Hacker News Reader

Comment by T-A | Hacker News Reader

> GPT4 gave the correct answer in 1/10th the time it would take Priya and cost a lot less. [...] Priya has tried to learn coding, but she has struggled. Not that she gets a lot of time to do it either.

1) Show Priya how to use GPT4 to do her current job in 1/10th of the time.

2) Let her use part (50%+) of the time saved to learn coding or whatever other skill she needs (not sure coding is it).

View on HN

Why? No really: why? Businesses are profit-seeking entities. Why not just fire nine people? You save money.

(Yes, this could be a really stupid decision for the company in question. I agree with you. I'm not saying what should happen. I'm saying what will happen, and that that is really bad.)

All small businesses are not pure profit maximizers. Anyway even if they were, there’s MORE money to be made by empowering employees to use powerful tools which can increase revenues and profits much more than just firing. Doing so generates goodwill which improves employee effectiveness and drive.

Small businesses get eaten. That's the entire point of modern venture capital, even! Find ways to out-operationalize small businesses in sub-optimally exploited markets while burning an ocean of cash, then seek profits when you no longer have to dump to be competitive.

It's okay to not think every technology under the sun is a good thing. You don't have to have heart-eyes emojis for the immiseration machine.

Sounds like a good idea to embrace new tech, then, instead of get eaten.

Cool, so we're back to "streamlining" and mass unemployment just in time to get eaten anyway.

Once more for emphasis: you do not have to cheer on the immiseration machine.

No, we’re not. Small businesses embracing technology to get more done with the same number of people is not “immiseration.”

And I take it from your use of Marxist terminology that you’re coming from that the perspective. 20th Century Marxists embraced technological advancement, and used that to advance the cause of the proletariat. The REAL immiseration is to embrace paleoconservative ludditism.

Most small businesses skimp by on very small margins - often run by "crafters" or "artistans" who generally couldn't tell you much about margins besides "cash in the bank"

I believe AGI can help small businesses have higher margins for the same service level - but lower cost.

Goodwill doesn’t pay the bills, most businesses don’t care, and most CEOs etc cannot acquire more customers and contracts fast enough to multiply the available work by ten in the time frame required.

Yeah, the idea that there is somehow Unlimited Demand to tap is just woefully underexposed to business realities. Demand at most businesses is a river, not a municipal waterline with a spigot at one end. It ebbs and it flows, and building for a hypothetical high-demand future is what you do when you have venture capital expecting it of you--not when you're building a "small business" (or even a medium to large one, if you want to be resilient).

It is, however, a really convenient excuse for people who don't regularly think about how people who don't work "above the API" actually exist in the world.

Seems most people miss this. If anything businesses will increase their profits instead

This! If you can 10X an employee, the obvious play is to retain all ten employees and get 10X widget output. Yes, you could fire nine people to retain 1X output—but your competition also has access to LLMs, and may choose the former option. Can you compete?

> If you can 10X an employee, the obvious play is to retain all ten employees and get 10X widget output.

Have you ever run a business that employed other people and relied on operational income rather than venture capital to function?

Because this "obvious play" is a screaming red flag for misunderstanding the fundamentals of how business works when money isn't free and speculative.

No, but I have been an employee in a business that had layoffs, then had reduced capacity, and had to scramble to hire new employees.

You can't be too reactive with talent. If you hired 10 people, you should be figuring out how to make that talent pool as happy and productive as possible, not firing them out of fear at the earliest opportunity.

Said business has to hire employees because they can't replace them with tools and AI.

When said business can replace the employees with tools and AI, they will.

"Replace" is very different from "Enhance" or "Augment." There are certainly roles that could be 100% replaced, but I think they'll be the exception, not the rule. Guess we'll find out!

This assumes that there is a need for 10x amount whichever service these entities are providing. For some cases that may be true, but for many (most?) it is not. In fact demand will most likely remain about the same, and all competitors will need to adpot LLMs and fire 9 people to remain afloat.

There is cutting costs and scaling. You need 9 other people for scaling. Imaging if you with GPT4 can replace 10 people with 1. What if you keep other 9 and train them to prompt on GPT? Now you have not 10 people working but 100 and a cost of 10.

I guess the follow-up question is: if you do value these 100 units of work (1 employee == 1 unit pre-GPT), why did you only have 10 workers at the beginning? Doesn't it mean that 10 units of work is what you thought was optimal? What are you going to do with 100?

Or maybe you're implying that reducing the costs changes this equation?

It also implies the market can handle, and needs, the scaling. In some industries, like say food services, the problem isn’t lack of people.

