Regulations are exactly what we would have less of IMHO in so many areas nowadays. Talking about Europe, so if you are american, I do not have an informed opinion.
AI is clearly not the problem here. Our preexisting society and massive wealth inequality is. I'll add though that many, many open models of high quality that smaller companies can build on exist and these models and companies are likely to be the most harmed by over-the-top regulation. OpenAI will survive jumping through hoops just fine.
Is GenAI more like cigarettes ("cool factor", but harmful in the long run), or is it more like cars (noisy and creates pollution, but super-convenient and therefore something most people should adopt)?
I still struggle with this every day. Maybe I'll figure it out someday soon...
Those two problems are some of the largest we have at the moment. They screw everything up.
We should not optimize for a shitty existence.
I actually think making the problem worse is a good side-effect. The larger and more direct a problem is impacting us the greater impetus for fixing it.
I really wish our society was able to solve problems before that become enormous, but that just doesn't seem to be something that happens very often. :/
Absolutely. We're incapable of the foresight and coordination needed to stave off these disasters. It doesn't help that "progress" has become the idol of our secular religion. Slowing down or stopping is just sacrilegious to modernity.
I completely agree. I have many, many friends who are commercial illustrators, and many of them are very anti-AI now due to popularity of DALLE. They feel that they've been cheated out of a job. I'm an artist and a software engineer, though, so I'm in the same boat of having had my output used to train LLMs. But the outrage, at least in the conversations I've had, seems to always be rooted in a fear about economic insecurity if these models take their jobs. To me, this doesn't mean the issue is that we should expect our output to not be used to train models, but that we should expect our governments to support us in the event that large swathes of professionals find their jobs suddenly far less profitable.
That would require some kind of profit sharing agreement from the companies that benefit from replacing employees to redistribute to society, which frankly I cannot ever seen happening.
And professional musicians in a previous century felt cheated out of a job by the phonograph. Time and tech change. Such is life. Painters hated photography when it was invented. Jobs are distroy and new ones made by tech. No one is crying about the typest in typing pools loosing jobs to word processor. Its just artists turn this time.
You're not wrong, but the heartlessness of that sentiment is striking. Real people get hurt. That in the long run we'll adapt as a society doesn't mean we should just write off those whose lives are ruined in the now.
Yet we tell the coal miners to go take job retraining because photovoltaic are a better power source than burning carbon. It suck for people being replaced but is better for the whole. Should we subsidise buggy whip manufactures because they were obsoleted. No one is well served by trying to fight the tides.
> Yet we tell the coal miners to go take job retraining because photovoltaic are a better power source than burning carbon.
Do we?
Last I checked (after decades in the mining industry from driving haul paks to geophysical exploration to selling global mineral intelligence software to Standard & Poor) :
Coal mining skills easily transfer to Lithium mining (etc).
Fitters and turners still fit and turn, belt splicers still splice belts, industrial electricians still pull wire and lace looms, mechanics are still required and will easily adapt to EV heavy machines as they've been working on those since the 1970s (electric shovels, ship loaders, etc).
Could you outline some things that coal miners do that aren't applicable to rare earth mining?
That's addressing a different point than I was making. I'm not saying that we need to keep old professions around or that we shouldn't adapt to changing times.
I'm saying that speaking about the people who are harmed when things change as if they don't matter is heartless.
We need to have empathy for them, and to help them where we can, not to talk about them and treat them like obsolete machinery. They're people, not machines.
Whenever a company is gleeful for regulation this is where my mind goes. Do they see regulation as an extra moat? We have the cash and years head start. It's love government procurement. Companies love it because once they are in, they rarely have competition and can charge extra.
I find this a bold claim, considering I extract several hundreds of dollars a month for the cost of $25 a month for GPT-4 in both language learning and wading my way through tech stacks very different from my own.
It could be that my experience is very much an outlier, but even if only 1% of GPT-4's subscribers got the value of it that I do, that would imply it very much is a net positive.
If original writing is so valuable, people will pay the premium for it in cases where it matters.
The reality people don’t want to accept is that in many cases, original work doesn’t matter. There is no value in an original booking reminder for my dentist or most travel agency blogs on reasons to go to Thailand.
Most courses don’t even need original work. Is there really any value in having profs generate thousands of different explanations of the same concept for the same topic? I don’t think so, especially as most don’t put much effort in.
Thousands of explanations have been written on how to do an integral or how hydrogen bonds to oxygen. That’s not useful in most cases.
> Most courses don’t even need original work. Is there really any value in having profs generate thousands of different explanations of the same concept for the same topic? I don’t think so, especially as most don’t put much effort in.
I actually used ChatGPT to generate homework assignments when I was a professor last year. I of course double-checked the questions to make sure they made sense, and I would solve them myself to make sure that the solution was something that I could realistically expect a student to be able to figure out.
It was honestly great; making homework and exams was a huge time sink for me the semester before (the supervising professors didn't give me any of the materials like he was supposed to, so I had to improvise), I was teaching another class the next semester, so having ChatGPT give me assignments took a task that took me about three or four hours before to about 30 minutes.
