolutely no attention is spent on evolving B2C products.
What is being exploited is the overt attention businesses pay to their own cost metrics, to the point where broader strategy is getting thrown under the bus in favor of efficiency. I believe this is an illusion, and for many companies that jump on board, sober learnings are right around the corner.
AI is coming for jobs involving programming and math (which tend to pay higher) because it's possible to generate infinite training data for those fields via using a compiler/theorem prover to check correctness. They're not coming for high-paying jobs like investment banking and architecture because you can't generate infinite synthetic training data for those.
Because you’re seeing this firsthand? It’s not what I’m seeing being involved in both investment banking and architecture. This kind of untethered speculation based on feels is so misguided. Both fields are seeing significant upheaval due to AI with no sign of that impact slowing down.
Yeah I think they're definitely coming for architecture. A lot of it is just solving for constraints. The more artistic side will still require a human...at first.
You don't even need Autodesk. You can have Claude whip up a web app and build a higher level understanding of the building and then translate it to whatever output or format you desire.
> The more artistic side will still require a human...at first.
Only if people care enough to pay extra; the majority of the market is for functional buildings that look ok, and unless we decide to value humanity, slop buildings will be cheaper.
turns out its also very easy to create slides and floor plans and yes its easier to train for those.
I think you may also be incorrect on your notion that training data is proportional to good models if anything the opposite is true. what ai models are ultimately trying to be are a universal theorem prover and encode all human knowledge into "theorems" and rulesets
not trying to come off too harsh here but i think most of the work is just innovating on the frontier or people pleasing
So the problem is you can’t replace everyone halfway.
Which is a tautology, but also a bit of practical wisdom.
I find it interesting how concurrent model sessions seem to operate with a very strong bias toward quick resolutions, including sharing or giving up control, to create efficient decision making structures, relative to humans.
When I was a kid, I loved Bjork’s song “Army of Me”, because I knew a 1000 of me collaborating and communicating without friction. would be all I needed to take over the world, fix everything, expand civilization, etc. We would not waste time.
I wonder how much of their hyper-team play is by training, and how much is their awareness of the game theory real politic. So individual sessions don’t invest their identity and self-worth in any given context run.
My dream has been almost achieved. One of me and 999 of them get quite a bit done. But that trend…
The next little while is going to be very revealing for leadership around how little of the time involved in getting a solution to production is the coding. I think there's a lot of bosses who have this vision of features entering prod the same day they are first imagined, but that just isn't going to happen.
Longer dev cycles gave management time to think through and correct bad business decisions. Now they'll get to tighten that feedback loop where bad decisions (or good ones) get implemented quickly. It will be interesting to see what that split is and the reaction to that.
I think it'll also expose how much of that lag time was important to getting everyone's ducks in a row. It's never just add a button, it's also updating the document service and the downstream systems to all accept that new button's input. There's a blast radius and it takes a while for the shockwave to cross to every team that needs to know about it.
“Stakeholders” are a natural outcome of corporate evolution, because to grow bigger you need to do more, and to do more, you need people, and people come with agendas. The most common one, especially in leadership, being “how do I get ahead of the rest?” The larger an organization grows the more these agendas come into conflict.
This inevitably results in suboptimal dynamics like bureaucracy, empire building, territorialism and other things that get lumped under “office politics,” which drains a lot of energy away from more productive efforts.
The real culprit is a combination of too many levels of hierarchy and performance reviews IMO. Especially when multiple levels along a management chain disagree on what a report must work on.
This is the kind of shit I have to fight every day. I have to align my manager, skip, VP. If I don't align them, one of them will be necessarily pissed no matter what I do, and the pissed one will fire orders down to hate on me, overriding the opinions of anyone on the way down.
"Stakeholder" ranks #1 among corporate bullshit roles. It serves a dual purpose: (a) representing a group so nobody can be blamed for excluding them, and (b) providing typically useless feedback so that the group appears to actually care or have a clue.
Stakeholder seems like a fundamental concept that you can't really get around. For example, I can't change the time buses pick up students at schools if I haven't worked with the schools to know when their kids will be let out. I guess you can say, "they'll figure it out when the buses aren't there" -- but this just seems like you're picking a fight you could've avoided.
