There are also those for whom that percentage is higher, let’s say 6-50%.
> I understand things and then apply my ability to formulate solutions
The AI is coming for that too.
You might just be lucky to be in circumstances that value your contributions or an industry or domain that isn’t well represented in the training data, or problem spaces too complex for AI. Not everyone is, not even the majority of devs.
People knocking out Jira tickets and writing CRUD webapps will end up with their livelihood often taken away. Or bosses will just expect more output for same/less pay, with them having to use AI to keep up.
Agree. It is just like 2 totally separate groups are arguing.
One very tiny slice of speciality/ rare industries where code is critical but overall small part of project costs. I can see if code / software is 5% of overall cost even heavy use of AI for code part is not moving the needle. So people in this group can feel confident in their indispensability.
Second group is much larger and peddling CRUD / JS Frontends and other copy/paste junk. But as per industry classification they are just part of same Coder/Developer/IT Engineer group. And their bleak prospects is not some future scenario, it is playing out right now with tons of them getting laid off. And whole lot of people with IT degrees, certifications are not finding any jobs in this field.
We've had reasonable effectiveness for CRUD. It's mainly the UI toolkits we use, but the plumbing it can do quite well. It's not 100% vibecoding but certainly a significant accelerator for parts of the job.
I agree with the 2 separate groups theory, but I don't buy that the group that produces "copy/paste junk" is the much larger group. I think in most mega-corps, there is a huge existing code base, there are huge organizational challenges, and there is huge hierarchy with most people not being the junior juniors. 90+% of the work is "not coding." Probably way, way more if we include the middle managers. At startups, there is a lot of "copy/paste junk" but also often a decent amount of push the boundary new stuff. I don't know. I've been in the industry for 8 years now and it's been really rare to see the actual coding being the bottleneck or even the thing that takes the majority of the time.
I don't mean this as a snarky jab. It's coming for anything software. I've used AI to accomplish front end development and reverse engineer proprietary USB hardware dongles in C, then rewriting the C into Rust to get easy desktop GUIs around it. Backend APIs, systems programming, embedded programming, they all seem equally threatened it's just a matter of time. Front end is easy to see in the AI web front ends but everything else is still easy pickings.
You are describing the toy projects that had us all amazed end of last year. Large, maintainable software that can serve paying customers is in a completely different galaxy.
> I've used AI to accomplish front end development and reverse engineer proprietary USB hardware dongles in C, then rewriting the C into Rust to get easy desktop GUIs around it. Backend
That is not hard. It’s just tedious and very slow to do manually. The hard part would be about designing a usb dongle and ensuring that the associated software has good UX. The reason you don’t see kernel devs REing devices is not because it’s impossible or that it requires expert knowledge. It’s because it’s like counting sands on the beach.
Whether something is tedious depends on the person and situation. If you're already an expert, you may find a lot of work that goes into your 4th USB device (especially if it's based on yet another chip and bespoke SDK) quite tedious, since lot of it is based on standard requirements/designs that you can't change.
You may also find re-ing stuff not tedious, due to what may be motivating you.
In any case, any work will have some things you just know how to do, or what to do, but previously (before LLM agents) no easy way to plow through them without pressing a lot of keyboard keys over long period of time.
There's rather a big difference between reverse engineering already working code and forward(?) engineering working code from nothing so that confidence seems misplaced.
As a manager of a full stack team, we've found AI falls short a lot more on front end. It has its weak points on both front and back, but the problems with backend are quite easy to feed back into it -- needs more performance, needs to pass this security audit, needs to deal with xyz system. The problems with frontend are more like this is ugly, it's clunky to use, people don't like it. People without years of frontend experience tend to lack the vocabulary required to get AI to fix it, period, and it ends up going around in loops.
It is irrelevant that complex frontend would be easy for AI or not. To me 1) how many unique complex frontends are needed out of total frontends that millions of sites out there need. 2) Will there be increase in need of such frontend engineers so other displaced folks can land a job there.
I think it will be far fewer to have any positive impact on IT engineers' overall job prospects.
