I think his point that you basically give a slop generator a fitness function in the form of tests, compilation scripts, and static analysis thresholds, was pretty good. I never really thought of forcing the slop generator to generate slop randomly until it passes tests. That’s a pretty interesting idea. Wasteful for sure, but I can see it saving someone a lot of time.
This is absolutely the crisis of aging hitting the software engineering labor pool hard. There are other industries where 60% or more of the trained people are retiring in 5 years. Software is now on the fast track to get there as well.
This is a great point. I think what is most jarring to me is the speed at which this is happening. I may be wrong but it felt like those other industries, it took at least a couple decades for it to happen, and it feels like tech is doing it in a matter of months?
Nah. It’s two different phenomena with the same end point. Those other industries lost young entrants because of the rise of the college pursuit. And yes that took decades. But for software we’re still 20 years out at least before the we have a retirement crisis.
Although, we already one back in 2000 when not enough working age people knew COBOL.
Anyway, it’s a historical process. It’s just one we’ve seen over and over and just don’t learn the lesson.
wastes its time doing these repetitive and boring tasks
To me, this is sort of a code smell. I’m not going to say that every single bit of work that I have done is unique and engaging, but I think that if a lot of code being written is boring and repetitive, it’s probably not engineered correctly.
It’s easy for me to be flippant and say this and you’d be totally right to point that out. I just felt like getting it out of my head.
If most of the code you write is meaningful code that’s novel and interesting then you are incredibly privileged. Majority of code I’ve seen in the industry is mostly boring and a lot of it just boilerplate.
I’d argue that most of the code is conceptually boilerplate, even when you have a framework to paper over it. There’s really nothing exciting about declaring an HTTP endpoint that’s going to slurp some JSON, massage it a bit, and shove it n your db. It’s a boring repetitive task, and I’m happy to let a tool do it for me.
Absolutely, coders should be spending time developing new and faster algorithms, things that AI cannot do, not figuring out the boilerplate of a dropbox menu on whatever framework. Heck, we dont even need frameworks with AI.
It’s more that the iterative slop generation is pretty energy intensive when you scale it up like this. Tons of tokens in memory, multiple iterations of producing slop, running tests to tell it’s slop and starting it over again automatically. I’d love the time savings as well. I’m just saying we should keep in mind the waste aspect as it’s bound to catch us up.
I don’t really find the waste argument terribly convincing myself. The amount of waste depends on how many tries it needs to get the answer, and how much previous work it can reuse. The quality of output has already improved dramatically, and there’s no reason to expect that this will not continue to get better over time. Meanwhile, there’s every reason to expect that iterative loop will continue to be optimized as well.
In a broader sense, we waste power all the time on all kinds of things. Think of all the ads, crypto, or consumerism in general. There’s nothing uniquely wasteful about LLMs, and at least they can be put towards producing something of value, unlike many things our society wastes energy on.
I do think there’s something uniquely wasteful about floating point arithmetic, which is why need specialized processors for it, and there is something uniquely wasteful about crypto and LLMs, both in terms of electricity but also in terms of waste heat. I agree that generative AI for solving problems is definitely better than crypto, and it’s better than using generative AI to produce creative works, do advertising and marketing, etc.
But it’s not without it’s externalities and putting that in an unmonitored iterative loop at scale requires us to at least consider the costs.
Eventually we most likely will see specialized chips for this, and there are already analog chips being produced for neural networks which are a far better fit. There are selection pressures to improve this tech even under capitalism, since companies running models end up paying for the power usage. And then we have open source models with people optimizing them to run things locally. Personally, I find it mind blowing that we’ve already can run local models on a laptop that perform roughly as well as models that required a whole data centre to run just a year ago. It’s hard to say when all the low hanging fruit is picked, will improvements start to plateau, but so far it’s been really impressive to watch.
Yeah, there is something to be said for changing the hardware. Producing the models is still expensive even if running the models is becoming more efficient. But DeepSeek shows us even production is becoming more efficient.
What’s impressive to me is how useful the concept of the stochastic parrot is turning out to be. It doesn’t seem to make a lot of sense, at first or even second glace, that choosing the most probable next word in a sentence based on the statistical distribution of word usages across a training set would actually be all that useful.
I’ve used it for coding before and it’s obvious that these things are most useful at reproducing code tutorials or code examples and not at all for reasoning, but there’s a lot of code examples and tutorials out there that I haven’t read yet and never will read. The ability of a stochastic parrot to reproduce that code using human language as it’s control input is impressive.