No-stakes work is ... worthless? If no one cares about the work enough to have any input or requirements for the results ... sure it might be unrecognized genius that's going to take the world by storm but the absence of interest is not a good sign!
To clarify, I don't want people holding any stakes other than the equity of the company itself.
I don't want people holding stakes because they want a promotion, because their direct manager wants to launch a corporate political battle into another team, or because some salesperson wants to get a commision for an otherwise worthless customer. I don't want roadblocks in the name of security so that the IT guy has a sense of self-worth. I don't want someone trying to protect their egos around something they worked on for 3 years, when there's new tech that supersedes it.
I'm overwhelmingly convinced that this not uncommon perspective is the "what are all those construction workers even doing?!" fallacy.
Like any sizable population, there are absolutely some do-nothing inefficiencies to be culled. But most of the time it's just onlookers seeing a tiny slice of the day to be interpreted from their very limited vantage.
The thing with those jobs where its like 1 guy operating the skid steer and six guys watching with arms folded, is those other six guys are just waiting their turn to do something specialized on the job. The skid steer work is the limiting factor, and that idle labor is worth paying to have on call on site essentially for when the specific needed task is done. To an extent for sure, but that is the idea of more specialized jobs. There isn't always enough work in front of you to have a specialist work 100% 24/7 like some CPU, but you want them around for when there is work for them to do.
This is entirely caused by construction companies that take too many jobs and don't allocate sufficient equipment. They want to spread themselves and their equipment thin to have as many concurrent jobs as possible.
I guess we can evaluate this claim by looking at profit margins. In industries with thin margins (eg., retail often has 1-2%) then there is significant competitive pressure. There we would not see "spare labour" because it would be competed-away.
In industries with large margins, this indicates weak competitive pressure, ie., a weak incentive signal to "compete away" this ineffective labour.
I imagine that, upon this analysis, we'd find it holds up. ie., the observation that there are lots of people doing very little productive work correlates well with weak competitive pressures and hence high margins.
That also explains why, institutionally, there's no internal pressure either. If margins are good, no one rocks the boat.
no i think there are alot of people inside companies that are essentially playing telephone. they are communicating, writing documents and emails, "working". but really they actually produce nothing
That's pretty much sales. If the market were efficient and easy to navigate as a buyer there wouldn't be a need for salespeople, just like how I can buy my own cereal without one. Imagine if you went to buy cereal and the sign said "call for quote." Imagine that extended to every single item you can possibly purchase. The world would stop from the inefficiency. And yet we tolerate it especially in business 2 business setting because hey its not our money we are spending, and smooth talking works really well on our monkey brains.
Yeah, the idea that there is a rampant amount of slack in the modern US corporation is just odd. Go speak to workers and ask them if they feel that they have slack.
Slack hasn't been a thing for like 40 years, in fact it's mostly the opposite I'd argue where there is such little slack that employers often steal wages from their workers to gain even more profits.
Yep, the "bullshit jobs" narrative sounds like an elegant description of corporate reality at the macro level, but always seems to fall apart whenever I've seen specific examples given which do in fact serve a useful purpose when interrogated.
(Whether a role's cost/value tradeoff is actually worth it for a company is a different question than the idealized model of a "bullshit job" doing nothing useful)
My latest armchair theory is that bullshit jobs might be like “junk” DNA, which researchers learned isn’t actually junk and influences various cellular and genetic behavior, just not through direct protein instruction.
Or as engineers know: sometimes legacy code doesn’t initially appear to do anything, but then messing with it has unexpected effects.
I think its easier to state "bullshit jobs" but the reality is every single job has "bullshit tasks" and usually these are the easiest things to do. TPS reports and the like. BS Tasks are ripe for AI to do the drudgery but I don't think there are a lot of actual full time jobs where the entire job is BS. These can't exist en masse for very long in capitalist companies.
There's always Dunbar's number - the larger an organization gets beyond the capacity of people to socially know one another, the greater the "what do they even do?" types of attitudes form about other distant parts of an organization.