But that's equally true for any type of system. Frontend isn't inherently easier than other systems, so i was just wondering why you singled it out. To me AI just seems better at backends and database design
OK, my examples seemed like biased against frontend which was not the intention.
The thrust was overall job prospects for people in software field. It is not that frontend is easy but it is definitely easy to get into. Considering there are far more frontend developers then say C++ system engineers or database designers so in sheer numbers they will be affect more.
Ah okay that's fair. In my country boot camps aren't a thing so frontend devs are rare and good frontend devs even more, so I think it depends on where in the world you are. We got an abundance of java devs here that i fear more for
There are periods of time where I might spend 80% of my time "coding", meaning I have minimal meetings and other responsibilities.
However, even out of that 80% of my time, what fraction is actually spent "writing code"?
AI can be an enormous accelerator for the time I'd normally spend writing lines of code by hand, but it doesn't really help with the rest of the work:
- Understanding the problem
- Waiting for the build system and tests to run
- Manually testing the app to make sure it behaves as I'd like
- Reviewing the diff to make sure it's clear
- Uploading the PR and writing a description
- Responding to reviewer feedback
There are times when AI can do the "write the code" portion 10x faster than I could, but if it's production code that actually matters, by the time I actually review the code, I doubt it's more than 2x.
>AI can be an enormous accelerator for the time I'd normally spend writing lines of code by hand, but it doesn't really help with the rest of the work:
- Understanding the problem - Waiting for the build system and tests to run - Manually testing the app to make sure it behaves as I'd like - Reviewing the diff to make sure it's clear - Uploading the PR and writing a description - Responding to reviewer feedback
What part of those you think it doesn't help with?
They can make it unnecessary for you to understand.
Consider hash tables. Nobody implements a hash table by hand any more.
I've written some, but not in this century.
Optimal hash table design is a specialist subject. Do you know about robin hood algorithms? Changing the random number generator's seed to discourage collision attacks?
A basic hash table starts to slow down around 70% full. Modern hash tables can get above 90% full before they have to expand.
Who keeps Knuth's Fundamental Algorithms handy any more? I own both the original edition and the revised edition. They're boxed up in the garage. I once read that book cover to cover. That was a long time ago.
That's not AI. That's solving the problem and putting it in a black box. That's how technology progresses.
That's obviously not what I'm talking about. If you're asking an AI to write an optimal hash table algorithm, something is clearly wrong. I'm talking specifically about understanding the business domain and problem you are trying to solve.
> That's not AI. That's solving the problem and putting it in a black box. That's how technology progresses.
The key word is solving. Meaning someone, after coming up with the solution, has taken times to prove that it works well in all usual and most extreme cases. With their reputation on the line.
That’s why you trust curl, ffmpeg, Knuth’s books,… but you don’t trust random cat on the internet. We don’t trust AI and the cost to review its output is not a great tradeoffs compared to just think and solve the problem.
That may be true I’m not gonna say one way or the other, but if AI comes for that then almost all knowledge work is effectively dead, so all that’s left would be sales or physical labor.
> The history of the last 250 years is inventing new professions as old ones are automated away.
Even if this still holds true ("past performance is no guarantee of future results") the part about it that people handwave away without thinking about or addressing is how awful the transitional period can be.
The industrial revolution worked out well for the human labor force in the long term, but there were multiple generations of people who suffered through a horrendous transition (one that was only alleviated by the rise of a strong labor movement that may not be replicable in the age of AI, given how it is likely to shift the leverage of labor vs. capital).
If you want to lean on history as an indication that massive sudden productivity changes will make things better for humanity in the long run, then fine, but then you have to acknowledge that (based on that same history) the transition could still be absolutely chaotic and awful for the lifespan of anyone who is currently alive.
The history of the last 250 was moving from agriculture to industrial work to service work. Now the last frontier is starting to be overtaken by automation too.
(And in all of those transitions millions where left behind without work or with very worse prospects. The people that took the new jobs were often a different group, not people who knew the old jobs and were already in their 30s and 40s).