I’ve been amazed by this idea ever since I learned about Markov chains, and arguably LLMs aren’t fundamentally different in nature. It’s simply a huge token space encoded in a multidimensional matrix, but the fundamental idea is the same. It’s really interesting how you start getting emergent properties when you scale something conceptually simple up. It might say something about the nature of our own cognition as well.
Wasteful for sure, but I can see it saving someone a lot of time.
Many degrowth proposals call for some aggregate reduction of energy use or material throughput. The issue with these proposals is that they conflict with the need to give the entire planet public housing, public transit, reliable electricity, modern water-sewage services, etc., which cannot be achieved by “shrinking material throughput”. According to modeling from Princeton University (this may be outdated), it suggests that zeroing emissions by 2050 will require 80 to 120 million heat pumps, up to 5 times an increase in electricity transmission capacity, 250 large or 3,800 nuclear reactors, and the development of a new carbon capture and sequestration industry from scratch. Degrowth policies, while not intending to result in ecological austerity, effectively do so through their fiscal commitment to budgetary constraints which inevitably require government cuts.
The reason for the above paragraph is to give an analogy to the controversy of “AI wastefulness”. Relying on manual labor for software development could actually lead to more wastefulness long term and a failure to resolve the climate crisis in time. Even though AI requires a lot of power, creating solutions faster (especially in green industries as well as emissions reduced from humans such as commuting to work) could lead to a better and faster impact on reducing emissions.
While an interesting point, it relies on a specific assumption - that LLMs are useful in solving the problems you’re talking about. Unfortunately, as we’ve seen from nearly all other advances in human productivity, we just take surplus labor and apply it to completely wasteful projects:
DRM
IP litigation
IP violation detection
Marketing and sales
Public relations (7x as many people in PR as in journalism)
Imperial militarism and the resultant defense against such
Imperial propaganda and the defense against such
FOMO
More entertaining entertainment
Actor salaries
Stadium concerts
Homographs of celebrities
Funkopops and other purely wasteful plastic treats
Corporate rebranding
Branded corporate swag
Drop shipping
The war on drugs
Student debt collection
Medical debt collection
Financial abstractions
Vanity
Fast fashion
Fried ranch dressing and other food where unhealthy is the point
Analyzing, exploiting, and litigating addiction
For profit prisons and the parole system
I could go on. This is what we choose to spend our surplus labor on. So AI time savings just isn’t going to save us. AI would have to fundamentally change the way solve certain problems, not improve the productivity of the billions of people who are already wasting most of the careers working on things that make the problems you’re talking about worse and not better.
Yes, neural network techniques are useful in scientific applications to fundamentally change how we solve problems. Helping a mid level programmer get more productive at building AdTech, MarTech, video games, media distribution, subscription services, ecommerce, DRM, and every other waste of human productivity relative to the problems you’re raising is done by LLMs, which are not useful for protein folding, physics simulations, materials analysis, and all the other critical applications of AI that we need to solve.
I was just implying after revolution and when we live in a socialist society, we would use AI for productive means and not for these wasteful projects. Your list above are mostly projects in a capitalist society that serve the ruling class’s interests.
The only real potential benefit of AI in a capitalist society, besides potentially using it to make tools, services, and content for workers and communist parties to fight back against the system, is the proletarianization and hopefully radicalization (toward socialism) of labor aristocrats as the deepening contradictions of capitalism lead to more unrest amongst the working class.
Yeah, so, I am not saying AI is too wasteful to exist. I am saying it’s too wasteful to be used for worker productivity in the current global capitalist system. Workers in China, Vietnam, India, Pakistan, Bulgaria, Nigeria, and Venezuela are going to use LLMs first and foremost to remain competitive in the race to the bottom for their wages in the global capitalist system, because that’s where we are at and will be for as long as it takes.
This is exactly what I’m talking about. Lots of privileged first worlders have no idea what this means for the globe as well. It’s not going to liberate the third world; instead it will mantle a much heavier burden of productivity upon them as wealth extraction of the meager white-collar is kicked into overdrive as databases for these are built in the periphery or even in poor places at home.
I mean look at where Colossus is being built. While we as workers might be able to use AI reasonably and responsibly; it has no more meaning and impact than using paper/plastic straws in comparison to the ruling class running giant data-centers and energy-intensive slop machines to our “open source!” equivalent.