Do the students need 100% of the assignments to be my completely unique perspective of Object Oriented Programming? I don't think so, a lot of this stuff is just practice, and they still got my perspective on things during the lectures.
> Is there really any value in having profs generate thousands of different explanations of the same concept for the same topic? I don’t think so, especially as most don’t put much effort in.
Would you be satisfied with a professor who is mindlessly reading back to you information from external course materials? I know I wasn't when I was in school. It was only the professors who made the effort to teach what _they_ knew who were worth learning from.
I mentioned in a sibling comment that I used ChatGPT to generate homework. I tried pretty hard to get it to generate questions that I thought would be fun and interesting, instead of kind of boring cookie-cutter stuff that was in every textbook (e.g. shapes and animals to represent classes and inheritance and interfaces...boring!). I was very happy with the results (my favorite was getting the students to implement a rough version of the Pokemon battle system in Java).
I was a bit of a weird professor since I had like 11 years of industry experience at the time, and had been a college dropout not long before I was a professor (I did have my bachelors and was pursuing a doctorate while teaching, don't worry), and I do think that I was able to put a unique spin on things and explain stuff from a more industrial perspective, but I honestly feel like AI helped me with that, instead of making the stuff boring by regurgitating the textbook.
Ever since the spinning loom, this claim has been made. It’s been made about the combine harvester, welding robots, computers, and now AI.
And it has never once been true as the improved productivity led to different jobs as human wants are pretty unlimited.
And the jobs eliminated of are jobs that today would be near impossible to fill, as the modern jobs are much better.
Next to nobody would take a job standing in a wheat field chopping it down with a knife on a stick. Yet people protested the combine to try and save them.
I don’t see why the programmer wouldn’t just be the AI’s overseer, in the same way a farm worker now drives a combine instead of slashing wheat by hand.
You posted basically the same thing I posted right before I did, and I completely agree with it.
It's not like we will ever get to a point where we say "all the potential work that could happen is done!". There's always more work to do, as human resources get freed up they get reallocated to other tasks.
It's sort of like Gustafson's Law [1]; it's easy to see all the work we have right now as automate-able and therefore assume "there won't be any jobs", but I think that's making an incorrect assumption that the number of jobs is a fixed size, and I just flatly think this is not true.
The problem is the jobs you imagine people moving to won't materialize because AI will do those too. In the past, efficiency increases lead to the realization of new modes of transactions becoming economically viable. The people made obsolete moved to entirely new jobs because the tech was narrowly applicable. AGI is fully general, there is no space of jobs that humans could do but AGI cannot do. This is the end game for human labor. Society is not ready, and probably never will be.
What makes you think that we're actually close to AGI? I'm actually asking, because it seems like a buzzword that the media throws around a lot of handwaves the details away and just acts like it's inevitable.
The latest thing I saw about it seemed to indicate that OpenAI was looking to build a dedicated nuclear power plant because AGI on current hardware might be possible with that much energy [1]. Even if they do crack AGI with that, it's not like its energy consumption will be justifiable.
This article[1] does a good job of laying out the argument from trends and arguing that taking the trends seriously suggests reaching AGI in a few years.
Admittedly, my own argument in favor of near-term AGI is much less sophisticated but is also largely based on extrapolating trends and noting the leaps in capabilities that can come from singular architectural innovations. The GPT family has been amazingly productive, and they result from largely a single architectural innovation. They have glaring deficiencies, but their failures are fairly well understood and overcoming them are basically engineering problems at this point. Whereas 10 years ago no one had any idea how to get from the SOTA at the time to AGI, I see a clear path to it from here.
Once AIs are significantly smarter than humans, working humans become a problem, like "working" toddlers would be a problem today. They are useless at best and worse than useless at worst. There will be no economic incentive to pay humans to work.
Well, there's no economic incentive to pay low-skill workers minimum wage. The incentive is violence! They voted saying you gotta, and you'll be fined (I guess that's economic), imprisoned, or killed if you refuse. :p
Currently you have to pay low skilled workers minimum wage only if you actually need the job done by some human. The law just says: either you don't employ a human at all, or you pay him at least minimum wage. This is compatible with mass unemployment. The alternative would be to force companies to hand paychecks to "employees" which don't do anything. Which seems like taxation+UBI by another name.
I don't really think AI training is "stealing", despite how much people seem to keep repeating that. You can't own a "style", and you don't get rights to everything inspired by your work, and you shouldn't be because that would be horrible. A lot of AI images that I've generated, as far as I can tell, are pretty much completely unique in the sense that, yes, they were trained on copyrighted data, they don't really look like anything that has ever existed before.
There's been plenty of cases where a big company is inspired by a small company's work, largely redid it, and no one cared, well before AI was a thing. "It! The Terror From Beyond Space" was basically copied by Alien twenty years later, and the latter is considered a masterpiece of cinema, and while most film-geeks know this about Alien, they still consider it unique enough in its own rights to be a great movie. I feel like, in many cases, the stuff generated by Stable Diffusion is transformative enough to be considered fair use. I guess it's a minority opinion, since a large part of the most recent writers strike that everyone seemed to get behind was complaining about AI.