Because information curated by an AI with validation by a subject matter expert is worth significantly more than that which was done without a human in the loop. There's no determinism with these systems, and from experience with GPT-4 doing classification tasks, it'll do everything exactly right until it doesn't.

If it's like many businesses in the classification world, they aren't lacking for work, so it's more like they'll be able to do 10x the work done by keeping the same number of people.

90% quality for 10% the price means that most decisionmakers will--and they don't need to collude to do this--take that and pocket the difference.

Why would 50% for 0.1% be different? 30% for 0.01%?

This is the enshittification spiral to which I refer.

Not if the output quality matters - which in medical and biological classification tasks, quality labeling really really does matter. Garbage in, garbage out.

If I was paying someone for labeling or classification, and the quality dropped to 90%, 50%, 30% accuracy, I'd quickly fire them.

What's your idea of "subject matter expert"? The term is used in court. Mine would probably include no more than double the number of people with doctorate degrees. It wouldn't include Priya right now, maybe after more education and/or a few years of exceptional experience.

> 1) Show Priya how to use GPT4 to do her current job in 1/10th of the time.

Why would I subsidize that as a business owner? I can fire 9/10ths of the people who do that job instead.

Train people to use GPT-related tech to become superhuman team members. You can do it all yourself, sure, but that doesn't scale. You can either train GPT-tech to use GPT to scale or train humans. Right now, reliability is the most difficult issue, so humans are essentially to ensuring the results are correct.

Humans are useful for training future iterations of the closed source thinking model. Once the model is sophisticated enough to train itself humans will no longer be required.

> Train people to use GPT-related tech to become superhuman team members.

then OpenAI either discontinue or slightly adjust the model such that it doesn't work now

there goes your business

Historically, in these situations your business is either about to become obsolete or is about to have much more competition. You could hire people to do the new tasks required to thrive, but it’s easier to retrain people you already know and who want to work for you.

It actually much easier to hire people who already have the skillset and pay them 5-10% more than they are making because of salary compression and inversion.

Let's be realistic here though: a big chunk of that is that there are plenty of people who decide they'll simply ignore the new direction, or they're too old to adapt with the times, and curmudgeon about it until someone's hand is forced.

That’s only true during stable times or when more mature markets are casting off that type of employee. Or, I suppose, if a company is incapable of understanding employees’ aptitudes behind their skills.

It’s also possible, like a lot of skilled knowledge work, they’re greatly constrained on talent relative to potential work. I think something folks don’t get is that AI is going to allow a lot more of everything to be done, and will still need humans to guide it. I suspect it’ll be a bit like the calculator and it’s acceleration of math fields. Human calculators became less useful, but humans could do much more math and I’d argue power tools don’t eliminate jobs, they allow the people that exist do 10x the work allowing 10x to be achieved with the same people. I think it’s cynical to think a rising tide sinks all boats.

I don't think anyone at all suggests that all boats will sink.

But the poorest and most vulnerable will. It happens every time. This time, we could do better--and we won't.

Is that true? The last few centuries have shown a near continuous improvement in the standard of living, particularly at the lower end of the distribution.

I think your windows are larger than mine. The Industrial Revolution did not improve standard of living for the people crushed in the gears, both metaphorical and physical. Remember: it was not until the modern age, almost but not quite within living memory, where cities were not a sink for population. Cities were places where normal people died and died ugly for most of human history, and much of the migration to them has historically been out of desperation.

Eventually, improvements through both technological means but also political ones (worker action says hi) did make things on-net better, but we live in the now and the now is going to simply vaporize an impossibly large set of jobs, particularly in developing countries. Turning one person into an LLM driver to lay off four or nine is not a net benefit.

As just one example: with modern text-to-speech, speech-to-text, and AI parsing, how much of every call center can just go away? Eight of ten? Nine? I have no idea, but it's a lot, and we have neither time nor inclination to prepare, globally or locally, for this.

(edit to add: the most appalling part of this oncoming train, as I have mentioned elsewhere in this thread, is just how shitty a future this one wants so desperately to be. A LLM-driven chatbot or phone system doesn't get tired. It doesn't get "too expensive" to continue to obstruct you and to stiff you. Not only is this primed to seriously hurt people who are below the API, but it's going to make the world suck more for the rest of us, too. Like, sure, "it makes writing code marginally easier"--it's going to make getting a refund for a messed-up Comcast bill an exercise in pain. That doesn't remotely net out, code's already easy enough.)

"The Industrial Revolution did not improve standard of living for the people crushed in the gears, both metaphorical and physical."