The effect of weakening labor aristocracy is secondary.
First, I am only meaning to provide my perspective, even if it turns out to be not perfect or 100% accurate to reality, and I am willing to be corrected or to learn from others. I’m not just some first worlder set in their ways.
I think you misunderstood some of the points I made, just as I interpreted freagle’s point about AI wastefulness to be in a broader sense than it was. I don’t believe AI as it exists now in the global capitalist system will liberate the third world. My guess is the ruling class’s use of the AI will probably have a greater impact against the working class’s interests than the impact the working class will have to counter AI through its own use. We have no control over AI’s existence. It’s a reality we have to live with. The only way AI would have an improvement on the working class’s lives across the entire planet would be for a socialist system to become dominant across the world and the global capitalist hegemony to be overthrown.
I’m not denying the ruling class’s use of AI is a much greater detriment to us than any gains we get from the weakening of the labor aristocracy. However, I believe as those people start losing their jobs, communist parties will need to start reaching out to them, educate them, and bring them to our cause so we can develop the numbers and power to overthrow the ruling class, which I believe is the upmost importance. I honestly don’t believe most labor aristocrats, especially in the West, will be radicalized until they become proletarianized and their material conditions greatly worsen.
My guess is the ruling class’s use of the AI will probably have a greater impact against the working class’s interests than the impact the working class will have to counter AI through its own use. We have no control over AI’s existence. It’s a reality we have to live with. The only way AI would have an improvement on the working class’s lives across the entire planet would be for a socialist system to become dominant across the world and the global capitalist hegemony to be overthrown.
That is exactly my point, friend, wasn’t trying to argue that the technology itself is inherently evil. It’s that you can’t expect people to embrace and understand this as a useful tool to the communist cause when it doesn’t even have a clear use-case there besides for coding projects. I understand it’s the reality we live with, it’s why I’m not suggesting “OPEN REVOLT AGAINST AI!” because it’s just the reality we’re in now. My point is that people are going to continue having a disgust for it and a backlash against it because it’s going to do more harm than good even with a handful of clear use-cases.
Trying to sell the “goods” of this technology when it’s doing so much harm already comes off as tasteless.
I’m not denying the ruling class’s use of AI is a much greater detriment to us than any gains we get from the weakening of the labor aristocracy. However, I believe as those people start losing their jobs, communist parties will need to start reaching out to them, educate them, and bring them to our cause so we can develop the numbers and power to overthrow the ruling class, which I believe is the upmost importance. I honestly don’t believe most labor aristocrats, especially in the West, will be radicalized until they become proletarianized and their material conditions greatly worsen.
Every six months the tone of these “why won’t you use my hallucinating slop generator” get more and more shrill.
I think his point that you basically give a slop generator a fitness function in the form of tests, compilation scripts, and static analysis thresholds, was pretty good. I never really thought of forcing the slop generator to generate slop randomly until it passes tests. That’s a pretty interesting idea. Wasteful for sure, but I can see it saving someone a lot of time.
I have to chuckle at this because it’s practically the same way that you have to manage junior engineers, sometimes.
It really shows how barely “good enough” is killing off all the junior engineers, and once I die, who’s going to replace me?
This is absolutely the crisis of aging hitting the software engineering labor pool hard. There are other industries where 60% or more of the trained people are retiring in 5 years. Software is now on the fast track to get there as well.
This is a great point. I think what is most jarring to me is the speed at which this is happening. I may be wrong but it felt like those other industries, it took at least a couple decades for it to happen, and it feels like tech is doing it in a matter of months?
Nah. It’s two different phenomena with the same end point. Those other industries lost young entrants because of the rise of the college pursuit. And yes that took decades. But for software we’re still 20 years out at least before the we have a retirement crisis.
Although, we already one back in 2000 when not enough working age people knew COBOL.
Anyway, it’s a historical process. It’s just one we’ve seen over and over and just don’t learn the lesson.
I’d much rather the slop generator wastes its time doing these repetitive and boring tasks so I can spend my time doing something more interesting.
To me, this is sort of a code smell. I’m not going to say that every single bit of work that I have done is unique and engaging, but I think that if a lot of code being written is boring and repetitive, it’s probably not engineered correctly.
It’s easy for me to be flippant and say this and you’d be totally right to point that out. I just felt like getting it out of my head.