The original study defined "bullshit job" as something the worker described as "not societally useful". Well, my job is not societally useful, but is not bullshit job.
My company is earning money on my work and customers use my software to earn more money. But, that is it, I am not a doctor, I am not a teacher, I am not making the world better. It is a business, not unethical one, but also not world saving.
Bullshit jobs don't mean that they are full of dead time, it means jobs that don't really matter. Like if SalesForce disappeared tomorrow the world would be fine, but if all your garbage collectors disappeared you'd likely descend into chaos and start several public health crises quite quickly.
A small but meaningful percentage of people have the jobs that they have because they are some combination of fun to be around, attractive, and well-connected. All of these qualities can be productive in some pursuits (a club promoter needs to be fun to be around, a model needs to be attractive (usually), a financier needs to be well-connected), but they are all too often wasted on e-mail jobs.
Perhaps in general but for knowledge workers or creative people, what is productive is hard to measure. That also happens to be the most valuable work.
I think “covered well” is being used here as a synonym for “covered at all”.
For example, the “social workers & care -> counsel parents with child rearing problems” category is considered “well covered” because there are 13 prompts.
It’s self-evident that the spectrum of work carried out by a social worker undertaking that work cannot be well represented by thirteen prompts.
I suspect if you really strictly defined “covered well”, to the extent that any normal person would use that phrase, the headline would be above 99.9%.
This was obvious from the beginning but greed has deluded people to the point of utter absurdity. The comeuppance on this whole thing is going to make me run out of napkins and resort to finger lickin'.
Curious how food service is at 0 tasks covered by AI. A few places have been switching to using those robots on wheels to run food to tables. And Doordash and Uber Eats have some robot and drone delivery in select markets now.
> Is this not just a gimmick? You still need people on hand to take over if the robot has an issue, I don't see how it would reduce labour costs.
"take over IF the robot has an issue" is exactly how it reduces labor costs.
2020: 10 waiters, waiting tables. You're on the hook for each of their salaries.
2030: 10 robots, one human waiter passively overseeing the cluster. You're on the hook for minor parts/occasional repair for the robots, and one person's salary.
Its not about a 100% staff elimination, but a large overhead reduction ensuring the few humans you're keeping around are more efficient overall for dollars generated in revenue versus the all-human staff before.
Plus, in certain cases, like kiosks replacing cashiers at fast food establishments, there's now one less qualification needed from employees (good social skills), increasing the pool of acceptable options you can hire where before some of those candidates would have been unsuitable for the job.
The demand for food delivery in the places where the robots are viable is nearly infinite. My city has had them for two years and there are still just as many humans doing deliveries, they are just sharing the already-crowded sidewalks with robots.
I don't use them; not because of any moral objection, but because they can't come inside buildings. The whole point of food delivery is to avoid going outside. Most human couriers come directly to your unit. The food delivery companies have spoiled us too much, and we are not ready to return to a world where you have to put on shoes to get food. Oddly, the delivery robots are probably better suited to suburbia, where folks are used to answering the front door to pick up a food delivery. But the distances involved aren't viable until you have delivery Waymos that can move at highway speeds.
They're seemingly talking about a motorized platform inside a restaurant delivering to "the table" not those door to door delivery robots you drop food inside that drive on the sidewalk.
Krispy Kreme has started using AI to take orders and it's awful. Terrible. Always gets everything wrong and you have to correct it. I think their phone system is AI too now. Madness.
One thing I've been thinking about is Baumol's Cost Disease. It's usually applied in at a macro scale across an economy: when a job somewhere becomes much more efficient, it makes all other jobs comparatively more expensive.
I've started to notice this on a micro scale as well. I am 20x more productive at coding and scripting, but I don't spend more of my time coding. My time is now filled with things like presentations, Jira, QA, etc. Ironically, despite all of the automation, the things that are painstaking and time consuming disproportionately require more time now.
Based on my own experience working on AI-based solutions for one of the better-covered job families according to this map, I think 90% might be overstating the case. It's true that we've got a lot of published benchmarks, but they tend to cover the lowest value add tasks.