Can you expand on that?

My impression is that life for the less well-off before the Industrial Revolution was worse.

Depends. How do you measure it and how much further back? Because, in the short term and in some ways, yeah, it was, insofar as small agriculture was kinda collapsing and large agriculture was starting to consolidate. A big driver of the move to cities in the Industrial Revolution was both climate and political/economic forces making even subsistence farming much harder to do than before, which incentivized urban migration. (The political/economic drivers of urbanization are as old as civilization; Roman indigents flocked to Rome because farming was hard and getting harder, when their lands weren't being seized and they weren't being outright turfed out. Egyptian grain becoming the dole of Rome attracted people without any other options. Again--cities as population sinks.)

So in some ways, and in the short term, the Industrial Revolution was better than just-plain-starving, sure. But most of the benefits of the Industrial Revolution accrued to everyone else and the mangled limbs accrued to the poor. It wasn't without significant worker action (and the requisite workers-getting-beat-to-shit-and-killed) that their lot improved materially.

Is subsistence or subsistence+ farming hard? Absolutely. Mind-bogglingly so. But the Industrial Revolution was fucking bad for the people caught at the bottom. Like--read Dickens.

I tend to think we have such global largesse that we could do better. But we won't, and a lot of the commentariat here cheers for never doing better.

> Cities were places where normal people died and died ugly for most of human history, and much of the migration to them has historically been out of desperation.

I think that captures some of the contradiction in your claims. Yes, cities could be terrible places to live and work. Yet people have moved to them throughout recent history because the alternative is worse.

That said, I don’t disagree with your point that there will be pain associated with this technological jump. I dont know of there will be more or less than with previous jumps. There are some interesting considerations.

One, this one hits knowledge workers who are in the middle class, instead of hitting those who use muscle or hand labor. That may change the outcome.

Two, governments are much more sophisticated and have much better policy tools to deal with disruption. While that doesn’t fix the root, it can help to prevent compounding problems and soften the impact.

Three, the tools that are doing the disrupting appear to be near zero marginal cost, which is different than say a factory which improved efficiency with a large up-front capital investment. This factor probably will make it worse, but I can see possibilities of it making this change less painful too.

Fourth, it isn’t really clear how this will play out. It kinda feels like we have seen the first demonstrations of a steam engine, and are trying to predict the course of the Industrial Revolution.

> I think that captures some of the contradiction in your claims. Yes, cities could be terrible places to live and work. Yet people have moved to them throughout recent history because the alternative is worse.

I think this is a misreading. People who had options didn't urbanize until they had to. When a family had too many sons to split and or when Roman aristocrats or English magnates pushed people off their land or when a bad climate situation made farming impractical, people moved--but it's a very, very recent historical development to urbanize (en masse) when other choices exist.

Yes, the alternatives have been worse and so industrial urbanization became more appealing than starving. Who the hell made them worse and whose progenitors now control the capital needed to destroy ever more labor?

(Don't take not addressing the rest as dismissal--your other points are all within a coherent universe, they're just techno-optimism that I have no reason to share so I have nothing to say to them.)

Based on this and other posts, I think you are saying not that technological progress is bad, but that we as a society fail to take full advantage of the opportunity that could come from the advances.

And of course you are right. But that’s a societal problem, not a technology problem. I would call if societal-optimism to hope that human nature will go away and collective action problems will suddenly disappear, and all boats will be lifted equally.

Nobody is asking for all boats to be lifted equally.

I’m saying we have a fundamental human responsibility not to burn them for fucking funsies.

This is entirely untrue. People left their share cropping farms to work in the factories precisely because it improved their standard of living. Just because it isn’t an improvement over yours doesn’t make it an improvement over theirs. Factory owners paid enough to incentivize people to work there, just like in any other capitalist structure. AFAIK there were no forced labor conditions in industrial England.

You missed the union wars to ensure that we didn't fall back into a techno-feudalism. I'm sure we're all looking forward for the Amazon Wars with a billion percent more killer drones to put down worker rights.

I once read a shocking statistic that children of the poor in London during the Industrial Revolution were 6 inches shorter than children of the rich. What raised the standard of living was bloody, persistent labor revolt. It wasn't pretty and as your comment shows it has been mostly forgotten.

And yet families flocked to work in factories because conditions they left were that much worse. Everything is relative, and they can’t judge their situation from yours - they had to choose based on their situation.

> It’s also possible, like a lot of skilled knowledge work, they’re greatly constrained on talent relative to potential work.