As a rule, I think articles like this are just kind of fear-mongering. Every new piece of tech has a risk of displacing workers, especially in the short term. This isn't new, it's the nature of progress: people really want stuff to be cheaper so we invent stuff to make things cheaper.
Will AI "replace" jobs? Sure, but so did ditch-digging machines, cars, airplanes, and computers themselves; people find other stuff to do.
When AI becomes good enough to code web apps and other simple applications, the people who will be decrying AI will be all of the web devs out of work.
Notto mention, all of the creative and entertainment industry jobs that will be destroyed permanently.
It's coming whether people want to accept it or not and we're going to reach a point where new roles will not be ready to absorb the loss.
A widely observed problem with AI we’ve created so far is that it seems best at destroying jobs people tend to like (at least, better than other things they might be paid to do) and replacing them with a smaller set of less-pleasant jobs.
At least, within the realm of basically-positive jobs. There’s separately the issue of its being by far the most productivity-enhancing for things that are harmful (scams, astroturfing, spamming)
I mean, you're not entitled to a fun job, and moreover I think people will find other fun jobs. It sucks if you loved painting, and you're risking getting replaced by a robot, but a) I think Stable Diffusion will still need to be human assisted for quite awhile to get results that don't suck, and the pictures will almost certainly need to be touched up by people to be useful in a commercial context, and b) There's really no limit on "potentially fun work".
I agree that making scams easier is a real threat though, and I have absolutely no idea how the hell we're going to properly address that. Even if the main service providers block scam-like language, there's nothing stopping anyone from using an open model and running Llamas or something.
Who said anything about entitled? It’s about better and worse.
“We automated all the things people wished they could do if they didn’t need money—but basically none of the stuff you all actually wished we’d automate” is pretty clearly bleak as fuck.
That's just not true, we've automated lots of stuff that people don't like and wish we'd automate, well before AI was in the picture.
"Computer" used to be a job title, but that was automated by, you know, computers. We've automated digging ditches, planting crops, transportation, and many, many, many more things. I'd venture to say that most of the stuff we've automated has been stuff people don't really like doing.
Even with AI, there's plenty of stuff it's automating that no one really likes doing. For me, it makes bizarre server logs substantially more digestible, so to a point where I've debated writing a systemd service that listens on journalctl and creates a separate stream of easier-to-understand server logs for me to look at.
I don't really see it as "bleak", and I certainly don't view it as "worse", I think it's just different.
Most art isn't people painting stuff for the Met, it's corporate stuff. I don't really think it's terribly fun for people who do that (though I'm not sure, I don't have much experience in that realm), so if we can automate that isn't that something "you all wished we'd automate"?
I really try to keep my comments on this topic as neutral and not emotionally based as I can. But here's my 100% emotional, non-neutral reaction:
> I don't really see it as "bleak"
I'm trying not to see it that way, but it's really hard. Almost everything that the advocates of these systems say about them strikes me as very dystopic.
> I don't really think it's terribly fun for people who do that (though I'm not sure, I don't have much experience in that realm)
I know a few people who make their living doing that, and they really do enjoy it.
A: “Good news, we automated a ton of work. Millions of jobs no longer need to be done.”
P: “Oh, great, now that you automated boring jobs that nobody wants to do, those workers can switch to working in the arts just like they always wanted!”
A: “…”
P: “… you automated the boring work we all hate, right?”
I am confident that I will not be one of "web devs who decry AI because I'm out of work" [1], but even if I were, it doesn't actually change anything about what I said, even if you could prove some kind of hypocrisy. I was laid off from three jobs last year (not for AI replacement), so I can empathize how much it sucks to lose your job, but that doesn't mean we just halt progress, and even if I were pissed off that ChatGPT replaced me, it wouldn't really change anything I said.
> Notto mention, all of the creative and entertainment industry jobs that will be destroyed permanently.
Sorry, citation needed. Everyone just keeps asserting it like I can just take it as an axiom, and I don't buy it. AI is a tool, it will be treated as a tool, if a tool can do something better than a person then that tool should be used, and if it's worse than a person, then they probably won't be replaced.
If ChatGPT were to start writing code that didn't suck and as a result started replacing lower-level developers and eventually higher-level developers, that would of course suck, but those developers would probably just learn how to use the AI tools and move on with the trend, at least eventually.
People have been saying this crap my entire life (and probably way before it), and I just don't accept the axiom.
[1] I don't do web dev, but you can replace it with "backend data processing crap dev".
> those developers would probably just learn how to use the AI tools and move on with the trend, at least eventually.
Many certainly will. But that sort of job would be very different from what being a dev has entailed when they chose the career. For many devs, the part of programming that they enjoy the most is precisely the part that people are talking about automating away.
That means that there would be devs who find themselves trapped in a job that they now hate.
Sample size of one, but I've found AI to be quite good at automating away the parts of engineering that I really dislike.