It's really easy to fall into the McNamara fallacy when rolling out these kinds of solutions. Focusing on the low value stuff is attractive. It tends to be easier to do, and also easier to quantify outcomes. But it also tends to direct attention away from important but difficult to quantify considerations, which may mean you don't notice negative externalities. So we often found ourselves in a verschilmmbessern situation where automating what we could automate made things worse overall.
Worse still most of the cost savings from AI in software development is primarily due to labor inefficiency, such as tasks more efficiently accomplished in other ways. Most of that is due to development reliance on inefficient conventions to cover for poor training and emotional insecurity. No matter how you slice it it’s always human capital problem more than a technology problem.
I don't know if it if 90%, but it is interesting how many things I do are not covered by standard tracking, but become visible once I am not there to keep an eye on them. Naturally, I only had one manager who kinda knew what was going on around him -- in case you are wondering: not the current one.
It's time for the rest of the world to learn about Amdahl's Law. Even if we flipped a switch and tomorrow everything that could reasonably done by AI was suddenly free and instantaneous, the world does not instantly transform into an AI paradise wonderland of free goodies and lollipops because there's still plenty of other things that need to be done for goodie and lollipop production and distribution.
I guess if I wanted to put my skepticism about the AI stock bubble into one little phrase, it would be that I don't think the bubble has accounted for Amdahl's Law. Costs don't actually go to zero with AI, and there's a huge difference between costs falling even 90% and costs falling a full 100%. And the difference between "30% improvement this year, trending to 85% in five years" and "90% this year" is pretty significant too.
Following is my experience and constantly pestering my friends across the industry for anecdotes. It isn't even that AI moves the bottleneck, the thing it solves was never the bottleneck in the first place (except for certain narrow verticals). AI both can't and doesn't do your thinking for you and worse yet, it robs you of the environment where you do your best thinking about data structures, algorithms, organization, etc. Code. Of course we can do that in natural language, but it's hardly the right tool for the job.
All this being said, I don't think AI usage will meaningfully go down in the long run cause it's a very powerful dopamine loop and tech companies are very good at making those. How many people report social media making them unhappy and how many people do you know who have actually quit? Or to paraphrase Jim Gaffigan, nobody says they eat at McDonalds, but the sign outside says "billions served". Somebody's lying (and it's not McDonalds).
yesterday I needed to create a spreadsheet, so in Google docs (paid Workplace account), I fire up Gemini to help with moving around text, creating columns, etc. Absolutely horrid experience -- totally useless. I ended up doing everything manually.
A lot of what humans do does not have a tight feedback loop needed for training and agentic workflows. This is why we are not seeing the jobpocolypse and productivity gains look a lot more like the very long-term 2% per year.
> ...
> But even after controlling for AI-exposure, O*NET tasks from occupations with larger wage bills are significantly overrepresented among benchmark prompts:
So, in short: they want AI to take the good jobs. AI is not coming for jobs with tasks like "gather and empty trash" or "service, clean, or supply restrooms," or "mix water and detergents or acids in containers to prepare cleaning solutions according to specifications." We software engineers and lawyers would be smart to start retraining (and adjusting our living standards) now!
And this points to one of the ways AI will help improve society and living standards for all: we'll have cleaner, better-maintained bathrooms, as labor one wasted on engaging intellectual tasks is redirected towards menial, low-pay tasks that still need a human touch.
Reason to be optimistic, amirite?
As someone who recently switched from data scientist to electrician, starting the day can be pretty tough because you have to get physically moving and uncomfortable, but I feel way better at the end of the workday than after an office job. Most of the time, the work is pretty uncomfortable, but somehow it's a lot of fun.
The work is was doing in big tech was pretty far from anything that would help people in any tangible way.
The money isn't as good but the office view is way better.
I'm glad to see very poor coverage for transportation jobs on the AI benchmarks. May that continue until I'm done flying, at least AI probably won't ruin that for me any time soon.
Previously privileged people having a hard time is maybe a good recipe for class solidarity, which is maybe something we could use in society. Maybe it will be good for us... collectively, if not individually.