And what would be the new professions that uniquely require humans, when even thinking and creative jobs are eaten by AI? Would there be a boom of demand for dancers and chefs, especially as millions lose their service jobs?
> The history of the last 250 years is inventing new professions as old ones are automated away.
Are you going to have the time and money in 2035 to spend 5 years retraining though? Will you be as cheap and fast as the young ones that will learn with you and ask for half your salary?
New professions will likely come no doubt. Whether existing professionals will be able to transition to those easily is another matter.
This is the kind of sleep walking that’s about to walk humanity into the next dark ages.
My parents say a lot of stuff like this. They tend to gloss over the untold suffering, great depressions and world wars that took us to get here.
The planets resources were also not in risk of running out. As the world is min maxed by billionaires, it nice the lower classes are drained of all capital, they will soon move to fighting each other for resources. the future is looking pretty grim for even the most optimistic of scenarios
Doordash and similar are experimenting with autonomous/remotely operated vehicles and porn is getting decimated once good enough uncensored video gen ai gets available. That doesn't sound like viable career choices either.
It's happening, but theres no law of the universe that says it has to be 1:1. Why are you so confident in this regard? 250 years is a very small slice of human history and could easily be the outlier.
I wonder though, can AI make the next JS framework. I mean that in sincerity, there was the leap from jQuery to React for ex. If an AI only knows jQuery and no one makes React, will React come out of AI.
I think it won't be like assembly, because it takes more information vs building blocks that have more dense information in them, kind of like how we use libraries and frameworks
Yeah that's my thing for my hardware projects, I'm not going to reach for an LLM to do it, I want to write the code myself/be present. For something new I would consider using LLM to generate something, like a computer vision implementation or something I don't already know. The end result I would know how it works, just for POC.
People didn't leap from jQuery to React. It's a lot easier to imagine an AI looking at jQuery and [insert any server side MVC framework] and inventing Backbone.
It can't. Framework hierarchy is largely based on social structure, rather than pure technical merit. Otherwise React would've been displaced long time ago.
Yes, but if/when that happens, it won't just affect software engineers. An AI that can do that can replace any white collar worker.
> People knocking out Jira tickets and writing CRUD webapps will end up with their livelihood often taken away.
I'm not sure anyone is actually working on those. People talk about spending all day writing CRUD apps here, but if you suggest there are already low code tools to build those, they will promptly tell you it's too complex for that to work.
>Yes, but if/when that happens, it won't just affect software engineers. An AI that can do that can replace any white collar worker.
Yes. Yes, that's exactly what we're going to see, and more swiftly than people are generally comfortable with. What are we going to do with all those cubicle dwellers?
a new paradigm of 3 day work weeks. share the salary of the days off with those less automatable, and work to automate everyone.
I wish some sort of dicussion like this could happen where the workers of the world get to see some of the gains of a new technology more immediately.
If "the state" wants to maintain legitimacy and protect its citizenry (one of the primary promises of a state) to avoid a period of social unrest, the likes of which has been unheard of for several generations, I think something like this should at least be a part of the discussion.
I don't think it will happen. For one thing, it hasn't in the US since the introduction of the forty hour work week in 1940.
But beyond that, I don't think most people want a three day work week. They would rather work five days and get the extra money. I worked at a company that did government contracting. We had a couple quarters without much in the way of orders, so instead of laying people off like you'd normally see in that situation, the company decided to go to a four day week, with a commensurate cut in pay.
I was thrilled, as a young single guy, to get Fridays off. I rented a room in someone's house and hit my monthly nut in about two weeks. But most of the people I worked with hated it. Some of them quit. A lot of them both needed the money and also had no idea what to do with themselves on that extra day.
5+ years in the software world is like 30 years in others...So...given lacking use-cases and humongous amounts of capital already wasted on chatbots...It's more like "we" are closer to closing curtains than to "just started"...
It feels its just around the corner. But when you turn 20th corner and its still behind the next one, maybe things are a bit different than they seem / clueless emotions make us believe.