Stuff like basic scaffolding and long repetitive if/elseif/else chains are things that I have zero fun doing, or looking at long debug logs from build systems makes my life considerably less irritating. It's not perfect at it, but it's also generally useful to paste a chunk of code I wrote years ago and telling me more or less what it does (because I forgot).
Obviously part of this is sort of predicated on the fact that it's still pretty bad at the fun stuff right now, and if it gets better at it then maybe I'll be right now, but in its current state I feel like it's made my life better.
> I've found AI to be quite good at automating away the parts of engineering that I really dislike
That doesn't surprise me. Different people enjoy different aspects of development, so YMMV.
> Stuff like basic scaffolding and long repetitive if/elseif/else chains are things that I have zero fun doing
I've seen other devs say similar things, and I agree that stuff sucks. But I only rarely have to do that sort of thing at all, so when I do, it's not a great burden. (That said, I actually do know at least one dev who enjoys that kind of thing.)
But it makes me wonder: am I an outlier here? Do others do this stuff often?
It happens less as I’ve gotten into more senior roles at my jobs, but I’m constantly starting new personal protects and I find that a lot of the annoying grunt work happens in the first few days of the project, which is the part I always dread. Being able to quickly ask ChatGPT for a quick couple stanzas of code for me to get started makes things more fun.
Also, I do need to look at server logs all the damn time and I never have fun doing it, especially for some of the more arcane weird stuff (getting an LTO6 drive to work on NixOS was especially irritating). A lot of the less-frequently-used services will often have logs that are really hard to parse. ChatGPT would be worth the price of admission based solely on that.
For that matter, build files are pretty much universally irritating to me. Being able to quickly get some personalized Maven or Gradle information really quick has saved me considerable time and effort.
There’s plenty of more bureaucratic parts of being a software engineer that it helps me with; I have gotten it to write letters of recommendation for me since those are fairly utilitarian at the end of the day, or to help me format slides in Beamer.
I don’t know; again for me it’s been extraordinarily helpful and made software engineering a lot more enjoyable. If I felt like my job were actually at risk from it in the near it’s possible I would have a worse opinion of it.
> I don’t know; again for me it’s been extraordinarily helpful and made software engineering a lot more enjoyable
I found it to be more of an impediment than help, in that I spent at least as much time and effort when using LLMs as assistants than not, and it makes the software engineering much less pleasant for me.
Different strokes and all that. As with all tools, what is great for one person can be terrible for another.
> But that sort of job would be very different from what being a dev has entailed when they chose the career.
I've been through that. I got into web dev when it was just HTML and maybe some JS (CSS didn't exist) and my interview was a half hour where I only had to demonstrate that I could use vi and create a C program that was basically just getenv("QUERY_STRING").
I miss the days of writing vanilla JS without a ton of dependencies.
I'm not a fan of the "framework of the month" nature of JS, and all the complex toolchains these days certainly has made the job less "fun", but the world isn't going to stop for me.
The only thing that doesn't change is incessant change.
You've written a lot but only touched on the smallest part that they're trying to automate:
> If ChatGPT were to start writing code that didn't suck and as a result started replacing lower-level developers and eventually higher-level developers, that would of course suck,
YES this is the future that they want and are pushing towards. AI will eventually write code and make entertainment that will surpass what coders can do. To think this isn't coming is naive.
First it's scaffolding and simple stuff. Next it's the harder engineering tasks you enjoy the most.
This is the future that is coming and the reason people have been saying this your entire life is because they're right. It's coming and it's coming NOW.
Don't believe me? Read up on what happened to graphic designers in China.
YOu didn't really read the rest of what I wrote. It would suck but then I'd figure out how to use the AI tools. Every job that has ever existed requires people to adapt.
I did and rather bluntly, your adaptation will be either accepting a new job doing something different or taking far less money to no money for your coding skills to continue doing what you're doing.
In short, coders are the modern day buggy whip manufacturers who are about to get BTFO by the up and coming AI/automobile. It's coming and your nativite hurts.
This is baseless fearmongering. This has been talked to death here and elsewhere. First this overestimates the capabilities of AI second it misses the historical precedent set by the creation of other GPTs throughout history. Yes some peoples lives were negatively influenced by the creation of the railroad or the printing press but that doesnt mean the downfall of society.
The problem of AI being concentrated in the hands of the powerful few is certainly a problem. Which is why its very clearly of the utmost importance that we stomp out AI doomerism now before they can convince congress to enact braindead laws limiting open source AI. Unfortunately Brandolini's law is in full effect and the meme of "The Terminator" and "The Matrix" are way too pervasive.
I‘m willing to bet money against the core statement of that article. Cut off date +20years. General terms to be negotiated.
I see this as an instance of jevrons law. AI will overall induce more demand for work because knowledge work can be leveraged with ai and thus it becomes more productive.
All capital-intensive technology will amplify preexisting wealth inequality. And that's never a net positive for society, because - well - the top 1% are only 1% of society. AI is just continuing the trend. But I don't think the damage so far is worse than advertising and social media combined.