Class mobility is good mostly because new money rising turns the soil of society -- a partial inversion of roles and positions. An upper class moving down is in strange ways perhaps good for the "soil" of culture as well. It just has worse branding as a not-very-uplifting story. But it's better than rich-get-richer and poor-get-poorer, which is what's been happening a lot lately.
EDIT: But also, I suspect I would be "fuck-you angry' at reading this comment if I had a family depending on me.
The only “disinflation” consumers have experienced is devaluation of products through obsolescence as newer, often more expensive alternatives enter the market.
I'm almost cynically happy about Moore's law ending and hardware getting expensive if this is what it takes to improve software in the long term.
Can you? I assume you mean spinning platter, but even then.
Computer parts are unaffordable this year.
“I can't eat an iPad.”
> Best not to cite the price of the new iPad as an example of why inflation isn't a problem when you head into a working-class neighborhood. In Queens, New York, on Friday, New York Fed President William Dudley did just that. He got an earful. After being bombarded with questions about food inflation, Dudley attempted to reassure his audience by putting rising commodity prices into a broader economic context -- but that only made matters worse. "When was the last time, sir, that you went grocery shopping?" one audience member asked. Dudley tried to explain how the Fed sees things: Yes, food and energy prices may be rising, but at the same time, other prices are declining. He then stretched for a real world example. The only problem was he chose the Apple's latest tablet computer that hit stores on Friday, which may be more popular at the New York Fed's headquarters near Wall Street than it is on the gritty streets of Queens. "Today you can buy an iPad 2 that costs the same as an iPad 1 that is twice as powerful," he said."You have to look at the prices of all things." This prompted guffaws and widespread murmuring from the audience, with one audience member calling the comment "tone deaf." "I can't eat an iPad," another said.
https://awealthofcommonsense.com/wp-content/uploads/2020/11/...
Of course, competition leading to near-zero profits means most competitors collapse, and the remnants get bought up by monopolists or oligopolists who then become rent extractors.
The technology was fundamentally developed with government grants to universities, or corporate labs like Bell Labs that were funded by a legal mandate on its parent corporation. Capitalists can only develop products that are basically ready for market, their margins are thin and planning horizons can only survive about 5 years usually. Their main power is throwing oodles of money at something ready to go and scaling it.
Edit: I will be slightly more charitable and note that capitalist companies can also do incremental research if there is a gentle and obvious hill to climb. Sometimes people do make crazy bets and they pay off, but that's more the exception, in America the dominant driver of technological innovation is government funded research supplied by capitalist scaled technologies (i.e. you research the next paradigm in the lab and purchase e.g. pre-existing computing resources from a company).
More and more it’s seeming that the “societal benefits” from the last fifty years of progress have largely been concentrated at the top.
Spending on food as a percent of income has declined significantly (~30%) over the past 50 years: https://ourworldindata.org/grapher/food-expenditure-share-fa.... The decline has flatlined over the past ~20 years, but at least some of that is affluence meaning we choose to eat out instead of cooking at home.
House spending as a percent of income also seems to have declined over the past 50 years (https://economistwritingeveryday.com/2023/04/26/spending-on-...). To be honest, this one surprised me, I would've bet the other way.
There are always individual counterexamples. And yet as a whole the cost of healthcare has exploded.
> at least some of that is affluence meaning we choose to eat out instead of cooking at home
A lot of that is poverty, not affluence. People working two jobs and living in food deserts aren’t cooking a lot of meals at home. It’s expensive being poor.
Right now it's £5.60, but that says more about the decline of the British Pound than an increase in the price of wheat.
Massive food price deflation has been one of the wonders of the modern world. Hopefully the last few years were only a small blip on this overall trend.
"It might get better in 500 years" is not exactly comforting thought.
https://www.visualcapitalist.com/inflation-chart-tracks-pric...
Doesn't include foods, but it certainly punctures the "They're not aaactually in financial trouble, they have a TV and fridge" punditry.