Long term its bleak, but short/medium term - not so much, if I get fired it won't be llm replacing me but rather company politics, budget changes etc. Which was the only real (very real) risk for past 15 years too, consistently. But it helps to not work for US company.
It is the lament of every generation of humans to think that they are the pinnacle of everything that has come before, we are just at the start of the so-called AI era, many very smart people coming up still haven’t really got their hands on all of the material available from a hardware and software standpoint. We are still at the early stages.
I am very optimistic. I just wish I was younger to take advantage like Junior high, high school age with my current resources damn… The oldest lament in the books.
We're not at the early stages. The LLM hype started almost for years ago! And for the past two or three years, progress has plateaued (if not regressed) and the economics, which still have to be figured out, point to it becoming an order of magnitude more expensive at least, compared to this aggressive marketing and VC phase.
I've been told for three years that it's the early days, and everything completely changes every few months, and these are the worst the tools will ever be. Meanwhile, I see very little technical progress, zero return on investment anywhere, negative infinity profitability on the side of the providers, and a fast growing realisation among the masses on all the myriad tasks they are systemically unsuitable for.
There are people and companies out there releasing entire vibe coded projects and for some upwards of 80% of the code they develop is AI-assisted/generated. Since around the end of 2025 and models like Opus 4.6, the SOTA has gotten good enough to work agentically on all sorts of dev tasks with pretty good degrees of success (harnesses and how you use them still matters, ofc).
> There are people and companies out there releasing entire vibe coded projects and for some upwards of 80% of the code they develop is AI-assisted/generated.
I mean this is just fingers-in-your-ears "LA LA LA I CAN'T HEAR YOU!!" stuff.
I still have a job so AI hasn't taken that yet. But the suggestion it's "no closer" is ridiculous. At least in my life/career/office this last 12 months seems to have been a real inflection point in how AI is being used for software development.
Discovery of the best solution in a problem space is not generative but only verificative. Meaning: the LLM can see if a solution is better than another, but it can't generate the best one from the start. If you trust it, you'll get sub-par solutions.
This is definitely an agent problem instead of an LLM problem. Anybody got something explorative like this working?
>> Or bosses will just expect more output for same/less pay, with them having to use AI to keep up.
Anecdotal evidence to support this.
I work with both dev and design teams. Upper management has already gone through several layoffs and offshoring of the two dev teams I work with. The devs they did keep were exactly what you said. The capable ones who reliably closed their Jira tickets. Never missed a deadline for building their features or components. And now? Their work has tripled and now the only help they get from management? "Start to figure out how to leverage AI, we're going to be a in hiring freeze for the next 10 months."
The double whammy of losing onshore team members and not getting any help from management to fix the problem they just created and essentially just telling them to figure out how to use AI to keep up is pretty staggering.
I would echo what one of the devs told me, "If this is the new "AI era" than you can count me right the fuck out of it."
>> I understand things and then apply my ability to formulate solutions
> The AI is coming for that too.
If this is true, then you'd have to conclude that AI is coming for everything. I'm still not convinced by that. But I am convinced that the part of software development that involves typing code manually into an IDE all day is likely gone forever.
It really doesn't have to come for everything to feel like it's taking everything. If it eliminates 10% of white collar jobs over the next decade, the impact will be felt everywhere.
I struggle to understand the logic (in general, the way people are talking), normally efficiencies come with increases in production and scale and use-cases.
So of 10% of lawyers get AI-d away, let’s say, the remaining 90% are 1.1x+ efficient and also up against other lawyers enjoying the same… work might go up. And on the customer side there is sooooo much BS with lawyers, but if both lawyer and customer can communicate faster or better with the LLMs, we should see more better cases with better dialog and case handling. Again, the total amount of lawyering could go up a lot. And then we have the cases prohibitive without the LLMs, now possible for big money. Better LLM empowered lawyers should be able to create new and more lawyer work.