That's my point basically. Since AI is owned by so few, all of the profits from it will flow to 0.01% of society or less and leave the rest of us with scraps.
And that's the entire point of companies like OpenAI.
Even that I call into question; a lot of image generation stuff has open models, and new models being pushed onto HuggingFace all the time, and if HuggingFace is too centralized for you, good news, it uses git-lfs and lots of other entities can run that as well for free.
As GPUs get better I think we will see a rise of a lot of local-AI. The companies that I think will benefit the most are Nvidia and to a lesser extent AMD (eventually). I would agree that that's too centralized, we should probably have more competitors in the GPU space, but I don't think OpenAI will be able to monopolize AI as a whole for forever.
I use AI and LLM's at my job every day and its made me at least 2x as efficient. Even if AI helped me complete all my tasks right away, there would still be 10 more things I need to do.
>It’s not to help society, but to sell AI workers at a fraction of the price of human workers to all the world’s companies, which would cause mass unemployment.
The cost of a model that is powerful enough to cause a dystopian level of unemployment is - as far as I can tell - considerably beyond the reaches of the private sector. The budgets of the wealthiest companies in history are being stretched to their limits, and they're able to churn out chatbot tech demos, which are admittedly fun, but aren't going to disrupt the entire workforce - or even meaningful percentages of it.
The people at the top of the industry are talking about multi-tens-of-trillion dollar training runs (not to mention the US grid requiring a massive overhaul with several new nuclear powerplants). They need those runs to achieve "it can perform as well as the average line-cook at McDonalds" level general intelligence. The private sector just isn't going to throw endless $trillions at this.
The most likely scenario for a generally intelligent model coming into existence, which can cause really dystopian levels of productivity replacement, is that it'll be funded by the public sector. The obvious social contract there will be that the taxpayer takes on risk in the near-term in exchange for long-term liberation from mandatory labour via UBI or a similar scheme.
People are tired of AI content, especially AI doomerism. Your article can be very thoughtful and even right, but enough people are tired of seeing similar headlines that they flag it off the front page.
ML is not a net positive for the society, but not because some actor steal the labor of some other human.
- it's not net positive because they are proprietary systems trainable only at giant scale so OR humans still have open universities and public research centers who train models on open datasets or it's the definitive proprietary software even in FLOSS form because the model is trained behind the closed door of some enterprises;
- it's not a net positive because it's used to make money in scenario where the current state of tech is simply way behind the minimum needed to be ready for production and as a result we get more and more dysfunctional services;
- it's not a net positive because to do really not very much it consume an immense amount of resources and we need these resources on other stuff in the present, short and mid term.
Automation of any work is not a threat is a target all humans have, who want to need doing more stuff that can be done by some automation? Who want to manually wash clothes when we can use a washing machine? The point is that automation MUST BE OPEN not only to avoid https://ckrybus.com/static/papers/Bainbridge_1983_Automatica... but more important to grant control of automation for formal owners not to 4 big power and all are then practically their slaves as Industrie 4.0 essentially imply.
A self-described "hardcore capitalist" realises that a small class of ultra wealthy people monopolise the labor of the masses to the detriment of the many? I can't tell if this is satire or not. God I miss n-gate.com
That's why Sam Altman has said he is in favor of an UBI. But that money would have to be collected from the AI company which owns the AI that does all the work. So only people from the country where this company pays taxes would receive this UBI. So US Americans in case of OpenAI.
94 comments
[ 3.4 ms ] story [ 173 ms ] threadIt seems like anything that will make this problem worse is a net negative for society
If the only way to make it a positive is to change everything underneath it, then...
I still struggle with this every day. Maybe I'll figure it out someday soon...
And there's a lot of people telling us we can't afford the time or money to replace the subfloor first
We should not optimize for a shitty existence.
I actually think making the problem worse is a good side-effect. The larger and more direct a problem is impacting us the greater impetus for fixing it.
This is that "tax" I hinted at in my article.
A little empathy is a good thing here.
Do we?
Last I checked (after decades in the mining industry from driving haul paks to geophysical exploration to selling global mineral intelligence software to Standard & Poor) :
Coal mining skills easily transfer to Lithium mining (etc).
Fitters and turners still fit and turn, belt splicers still splice belts, industrial electricians still pull wire and lace looms, mechanics are still required and will easily adapt to EV heavy machines as they've been working on those since the 1970s (electric shovels, ship loaders, etc).
Could you outline some things that coal miners do that aren't applicable to rare earth mining?
I'm saying that speaking about the people who are harmed when things change as if they don't matter is heartless.
We need to have empathy for them, and to help them where we can, not to talk about them and treat them like obsolete machinery. They're people, not machines.
I find this a bold claim, considering I extract several hundreds of dollars a month for the cost of $25 a month for GPT-4 in both language learning and wading my way through tech stacks very different from my own.
It could be that my experience is very much an outlier, but even if only 1% of GPT-4's subscribers got the value of it that I do, that would imply it very much is a net positive.