Yep. If I have 10 people doing Job X, then ChatGPT makes a person doing Job X ten times more productive, I can fire 9 people doing job X and retain the same output, which gives me a temporary profit boost. But wait! My competitor also had 10 people doing Job X, and now they're producing ten times the output of my company!

If ChatGPT/Copilot X made developers working at my company twice as productive, I wouldn't be thinking about firing half of them, I would be thinking about which of the 8000 features and improvements on our backlog we could finally prioritize...

So workers are 10x more productive! Are there not enough papers to annotate? The writer still needs a specialist to make sure the LLM is going the right thing.

Aren’t the comparisons to blue collar labor a little much here?

If I hire 10 people to dig the ditches, and then I get a backhoe, yes, one guy is needed to dig the ditch.

But backhoes are expensive investments, LLMs are not.

LLMs gives everyone of those ditch diggers a backhoe. Now I have an army of backhoes! Unless the world runs out of a need for ditches, time to start a ditch digging empire.

Interesting. Who's going to buy products and services? 1/10th of the remaining employees? What would happen with to the 9/10 unemployed people? Imagine a really big chunk of the human population trying to digest the idea of not being able to work, eat, feed their families, etc...

The industrial revolution switched work from almost exclusively manual labor to a more intellectual one while improving manual labor itself (better tools, etc.).

This one is cranking up a notch times 100 by trying to replace basically the human out of the equation altogether (This is the end goal of automation).

It's going to be a really interesting decade tbh.

You see the Catch-22 situation? Author is the manager of Priya.

This means Priya can replace the manager, but manager needs to train Priya for that. Obviously that is not going to happen as Author him/herself is very apprehensive about his/her future earning potential.

This would be the problem in the new world, people with lesser power (of knowledge) would get crushed.

Exactly, while there is some merit to the article's argument, I don't think it holds any water in this case. GPT4 will lie to you [0] and you need the human to figure out when that's the case.

[0] https://simonwillison.net/2023/Apr/7/chatgpt-lies/

> you need the human to figure out when that's the case

Do you? I mean, you and I think you do in order to do a good job, but does that hold water? The last decade or so of technological "progress" in the commercial space should leave few illusions about the willingness to obstruct processes and damage the interactions between people to save a few bucks.

As tech gets more capable, the human interactions with it are getting worse. And if you hate phone trees now (I do!), GPT4's contribution is adding ever deeper, infinitely patient tools for stalling you when you have to call for things.

Hell, we already have plenty of stans in the HN comment sections actively claiming that no, of course you shouldn't hold LLM purveyors responsible for their hallucinations, the desired universe is liability-free! If you've already bought into societal enshittification for a buck and there are no penalties, why would you worry about being right?

> you need the human to figure out when that's the case.

Yes, but you will need fewer people. And thas is with the current level of GPT. If the GPTs proficiency increases, fewer and fewer people will be needed. That’s a problem.

Try and reframe the same argument except with computers to see that it doesn't add up. When computers started becoming widely spread you could've easily made the same argument. It isn't a problem, it's a boon to make our lives easier.

More probable scenario:

1- Hire someone cheaper than Priya and show him to tricks to run GPT-4. Make him work 24/7 non-stop because it's an upgrade from his naan flipping job.

2- Fire Priya.

> Priya has tried to learn coding, but she has struggled.

If there's one thing that will get automated first, it's coding (taking simple explicit specifications and turning them into code).

Unless she can learn software engineering (understanding problem areas well enough you can devise a detailed spec, and maybe implement it) she'll be out of a job as well with just coding skills.

People who haven't dealt with cheap offshored labor or bootcamp grads from the latest craze don't realize how much that workflow was similar to working with GPT4. Except GPT4 requires no training and gives answers in seconds.

> Show Priya how to use GPT4 to do her current job in 1/10th of the time

And then if they have 10 people doing her job, 9 people are out of a job.

1) Competition is brutal. Many educated remain unemployed and the amount of work is not 10x. It also makes more sense to keep one person do 10 peoples job if there were more people doing it and avoid multiple people trying to co-ordinate.

2) Most if not all employers are not going to let 50% of the work time to go towards non income generating activities.

This is hopeless fantasy.

What skills, exactly, would those be?

This is the suggestion that everybody can just learn to code. I remember Code.org and #HourOfCode back in 2014. Still, it is hard both to find a senior software developer and to become one.

> 2) Let her use part (50%+) of the time saved to learn coding or whatever other skill she needs (not sure coding is it).

People are still pushing this meme on HN, even after GPT-4?

Reply on news.ycombinator.com