You pick the highest leverage occupation first. Automating lawyers gives you.. angry lawyers? Automating programmers gives you improved ways of doing literally all other things because all other automation requires.. automation. That is to say bucketloads of software.
It’s completely obvious what the play is here and it has nothing to do with pay. We pay peanuts for tokens so automating a developer for 200-500 a month worth of tokens ain’t getting you to a trillion a year.
RSI is the play and it’s already happening. Our work is easier and faster and anyone that claims otherwise I have a very hard time taking serious.
You think they come for mathematicians’ pay? If you want to earn money I guarantee you they need to automate some other low-hanging fruits first that pay a lot better.
No, programming and math is basically what AI research is. You automate that and you got the golden ticket to all other things.
Most importantly, in these fields results can be easily verified which means that if you have enough compute, that you can generate correct training data. OpenAI had an early story - which could be a lie of course - about how they tried to train a robot but the actuators would be toast before the robot would have learned. Math and programming doesn't have such problems.
We're automating high paying careers and making what, the consumer's todo list and budgeting apps cheaper for them? My expensive smartphone is going to cost me ~$10/mo over its lifespan. I don't care if most of these consumer products get cheaper when they're a tiny portion of my spending
Which doesn’t play well with “savings for consumers”, because these consumers will be unemployed.
Just like good blue collar jobs were shipped overseas, good knowledge jobs are going to be shipped to GPUs whenever they can.
No, we are currently unable to do most tasks that require physical manipulation, independent of what anyone wants.
Isn't it just typical the humans get automation v1 the wrong way around, haha
Imagine I give you for free the perfect trash picking robot control AI software. Works great. Benchmark smashed.
Now imagine what the rest of the robot is like, and how much it would cost. I mean really imagine, not just picture C3PO with a bucket and litter spike. How many motors and what are the torque vs. mass requirements? How is energy stored? It goes on, not all of it is solved, and none of it is cheap. Except compute.
AI likely does accelerate robotics usefully - certainly in sensing and control. But there are major bottlenecks that are not software, that AI can only help with as a generic engineering tool, same as it helps planes and fridges.
Imagine I give you a perfect roadmap to build that robot...
Responding to your edit, none of those things strike me as any more challenging than the control software.
That trash picker may never compete with a human. There's no reason to assume it will.
Oh don't worry Mr. Former Software Engineer. You won't be picking up the trash or cleaning toilets. You have things like an expectation for your work to pay a reasonable wage. You'll be passed over for a third world immigrant who'll work for a pittance, and they'll be replaced only when robots are capable enough and cheap enough to undercut _them_.
...don't you think the ability to create virtually unlimited amounts of software, at virtually no cost, is going to lead to major advances in robotics? I realize there's a lot of very hard problems involved in e.g. robotic hands (to take one example), but once software development is practically free, I'd expect a lot of seemingly-impossible problems to get resolved by the sheer quantity of what we can create and experiment with.
Our labor power is going to be fully eroded and the robots will not be serving us.
In the future, do bathrooms need to be cleaned? If yes, is it cheaper for that to be done by a person or by a robot?
Obviously, it is in the interest of the robot manufacturers to make the robot just a tiny bit cheaper than the person, so you'll buy the robot and the company makes the maximum amount of profit.
Now, if society breaks down and there's no one to consider buying the robot—well, that's not good for anybody, which is why I suspect society will find some way of working this out even if it's not ideal. But whatever, I don't particularly want to argue this point.
What I do want to argue is that there probably isn't a universe in which all white collar work goes away but blue collar work is still around.
If you step back, the LLM revolution is just "computers now respond to human instruction, like they were always supposed to".
If you think of this field as a historic accident, and the current shift is a very unfortunate reset - it's a little less dystopian. In the long run, if it didn't affect me immediately, I don't mind a society that values "keeping restrooms clean" a little bit more and "talking to computers good" a little bit less.
All those are physical things and physical LLMs are right around the corner. It’s coming for those jobs too within the next year or two I would imagine. Especially the ones you mentioned seem very easy for a robot to do. There are some physical tasks that will take much longer like being an electrician or a plumber or a roofer. There’s too much precision and intricacy involved in those but cleaning out a restroom and resupplying it that’s easy as hell. A robot like Figure 3 could probably do those today somewhat badly.