If most of the code you write is meaningful code that’s novel and interesting then you are incredibly privileged. Majority of code I’ve seen in the industry is mostly boring and a lot of it just boilerplate.
This is possible but I doubt it. It’s your usual CRUD web application with some business logic and some async workers.
So then you do write a bunch of boilerplate such as HTTP endpoints, database queries, and so on.
Not really. It’s Django and Django Rest Framework so there really isn’t a lot of boilerplate. That’s all hidden behind the framework
I’d argue that most of the code is conceptually boilerplate, even when you have a framework to paper over it. There’s really nothing exciting about declaring an HTTP endpoint that’s going to slurp some JSON, massage it a bit, and shove it n your db. It’s a boring repetitive task, and I’m happy to let a tool do it for me.
Absolutely, coders should be spending time developing new and faster algorithms, things that AI cannot do, not figuring out the boilerplate of a dropbox menu on whatever framework. Heck, we dont even need frameworks with AI.
It’s more that the iterative slop generation is pretty energy intensive when you scale it up like this. Tons of tokens in memory, multiple iterations of producing slop, running tests to tell it’s slop and starting it over again automatically. I’d love the time savings as well. I’m just saying we should keep in mind the waste aspect as it’s bound to catch us up.
I don’t really find the waste argument terribly convincing myself. The amount of waste depends on how many tries it needs to get the answer, and how much previous work it can reuse. The quality of output has already improved dramatically, and there’s no reason to expect that this will not continue to get better over time. Meanwhile, there’s every reason to expect that iterative loop will continue to be optimized as well.
In a broader sense, we waste power all the time on all kinds of things. Think of all the ads, crypto, or consumerism in general. There’s nothing uniquely wasteful about LLMs, and at least they can be put towards producing something of value, unlike many things our society wastes energy on.
I do think there’s something uniquely wasteful about floating point arithmetic, which is why need specialized processors for it, and there is something uniquely wasteful about crypto and LLMs, both in terms of electricity but also in terms of waste heat. I agree that generative AI for solving problems is definitely better than crypto, and it’s better than using generative AI to produce creative works, do advertising and marketing, etc.
But it’s not without it’s externalities and putting that in an unmonitored iterative loop at scale requires us to at least consider the costs.
Eventually we most likely will see specialized chips for this, and there are already analog chips being produced for neural networks which are a far better fit. There are selection pressures to improve this tech even under capitalism, since companies running models end up paying for the power usage. And then we have open source models with people optimizing them to run things locally. Personally, I find it mind blowing that we’ve already can run local models on a laptop that perform roughly as well as models that required a whole data centre to run just a year ago. It’s hard to say when all the low hanging fruit is picked, will improvements start to plateau, but so far it’s been really impressive to watch.
Yeah, there is something to be said for changing the hardware. Producing the models is still expensive even if running the models is becoming more efficient. But DeepSeek shows us even production is becoming more efficient.
What’s impressive to me is how useful the concept of the stochastic parrot is turning out to be. It doesn’t seem to make a lot of sense, at first or even second glace, that choosing the most probable next word in a sentence based on the statistical distribution of word usages across a training set would actually be all that useful.
I’ve used it for coding before and it’s obvious that these things are most useful at reproducing code tutorials or code examples and not at all for reasoning, but there’s a lot of code examples and tutorials out there that I haven’t read yet and never will read. The ability of a stochastic parrot to reproduce that code using human language as it’s control input is impressive.
I’ve been amazed by this idea ever since I learned about Markov chains, and arguably LLMs aren’t fundamentally different in nature. It’s simply a huge token space encoded in a multidimensional matrix, but the fundamental idea is the same. It’s really interesting how you start getting emergent properties when you scale something conceptually simple up. It might say something about the nature of our own cognition as well.
Many degrowth proposals call for some aggregate reduction of energy use or material throughput. The issue with these proposals is that they conflict with the need to give the entire planet public housing, public transit, reliable electricity, modern water-sewage services, etc., which cannot be achieved by “shrinking material throughput”. According to modeling from Princeton University (this may be outdated), it suggests that zeroing emissions by 2050 will require 80 to 120 million heat pumps, up to 5 times an increase in electricity transmission capacity, 250 large or 3,800 nuclear reactors, and the development of a new carbon capture and sequestration industry from scratch. Degrowth policies, while not intending to result in ecological austerity, effectively do so through their fiscal commitment to budgetary constraints which inevitably require government cuts.