As it stands I see people selling services that are subsidized by VC, template jobs we’d be doing faster with copy paste but it’s not copyright infringement when OpenAI does it, and a rush for valuations to soak up VC because the business model isn’t there. I’m seeing a huge uptick in visual bugs on large commercial platforms and customer facing apps, and don’t feel OpenAI is gonna kill Office anytime soon… or Chromium… or Steam… or emacs…
Call me an optimist, but I think those LLM pump and dumpers are creating a wave of fear that would be quite different if they weren’t lying and trying to boost an IPO. Chat GPT 2 was too dangerous to release, lul, and the class action suits are just getting started.
An actual lawyer replacing tech company should sell lawyering for infini-money, not pens that’ll totally 10x your lawyering (bro).
And what do those 10% of lawyers do? Every other industry also got reduced by 10+%, its not like they have a job elsewhere.
So.... they just starve in the streets?
Even if some other, arguably better job comes along, would they retrain for it? (You can say yes, but take a look at the long history of people choosing to join a cult and vote for an orange moron instead of learning a new skill).
Either you're convinced you won't be too badly affected and will gladly watch huge swaths of people suffer, or you're deluded enough to think that it will really, truly be different this time. In any case, I hope you get the worst results of what you preach.
Even if AI advances continue, for quite a while there's likely still going to be the 'Steve Jobs' role. That is, even if AI coding agents can, in the future, replace entire teams of SWEs, competently making all implementation decisions with no guidance from a tech-savvy human, the best software will likely still involve a human deciding what should be built and being very picky about how, exactly, it should externally behave.
I don't know if it makes sense to call that person an SWE, and some people currently employed as SWEs either won't be good at this or aren't interested in doing it. But the existing pool of SWEs is probably the largest concentration of people who'll end up doing this job, because it's the largest concentration of people who've thought a lot about, and developed taste with respect to, how software should work.
This matches what I'm seeing. I've been building software for a long time, but building more now with AI than I ever could with a traditional team. But the throughput that's helpful is from knowing what to build and what tradeoffs matter. The AI doesn't have that. It's a force multiplier on experience, not a replacement for it.
Yes, AI is coming for solution formulation, absolutely, but not all of it, because it is actually a statistical machine with context limit.
Until the day LLMs are not statistical machine with a context limit, this will hold. Someone need to make something that has intent and purpose, and evidently now not by adding another 10T to the LLM parameter count.
> because it is actually a statistical machine with context limit.
So are humans.
Machines have surpassed humans by magnitudes in many capabilities already (how many billion multiplications can you do per second?)
And I argue that current LLMs have surpassed many of my capabilities already.
For example GPT/Opus can understand and document some ancient legacy project I never saw before in minutes. I would take a week+ to do the same and my report would probably have more mistakes and oversights than the one generated by the LLM.
AI advocates are _way_ too confident about the nature human cognition. Questions that have been debated by philosophers and cognitive scientists for decades are now "obvious" according to you people, though you never provide any argument to support your statements.
>Machines have surpassed humans by magnitudes in many capabilities already (how many billion multiplications can you do per second?)
Do you have any idea how many calculations it takes for a human to put a ball through a hoop while running across a court?
It could be millions or billions in a second. Manifesting consciousness, coordinating body movements, and everything else all at the same time takes calculations.
You may not be aware that your brain is doing multiplication, or any other kinds of math, constantly, but it is.
I agree we do some marvelous things in sports but if we extrapolate from this table tennis robot, it's clear machines can/will do just as well there too:
That table tennis robot is not conscious. That table tennis robot does one thing well. A human is capable of far more. There is far more going on for a human playing table tennis than a robot. It doesn't matter if the table tennis robot plays table tennis better, it can't also play hocket, soccer, football, basketball, chess, polo, baseball, or many other things one human can do.
The human condition is nothing but a massive amount of calculations under the hood. You don't feel it, or understand it, but it's there. Everything in nature is math, every physical phenomena has a cause and effect rooted in mathematics, and it's no surprise that humans are great at subconsciously calculating myriad things on-the-fly, as life is happening around us.
We are not pre-trained using the summary of all human knowledge over all of history. Yet we make certain decisions with much more ease.