Energy consumption, for example. Or change in incentives for producing original writing. Etc etc.
I don’t think we fully comprehend the negatives yet (or, to be fair, the positives)
The reality people don’t want to accept is that in many cases, original work doesn’t matter. There is no value in an original booking reminder for my dentist or most travel agency blogs on reasons to go to Thailand.
Most courses don’t even need original work. Is there really any value in having profs generate thousands of different explanations of the same concept for the same topic? I don’t think so, especially as most don’t put much effort in.
Thousands of explanations have been written on how to do an integral or how hydrogen bonds to oxygen. That’s not useful in most cases.
I actually used ChatGPT to generate homework assignments when I was a professor last year. I of course double-checked the questions to make sure they made sense, and I would solve them myself to make sure that the solution was something that I could realistically expect a student to be able to figure out.
It was honestly great; making homework and exams was a huge time sink for me the semester before (the supervising professors didn't give me any of the materials like he was supposed to, so I had to improvise), I was teaching another class the next semester, so having ChatGPT give me assignments took a task that took me about three or four hours before to about 30 minutes.
Do the students need 100% of the assignments to be my completely unique perspective of Object Oriented Programming? I don't think so, a lot of this stuff is just practice, and they still got my perspective on things during the lectures.
Would you be satisfied with a professor who is mindlessly reading back to you information from external course materials? I know I wasn't when I was in school. It was only the professors who made the effort to teach what _they_ knew who were worth learning from.
I mentioned in a sibling comment that I used ChatGPT to generate homework. I tried pretty hard to get it to generate questions that I thought would be fun and interesting, instead of kind of boring cookie-cutter stuff that was in every textbook (e.g. shapes and animals to represent classes and inheritance and interfaces...boring!). I was very happy with the results (my favorite was getting the students to implement a rough version of the Pokemon battle system in Java).
I was a bit of a weird professor since I had like 11 years of industry experience at the time, and had been a college dropout not long before I was a professor (I did have my bachelors and was pursuing a doctorate while teaching, don't worry), and I do think that I was able to put a unique spin on things and explain stuff from a more industrial perspective, but I honestly feel like AI helped me with that, instead of making the stuff boring by regurgitating the textbook.
Energy consumption has led to all the prosperity humanity enjoys.
We should use more to do more, which we are doing with AI.
We need cleaner sources though. You know what will help incentivize more investment in clean energy??? More energy use.
That's an odd choice of words. You seem to be operating under the assumption that wealth extraction is a good thing for a society?
Ever since the spinning loom, this claim has been made. It’s been made about the combine harvester, welding robots, computers, and now AI.
And it has never once been true as the improved productivity led to different jobs as human wants are pretty unlimited.
And the jobs eliminated of are jobs that today would be near impossible to fill, as the modern jobs are much better.
Next to nobody would take a job standing in a wheat field chopping it down with a knife on a stick. Yet people protested the combine to try and save them.
I don’t see why the programmer wouldn’t just be the AI’s overseer, in the same way a farm worker now drives a combine instead of slashing wheat by hand.
It's not like we will ever get to a point where we say "all the potential work that could happen is done!". There's always more work to do, as human resources get freed up they get reallocated to other tasks.
It's sort of like Gustafson's Law [1]; it's easy to see all the work we have right now as automate-able and therefore assume "there won't be any jobs", but I think that's making an incorrect assumption that the number of jobs is a fixed size, and I just flatly think this is not true.
[1] https://en.wikipedia.org/wiki/Gustafson%27s_law
We need orders of magnitude fewer AI overseers than we need humans to do the work of the AI. This will flip society on its head.
We don’t miss the jobs. Indeed, our society would collapse if we needed that many farm workers today.
It will flip society on its head. For the better, as there is tons of additional productive capacity.
The latest thing I saw about it seemed to indicate that OpenAI was looking to build a dedicated nuclear power plant because AGI on current hardware might be possible with that much energy [1]. Even if they do crack AGI with that, it's not like its energy consumption will be justifiable.
[1] https://www.independent.co.uk/tech/openai-nuclear-fusion-ene...
Admittedly, my own argument in favor of near-term AGI is much less sophisticated but is also largely based on extrapolating trends and noting the leaps in capabilities that can come from singular architectural innovations. The GPT family has been amazingly productive, and they result from largely a single architectural innovation. They have glaring deficiencies, but their failures are fairly well understood and overcoming them are basically engineering problems at this point. Whereas 10 years ago no one had any idea how to get from the SOTA at the time to AGI, I see a clear path to it from here.
[1] https://situational-awareness.ai/from-gpt-4-to-agi/
But that's a big "if." They seem to be parroting stuff without knowing or caring what's true.
Why would anyone need a human here when the AGI could do that too?
* Not to be confused with an AI Whisperer, who performs the same tasks but didn't complete his certification.