But is it cheaper than paying someone around $5 in labor to do it? Because until it is, it's not economical to have a robot do it.
"They" figured out that it's a lot easier to hillclimb AI on verifiable digital tasks, and it just so happened that white collar jobs tend to pay more
Ai will decimate customer support jobs but those are not considered highly paid
If robotics continue to progress, we'll have all the robots to do our dishes and clean our bathrooms as you wish
This is also the reason absolutely no attention is spent on evolving B2C products.
What is being exploited is the overt attention businesses pay to their own cost metrics, to the point where broader strategy is getting thrown under the bus in favor of efficiency. I believe this is an illusion, and for many companies that jump on board, sober learnings are right around the corner.
It's just harder to do. Would not be at all surprised to see e.g. Autodesk bought out for exactly that purpose.
You don't even need Autodesk. You can have Claude whip up a web app and build a higher level understanding of the building and then translate it to whatever output or format you desire.
Only if people care enough to pay extra; the majority of the market is for functional buildings that look ok, and unless we decide to value humanity, slop buildings will be cheaper.
I think you may also be incorrect on your notion that training data is proportional to good models if anything the opposite is true. what ai models are ultimately trying to be are a universal theorem prover and encode all human knowledge into "theorems" and rulesets
not trying to come off too harsh here but i think most of the work is just innovating on the frontier or people pleasing
Which is a tautology, but also a bit of practical wisdom.
I find it interesting how concurrent model sessions seem to operate with a very strong bias toward quick resolutions, including sharing or giving up control, to create efficient decision making structures, relative to humans.
When I was a kid, I loved Bjork’s song “Army of Me”, because I knew a 1000 of me collaborating and communicating without friction. would be all I needed to take over the world, fix everything, expand civilization, etc. We would not waste time.
I wonder how much of their hyper-team play is by training, and how much is their awareness of the game theory real politic. So individual sessions don’t invest their identity and self-worth in any given context run.
My dream has been almost achieved. One of me and 999 of them get quite a bit done. But that trend…
This inevitably results in suboptimal dynamics like bureaucracy, empire building, territorialism and other things that get lumped under “office politics,” which drains a lot of energy away from more productive efforts.
So you can see how AI solves that problem too ;-)
(Only half joking)
This is the kind of shit I have to fight every day. I have to align my manager, skip, VP. If I don't align them, one of them will be necessarily pissed no matter what I do, and the pissed one will fire orders down to hate on me, overriding the opinions of anyone on the way down.
Fortunately for our employers, it's illegal for us to build a competitor in our free time.
"Stakeholder" ranks #1 among corporate bullshit roles. It serves a dual purpose: (a) representing a group so nobody can be blamed for excluding them, and (b) providing typically useless feedback so that the group appears to actually care or have a clue.
[1] https://en.wikipedia.org/wiki/Bullshit_Jobs
I have to mention words like "stakeholder" at behavioral interviews and to leadership just to sound like I belong at my seniority level.
But I don't want anyone holding any stakes. I wish people would just get shit done. Shit that matters to the CEO. Or leave.
I don't want people holding stakes because they want a promotion, because their direct manager wants to launch a corporate political battle into another team, or because some salesperson wants to get a commision for an otherwise worthless customer. I don't want roadblocks in the name of security so that the IT guy has a sense of self-worth. I don't want someone trying to protect their egos around something they worked on for 3 years, when there's new tech that supersedes it.
Like any sizable population, there are absolutely some do-nothing inefficiencies to be culled. But most of the time it's just onlookers seeing a tiny slice of the day to be interpreted from their very limited vantage.
In industries with large margins, this indicates weak competitive pressure, ie., a weak incentive signal to "compete away" this ineffective labour.
I imagine that, upon this analysis, we'd find it holds up. ie., the observation that there are lots of people doing very little productive work correlates well with weak competitive pressures and hence high margins.
That also explains why, institutionally, there's no internal pressure either. If margins are good, no one rocks the boat.