The reason for the above paragraph is to give an analogy to the controversy of “AI wastefulness”. Relying on manual labor for software development could actually lead to more wastefulness long term and a failure to resolve the climate crisis in time. Even though AI requires a lot of power, creating solutions faster (especially in green industries as well as emissions reduced from humans such as commuting to work) could lead to a better and faster impact on reducing emissions.
While an interesting point, it relies on a specific assumption - that LLMs are useful in solving the problems you’re talking about. Unfortunately, as we’ve seen from nearly all other advances in human productivity, we just take surplus labor and apply it to completely wasteful projects:
I could go on. This is what we choose to spend our surplus labor on. So AI time savings just isn’t going to save us. AI would have to fundamentally change the way solve certain problems, not improve the productivity of the billions of people who are already wasting most of the careers working on things that make the problems you’re talking about worse and not better.
Yes, neural network techniques are useful in scientific applications to fundamentally change how we solve problems. Helping a mid level programmer get more productive at building AdTech, MarTech, video games, media distribution, subscription services, ecommerce, DRM, and every other waste of human productivity relative to the problems you’re raising is done by LLMs, which are not useful for protein folding, physics simulations, materials analysis, and all the other critical applications of AI that we need to solve.
I was just implying after revolution and when we live in a socialist society, we would use AI for productive means and not for these wasteful projects. Your list above are mostly projects in a capitalist society that serve the ruling class’s interests.
The only real potential benefit of AI in a capitalist society, besides potentially using it to make tools, services, and content for workers and communist parties to fight back against the system, is the proletarianization and hopefully radicalization (toward socialism) of labor aristocrats as the deepening contradictions of capitalism lead to more unrest amongst the working class.
Yeah, so, I am not saying AI is too wasteful to exist. I am saying it’s too wasteful to be used for worker productivity in the current global capitalist system. Workers in China, Vietnam, India, Pakistan, Bulgaria, Nigeria, and Venezuela are going to use LLMs first and foremost to remain competitive in the race to the bottom for their wages in the global capitalist system, because that’s where we are at and will be for as long as it takes.
This is exactly what I’m talking about. Lots of privileged first worlders have no idea what this means for the globe as well. It’s not going to liberate the third world; instead it will mantle a much heavier burden of productivity upon them as wealth extraction of the meager white-collar is kicked into overdrive as databases for these are built in the periphery or even in poor places at home.
I mean look at where Colossus is being built. While we as workers might be able to use AI reasonably and responsibly; it has no more meaning and impact than using paper/plastic straws in comparison to the ruling class running giant data-centers and energy-intensive slop machines to our “open source!” equivalent.
The effect of weakening labor aristocracy is secondary.
First, I am only meaning to provide my perspective, even if it turns out to be not perfect or 100% accurate to reality, and I am willing to be corrected or to learn from others. I’m not just some first worlder set in their ways.
I think you misunderstood some of the points I made, just as I interpreted freagle’s point about AI wastefulness to be in a broader sense than it was. I don’t believe AI as it exists now in the global capitalist system will liberate the third world. My guess is the ruling class’s use of the AI will probably have a greater impact against the working class’s interests than the impact the working class will have to counter AI through its own use. We have no control over AI’s existence. It’s a reality we have to live with. The only way AI would have an improvement on the working class’s lives across the entire planet would be for a socialist system to become dominant across the world and the global capitalist hegemony to be overthrown.
I’m not denying the ruling class’s use of AI is a much greater detriment to us than any gains we get from the weakening of the labor aristocracy. However, I believe as those people start losing their jobs, communist parties will need to start reaching out to them, educate them, and bring them to our cause so we can develop the numbers and power to overthrow the ruling class, which I believe is the upmost importance. I honestly don’t believe most labor aristocrats, especially in the West, will be radicalized until they become proletarianized and their material conditions greatly worsen.
That is exactly my point, friend, wasn’t trying to argue that the technology itself is inherently evil. It’s that you can’t expect people to embrace and understand this as a useful tool to the communist cause when it doesn’t even have a clear use-case there besides for coding projects. I understand it’s the reality we live with, it’s why I’m not suggesting “OPEN REVOLT AGAINST AI!” because it’s just the reality we’re in now. My point is that people are going to continue having a disgust for it and a backlash against it because it’s going to do more harm than good even with a handful of clear use-cases.
Trying to sell the “goods” of this technology when it’s doing so much harm already comes off as tasteless.
Agreed.