There's been plenty of cases where a big company is inspired by a small company's work, largely redid it, and no one cared, well before AI was a thing. "It! The Terror From Beyond Space" was basically copied by Alien twenty years later, and the latter is considered a masterpiece of cinema, and while most film-geeks know this about Alien, they still consider it unique enough in its own rights to be a great movie. I feel like, in many cases, the stuff generated by Stable Diffusion is transformative enough to be considered fair use. I guess it's a minority opinion, since a large part of the most recent writers strike that everyone seemed to get behind was complaining about AI.
As a rule, I think articles like this are just kind of fear-mongering. Every new piece of tech has a risk of displacing workers, especially in the short term. This isn't new, it's the nature of progress: people really want stuff to be cheaper so we invent stuff to make things cheaper.
Will AI "replace" jobs? Sure, but so did ditch-digging machines, cars, airplanes, and computers themselves; people find other stuff to do.
When AI becomes good enough to code web apps and other simple applications, the people who will be decrying AI will be all of the web devs out of work.
Notto mention, all of the creative and entertainment industry jobs that will be destroyed permanently.
It's coming whether people want to accept it or not and we're going to reach a point where new roles will not be ready to absorb the loss.
At least, within the realm of basically-positive jobs. There’s separately the issue of its being by far the most productivity-enhancing for things that are harmful (scams, astroturfing, spamming)
I agree that making scams easier is a real threat though, and I have absolutely no idea how the hell we're going to properly address that. Even if the main service providers block scam-like language, there's nothing stopping anyone from using an open model and running Llamas or something.
Who said anything about entitled? It’s about better and worse.
“We automated all the things people wished they could do if they didn’t need money—but basically none of the stuff you all actually wished we’d automate” is pretty clearly bleak as fuck.
"Computer" used to be a job title, but that was automated by, you know, computers. We've automated digging ditches, planting crops, transportation, and many, many, many more things. I'd venture to say that most of the stuff we've automated has been stuff people don't really like doing.
Even with AI, there's plenty of stuff it's automating that no one really likes doing. For me, it makes bizarre server logs substantially more digestible, so to a point where I've debated writing a systemd service that listens on journalctl and creates a separate stream of easier-to-understand server logs for me to look at.
I don't really see it as "bleak", and I certainly don't view it as "worse", I think it's just different.
Most art isn't people painting stuff for the Met, it's corporate stuff. I don't really think it's terribly fun for people who do that (though I'm not sure, I don't have much experience in that realm), so if we can automate that isn't that something "you all wished we'd automate"?
> I don't really see it as "bleak"
I'm trying not to see it that way, but it's really hard. Almost everything that the advocates of these systems say about them strikes me as very dystopic.
> I don't really think it's terribly fun for people who do that (though I'm not sure, I don't have much experience in that realm)
I know a few people who make their living doing that, and they really do enjoy it.
A: “Good news, we automated a ton of work. Millions of jobs no longer need to be done.”
P: “Oh, great, now that you automated boring jobs that nobody wants to do, those workers can switch to working in the arts just like they always wanted!”
A: “…”
P: “… you automated the boring work we all hate, right?”
> Notto mention, all of the creative and entertainment industry jobs that will be destroyed permanently.
Sorry, citation needed. Everyone just keeps asserting it like I can just take it as an axiom, and I don't buy it. AI is a tool, it will be treated as a tool, if a tool can do something better than a person then that tool should be used, and if it's worse than a person, then they probably won't be replaced.
If ChatGPT were to start writing code that didn't suck and as a result started replacing lower-level developers and eventually higher-level developers, that would of course suck, but those developers would probably just learn how to use the AI tools and move on with the trend, at least eventually.
People have been saying this crap my entire life (and probably way before it), and I just don't accept the axiom.
[1] I don't do web dev, but you can replace it with "backend data processing crap dev".
Many certainly will. But that sort of job would be very different from what being a dev has entailed when they chose the career. For many devs, the part of programming that they enjoy the most is precisely the part that people are talking about automating away.
That means that there would be devs who find themselves trapped in a job that they now hate.
Stuff like basic scaffolding and long repetitive if/elseif/else chains are things that I have zero fun doing, or looking at long debug logs from build systems makes my life considerably less irritating. It's not perfect at it, but it's also generally useful to paste a chunk of code I wrote years ago and telling me more or less what it does (because I forgot).
Obviously part of this is sort of predicated on the fact that it's still pretty bad at the fun stuff right now, and if it gets better at it then maybe I'll be right now, but in its current state I feel like it's made my life better.
That doesn't surprise me. Different people enjoy different aspects of development, so YMMV.
> Stuff like basic scaffolding and long repetitive if/elseif/else chains are things that I have zero fun doing
I've seen other devs say similar things, and I agree that stuff sucks. But I only rarely have to do that sort of thing at all, so when I do, it's not a great burden. (That said, I actually do know at least one dev who enjoys that kind of thing.)
But it makes me wonder: am I an outlier here? Do others do this stuff often?
Also, I do need to look at server logs all the damn time and I never have fun doing it, especially for some of the more arcane weird stuff (getting an LTO6 drive to work on NixOS was especially irritating). A lot of the less-frequently-used services will often have logs that are really hard to parse. ChatGPT would be worth the price of admission based solely on that.