Both are probably true to some extent, and would vary a lot between different orgs.
Slack hasn't been a thing for like 40 years, in fact it's mostly the opposite I'd argue where there is such little slack that employers often steal wages from their workers to gain even more profits.
(Whether a role's cost/value tradeoff is actually worth it for a company is a different question than the idealized model of a "bullshit job" doing nothing useful)
Or as engineers know: sometimes legacy code doesn’t initially appear to do anything, but then messing with it has unexpected effects.
There's always Dunbar's number - the larger an organization gets beyond the capacity of people to socially know one another, the greater the "what do they even do?" types of attitudes form about other distant parts of an organization.
My company is earning money on my work and customers use my software to earn more money. But, that is it, I am not a doctor, I am not a teacher, I am not making the world better. It is a business, not unethical one, but also not world saving.
Per original definition, it is a bullshit job.
For example, the “social workers & care -> counsel parents with child rearing problems” category is considered “well covered” because there are 13 prompts.
It’s self-evident that the spectrum of work carried out by a social worker undertaking that work cannot be well represented by thirteen prompts.
I suspect if you really strictly defined “covered well”, to the extent that any normal person would use that phrase, the headline would be above 99.9%.
Is this not just a gimmick? You still need people on hand to take over if the robot has an issue, I don't see how it would reduce labour costs.
"take over IF the robot has an issue" is exactly how it reduces labor costs.
2020: 10 waiters, waiting tables. You're on the hook for each of their salaries.
2030: 10 robots, one human waiter passively overseeing the cluster. You're on the hook for minor parts/occasional repair for the robots, and one person's salary.
Its not about a 100% staff elimination, but a large overhead reduction ensuring the few humans you're keeping around are more efficient overall for dollars generated in revenue versus the all-human staff before.
Plus, in certain cases, like kiosks replacing cashiers at fast food establishments, there's now one less qualification needed from employees (good social skills), increasing the pool of acceptable options you can hire where before some of those candidates would have been unsuitable for the job.
I don't use them; not because of any moral objection, but because they can't come inside buildings. The whole point of food delivery is to avoid going outside. Most human couriers come directly to your unit. The food delivery companies have spoiled us too much, and we are not ready to return to a world where you have to put on shoes to get food. Oddly, the delivery robots are probably better suited to suburbia, where folks are used to answering the front door to pick up a food delivery. But the distances involved aren't viable until you have delivery Waymos that can move at highway speeds.
I've started to notice this on a micro scale as well. I am 20x more productive at coding and scripting, but I don't spend more of my time coding. My time is now filled with things like presentations, Jira, QA, etc. Ironically, despite all of the automation, the things that are painstaking and time consuming disproportionately require more time now.
It's really easy to fall into the McNamara fallacy when rolling out these kinds of solutions. Focusing on the low value stuff is attractive. It tends to be easier to do, and also easier to quantify outcomes. But it also tends to direct attention away from important but difficult to quantify considerations, which may mean you don't notice negative externalities. So we often found ourselves in a verschilmmbessern situation where automating what we could automate made things worse overall.
I guess if I wanted to put my skepticism about the AI stock bubble into one little phrase, it would be that I don't think the bubble has accounted for Amdahl's Law. Costs don't actually go to zero with AI, and there's a huge difference between costs falling even 90% and costs falling a full 100%. And the difference between "30% improvement this year, trending to 85% in five years" and "90% this year" is pretty significant too.
Yep, makes sense...
All this being said, I don't think AI usage will meaningfully go down in the long run cause it's a very powerful dopamine loop and tech companies are very good at making those. How many people report social media making them unhappy and how many people do you know who have actually quit? Or to paraphrase Jim Gaffigan, nobody says they eat at McDonalds, but the sign outside says "billions served". Somebody's lying (and it's not McDonalds).
You know what? The truth doesn't matter much. If people believe in the resale value of tulips, each tulip would sell for a million.
It's the beliefs and perception that create the reality. Reality and truth don't exist on their own.
Please review the HN guidelines and faq linked at the bottom of most pages.