We are much more limited, but we fundamentally work differently. Hence adding more parameter like certain companies are doing isn't necessarily going to help. We need to rethink how LLM work, or how it work in tandem with something that's completely different.
I think it's doable, I just don't believe it's LLM, and I don't think anyone now knows what it is.
That is not what the education system does. That's an obvious distortion of reality. People train over billions of documents to statistically predict the next word to gain and understanding of language. LLMs do this statistical processing in order to mimic humans natural language learning ability. And there has been continued evidence of the limitations of this approach to accurately mimic the totality of human cognition.
Yours is a “God of the gaps” argument. You will remain technically correct (the best kind of correct!) long after the statistical machine has subsumed your practical argument, context limit and all.
I fall into the "pessimistic heavy user" camp, I burn thousands of $ worth of SOTA tokens monthly but it just makes me more acutely aware of the limitation and amount of work I need to do to work around them and what kind of decision that I should reserve to myself instead of trusting the LLMs to do.
I can give you the exact mathematical formula used to statistically optimize the output of a neural network from input examples. Can you do the same for the brain?
To some degree yes, in practice, not so much. In practice you have to be in the world, talk to people, know how to talk to people, know how to listen, and be able to understand the difference between what people say and what they actually need. Not want. Need.
This was something I learnt in my very first job in the 1980s. I worked for someone who did industrial automation beyond PLCs and suchlike. He spent 6 months working in the company. On the factory floor, in the logistics department, in procurement, in accounting and even shadowing the board. Then he delivered a proposal for how to restructure the parts of the company, change manufacturing processes, and show how logistics and procurement could be optimized if you saw them as two parts in a bigger dance.
He redesigned the company so that it could a) be automated, and b) leverage automation to increase the efficiency of several parts of the business. THEN, he started planning how to write the software (this was the 80s after all), and then we started implementing it.
Now think about what went into this. For instance we changed a lot of what happened on the factory floor. Because my boss had actually worked it. So he knew what pain points existed. Pain points even the factory workers didn't know how to address because they didn't know that they could be addressed.
I was naive. I thought this was how everyone approached "software projects". People generally don't. But it did teach me that the job isn't writing code. It is reasoning about complex systems that often are not even known to those who are parts of it.
And this is for _boring_ software that requires very little creativity and mostly zero novelty. Now imagine how you do novel things.
> People knocking out Jira tickets and writing CRUD webapps will end up with their livelihood often taken away. Or bosses will just expect more output for same/less pay, with them having to use AI to keep up.
You make it sound like it is a bad thing that certain tasks become easier.
I spent a lot of time writing CRUD stuff. Because the things i really want to work on depend on them. I don't enjoy what is essentially boilerplate. Who does? If you can do the same job in 1/20 the time, then how is this a bad thing?
It is only a bad thing if writing CRUD webapps is the limit of your ability. We don't argue for banning excavators because it puts people with shovels out of work. We find more meaningful things for them to do and become more productive. New classes of work becomes low-skilled jobs.
If you have been doing software for a while, you are probably doing some subset of this. But these things are hard to articulate. It is hard to articulate because it is not something we think about. Like walking: easy for us to do, hard to program a robot to do it.
>To some degree yes, in practice, not so much. In practice you have to be in the world, talk to people, know how to talk to people, know how to listen, and be able to understand the difference between what people say and what they actually need. Not want. Need.
1 person needs to do that. The other 100 not doing that currently to begin with, but doing the AI-automatable work?
We used to say that (not long ago, even) about the code-writing part. Why do we believe that LLMs are going to stop there? Why do we think they won't soon be able to talk to people, listen, and determine what they need? I think it's mostly a cope.
If they can do those things they can effectively replace any white collar job. That’s about 45% of the workforce. Societies tend to collapse around 25-30% unemployment.
Imagine 45% of higher than average paying jobs gone.
If that happens we’ll either figure out a new economic system, or society will collapse.
Also saying robots are walking just fine is misleading for any definition of just fine that is anywhere near as good as a human.