For that matter, build files are pretty much universally irritating to me. Being able to quickly get some personalized Maven or Gradle information really quick has saved me considerable time and effort.
There’s plenty of more bureaucratic parts of being a software engineer that it helps me with; I have gotten it to write letters of recommendation for me since those are fairly utilitarian at the end of the day, or to help me format slides in Beamer.
I don’t know; again for me it’s been extraordinarily helpful and made software engineering a lot more enjoyable. If I felt like my job were actually at risk from it in the near it’s possible I would have a worse opinion of it.
> I don’t know; again for me it’s been extraordinarily helpful and made software engineering a lot more enjoyable
I found it to be more of an impediment than help, in that I spent at least as much time and effort when using LLMs as assistants than not, and it makes the software engineering much less pleasant for me.
Different strokes and all that. As with all tools, what is great for one person can be terrible for another.
I've been through that. I got into web dev when it was just HTML and maybe some JS (CSS didn't exist) and my interview was a half hour where I only had to demonstrate that I could use vi and create a C program that was basically just getenv("QUERY_STRING").
I miss the days of writing vanilla JS without a ton of dependencies.
I'm not a fan of the "framework of the month" nature of JS, and all the complex toolchains these days certainly has made the job less "fun", but the world isn't going to stop for me.
The only thing that doesn't change is incessant change.
YES this is the future that they want and are pushing towards. AI will eventually write code and make entertainment that will surpass what coders can do. To think this isn't coming is naive. First it's scaffolding and simple stuff. Next it's the harder engineering tasks you enjoy the most.
This is the future that is coming and the reason people have been saying this your entire life is because they're right. It's coming and it's coming NOW.
Don't believe me? Read up on what happened to graphic designers in China.
https://restofworld.org/2023/ai-china-video-game-layoffs-ill...
Next up: web devs
In short, coders are the modern day buggy whip manufacturers who are about to get BTFO by the up and coming AI/automobile. It's coming and your nativite hurts.
The problem of AI being concentrated in the hands of the powerful few is certainly a problem. Which is why its very clearly of the utmost importance that we stomp out AI doomerism now before they can convince congress to enact braindead laws limiting open source AI. Unfortunately Brandolini's law is in full effect and the meme of "The Terminator" and "The Matrix" are way too pervasive.
I see this as an instance of jevrons law. AI will overall induce more demand for work because knowledge work can be leveraged with ai and thus it becomes more productive.
It's not even a net destructive tool like a gun.
AI might be used to create a picture of you! This is way more dangerous than any firearm, or dare I say, even a nuclear weapon.
/s for the /s impaired
And that's the entire point of companies like OpenAI.
As GPUs get better I think we will see a rise of a lot of local-AI. The companies that I think will benefit the most are Nvidia and to a lesser extent AMD (eventually). I would agree that that's too centralized, we should probably have more competitors in the GPU space, but I don't think OpenAI will be able to monopolize AI as a whole for forever.
Reduced work with everything provided would of course be optimal.
AI is a good thing
The cost of a model that is powerful enough to cause a dystopian level of unemployment is - as far as I can tell - considerably beyond the reaches of the private sector. The budgets of the wealthiest companies in history are being stretched to their limits, and they're able to churn out chatbot tech demos, which are admittedly fun, but aren't going to disrupt the entire workforce - or even meaningful percentages of it.
The people at the top of the industry are talking about multi-tens-of-trillion dollar training runs (not to mention the US grid requiring a massive overhaul with several new nuclear powerplants). They need those runs to achieve "it can perform as well as the average line-cook at McDonalds" level general intelligence. The private sector just isn't going to throw endless $trillions at this.
The most likely scenario for a generally intelligent model coming into existence, which can cause really dystopian levels of productivity replacement, is that it'll be funded by the public sector. The obvious social contract there will be that the taxpayer takes on risk in the near-term in exchange for long-term liberation from mandatory labour via UBI or a similar scheme.
People are tired of AI content, especially AI doomerism. Your article can be very thoughtful and even right, but enough people are tired of seeing similar headlines that they flag it off the front page.
- it's not net positive because they are proprietary systems trainable only at giant scale so OR humans still have open universities and public research centers who train models on open datasets or it's the definitive proprietary software even in FLOSS form because the model is trained behind the closed door of some enterprises;
- it's not a net positive because it's used to make money in scenario where the current state of tech is simply way behind the minimum needed to be ready for production and as a result we get more and more dysfunctional services;
- it's not a net positive because to do really not very much it consume an immense amount of resources and we need these resources on other stuff in the present, short and mid term.
Automation of any work is not a threat is a target all humans have, who want to need doing more stuff that can be done by some automation? Who want to manually wash clothes when we can use a washing machine? The point is that automation MUST BE OPEN not only to avoid https://ckrybus.com/static/papers/Bainbridge_1983_Automatica... but more important to grant control of automation for formal owners not to 4 big power and all are then practically their slaves as Industrie 4.0 essentially imply.