It's important (and calming) to understand that since the Industrial Revolution started ~250 years ago, we've automated away most jobs several times over, while employment levels have stayed pretty constant.
"Automating half the jobs" is the same as "double productivity per worker".
When the doubling happens in 5 years rather than 50, it might be more disruptive, but I'm convinced we're on the verge of huge improvements in human standard of living!
What in the current state of world affairs outside of IT do you think is indicative of that potential for huge improvements in human standard of living?
If we double productivity per worker, we have twice as much wealth on average.
I know there are angry people convinced that this will all be consumed by billionaires and jews, but historically that is not at all the track record of the last 250 years, and I expect that to continue.
>If we double productivity per worker, we have twice as much wealth on average.
That's not true. There are other factors at play such as demand.
If we make the average IT worker twice as productive, that doesn't mean now every IT worker is being paid twice as much, because most users aren't going to care if there are twice as many options on the app store, or twice as many bug fixes per release.
Consumption in a society will always be roughly equal to production.
There are differences due to import/export balance, investments, government borrowing etc, but as a first approximation, if GDP increases by 10%, consumption will rise by a similar amount.
About your IT worker example:
Let's say s/he produces $150k/year in value and is paid $140k. If AI makes them produce $300k of value, they may not automatically get a raise. But it becomes very attractive for another employer to hire them for $200k or $250k, or even $280k.
In the medium/long term, I don't see why wages would keep proportional to produced value.
If you are going to bring up history you should really look into what it took to redistribute wealth from oligarchs in the past.
The fact that oligarchy now has more resources than ever in the history of humankind, a means to mass surveillance and generating mass propaganda, those wealth redistributions are looking much MUCH harder to accomplish.
Yea, historically it will inevitably happen. Realistically it will be after the new version of fuedalism and dark ages. So strap in for the next 400 years aren’t looking too good
Look at how the billionaires are talking about AI: Their clear, unambiguous goal is basically to replace all white collar "knowledge" jobs. And there's currently nothing regulatory that's stopping them--they just need to wait for the state of the art to improve. Once AI is "good enough" if it ever is, they won't even think twice about 45% unemployment. What are we unemployed workers going to do about it? There's no effective labor organization left. Workers have basically no political power or seat at the table. We're not going to get violent--the police/military are already owned by the billionaire class. We're just going to eventually become economically irrelevant and die off.
> We're just going to eventually become economically irrelevant and die off.
As harsh it may sound, it seems rather likely to me. It is not like s/w engineers have helped struggling workers in other sectors other than sanctimonious "Learn to code" advice. So software folks can't expect any solidarity or help from others.
The fundamental issue isn't unemployment due to automation, but the fact that society cannot benefit from unemployment.
It should be something for us to celebrate, because it means greater freedom for humans to pursue something else rather than spending time doing drudgery.
Put it another way, the issue is that resources are not shared more equitably. This is especially egregious considering that LLMs are trained on all human knowledge. We've all been contributing to this enterprise, and what we may end up getting in return is unemployment.
45% of folks sitting on their hands are going to have the free time to talk, and this group of people are skilled at organization. Are you planning on throwing your hands up and passively accepting whatever comes your way?
And at least in the US they have >45% of all the small arms weaponry. There is no bunker strong enough nor private army big enough if 100M people come for you.
They're probably be betting that the technology they will need to defend their bunkers, think autonomous kill-bots or whatever, will emerge before people start to riot.
Or they're planning to build an Elysium-like colony in the ocean or space, to keep the billionaire class far from danger.
I get that it is popular to hate billionaires these days, but realistically, they did not get to be billionaires by being stupid. It runs directly counter to their own interests to induce anything like 45% unemployment. They will get poorer, the world they live in right along with the rest of us will get noticeably shittier, etc.
More likely they figure out what to do with a bunch of idle talent. Or the coming generation of trillionaires will.
"Poorer?" That implies they still rely on money. But money is just a form of fiat power. The billionaires/trillionaires will not much need fiat currency or fiat power, because they're building real power.. by extension, they won't really need to worry very much about the rest of us, either..
We never noticed how easy the code writing part had already become because it happened slowly. Through mechanical means, through the ability to re-use code, and through code generation.
Heck, even long before LLMs about 10% to 30% of my code was already automatically generated. By tooling, by IDLs and by my editor just being able to infer what my most likely input would be.
> We have robots walking just fine now, by the way.
I don't think you got the point I was trying to make.
True, but I guess I see a distinction between scaffolded/templated boilerplate or autocomplete and actual application logic. People have generated boilerplate from templates for ages, as you say. RoR maybe a pretty good example, but there wasn't even early-days AI involved in doing that.
>> We used to say that (not long ago, even) about the code-writing part. Why do we believe that LLMs are going to stop there? Why do we think they won't soon be able to talk to people, listen, and determine what they need?
Because they are currently "generative AI" meaning... autocomplete. They generate stuff but fall down at thinking and problem solving. There is talk of "reasoning models" but I think that's just clever meta-programming with LLMs. I can't say AI won't take that next step, but I think it will take another breakthrough on the order of transformers or attention. Companies are currently too busy exploiting the local maxima of LLMs.
Walking was given as an example of "hard to program a robot to do it" by GP. Well, now we have robots that can walk.
What evidence is there that LLMs have hit a ceiling at being able to do things like talk to users or stakeholders to elicit requirements? Using LLMs to help with design and architecture decisions is already a pretty common example that people give.
Something like five to ten years ago, when AI hype was starting to hit media, one of the claims was that AI would come for middle-management first. Since middle-management can generally be described as collecting information from underlings and reporting information to upper management, their work was supposed to be easy to automate with AI. As far as I can tell, this hasn't proven to be true at all, and we software engineers proudly wrote ourselves out of work by constantly publishing our source code and discussing it openly.
No, I never believed in fully automated tale by Tesla, but as the LLMs improve my personal estimate for the date of human-level AGI is rapidly moving to "present". Before GPT-2 I had it somewhere in 2100, at GPT-2 I thought maybe by 2060 if we are lucky. Now I think it is 2035 or maybe even sooner.
I like to see the optimism, even if I don't share it. I think it's incredible hubris that humans think we are about to reinvent our own level of intelligence, just because we made a machine that talks pretty.
Your own comment in my timeline is 7 years out of date. GPT-2 talked pretty, that was its whole thing. If you are trying to claim there's no difference between 5.5 and 2 you are delusional (hallucinating?).
I think I was fairly clear, I said that I think it is hubris to think what we have created is anything even slightly like human intelligence. It talks very pretty (a lot of work has gone into this aspect in particular), and it does demonstrate the extent to which, as individuals, most of us do not have especially unique thoughts nor problems to solve. It exposes how quickly humans jump to anthropomorphizing pretty much anything.
Is it a handy tool? Yep! I use it every day. But it is laughable to think this is the path to AGI. The most common counterargument on HN is some variation of "but you can't prove that this isn't just like how a human thinks". A conspiracy theory at best, just reinforcing the fact that we know very little about how even simple non-human brains function.
You do you. I stick to the simplest reasonable definitions. From my perspective we are already in AGI, just the intelligence isn't quite on human level yet across the board.
I am yet to see anyone saying it's just like human, so it looks like you are mostly hallucinating that too.
You didn't address my point on GPT-2 vs 5.5. Your only relevant claim is that 5.5 talks very pretty vs 2 just pretty I assume. Well, you have to be blind to claim this is the main difference.
I agree with the statement and think a lot of people miss this, but I also wonder how many people probably don't care for good, they only care for 'good enough'.
There are also those for whom that percentage is higher, let’s say 6-50%.
> I understand things and then apply my ability to formulate solutions
The AI is coming for that too.
You might just be lucky to be in circumstances that value your contributions or an industry or domain that isn’t well represented in the training data, or problem spaces too complex for AI. Not everyone is, not even the majority of devs.
People knocking out Jira tickets and writing CRUD webapps will end up with their livelihood often taken away. Or bosses will just expect more output for same/less pay, with them having to use AI to keep up.