Will AI Replace Most Tech Jobs

[alex123711]

AI is already pretty impressive and seems to be getting better all the time. I have seen some pretty amazing things it can do e.g be used to program apps and websites and other automation with just prompts. I have seen some people in these fields that were previously said that AI will not replace them now changing their minds. I think AI overall will be a benefit to automate all the mundane jobs/ tasks like driving etc.

How much of a threat is AI to tech (and all jobs in general) and is there anything you are doing to adapt/ overcome these challenges?

[Jean]

Ai will probably replace any job that don't require a physical body on a specific site.
It went from a slightly dumb assistant to a smarter than average human assistant in less than a year.
It wasn't able to come up with a reasonable plan for our boar hunt (and by reasonable, i mean one were we aren't very likely to kill bystanders), so it is either already favoring non human life forms, or we under estimate how complex our ability to plan group action is.
For some reason, it isn't able to produce convincing music with just prompts, like it can for illustration.
Maybe our musicality is more complex than we thought?
I'll adapt by making music, jokes, and hunting plan for AI.

[The Old Man]

I think a lot of white-collar jobs that don't require creative thinking will be prone to elimination by AI. Factories were automated a long time ago and many many jobs were eliminated. The same will now happen in the office.

[Slevin]

No, apparently most people think tech “just exists”, but there’s actually a huge amount of labor in supply chains of speccing out new servers-> ordering hardware -> building server racks -> getting the servers running /alive -> debugging issues of the servers -> finding issues in the servers -> debugging and fixing whatever broke in the infrastructure -> making sure the infrastructure and software are building + running + booting properly.

So even if the coding is owned / unkept by AI, the AI still needs taken care of / upkept. OpenAI isn’t the current winner of the AI game, Nvidia is, and everyone selling and upkeeping compute is. Nvidia helps you build AIs, and building and upkeeping all the infrastructure will be the biggest goldmine there is. AIs can’t automate that until they get physical dexterous robots, which is nowhere near right now.

I.e. be the guy assembling and fixing the mining robots in the gold rush.

Second note: my job is still mostly not coding. I get that the AI can code, that’s cool and all, but even if some autonomous text interpreter can write my PRs for me, there’s still more work for me to do than I have time for.

Also, I still solve novel problems in domains that are barely documented / undocumented on the internet. There’s still a lot of steps in between where we are now and the AI being able to completely do my coding work. And it is hard to overstate how important AI development to this point is dependent on freely available posted in good faith information in the internet… And nobody is writing out new info / documentation that is freely available to be scraped by AIs anymore. How well will AIs do training in heavily poisoned / AI slop information wells? Hard to know, but I don’t expect linear / super linear scaling from this point on with current techniques.

[zbigi]

Second note: my job is still mostly not coding. I get that the AI can code, that’s cool and all, but even if some autonomous text interpreter can write my PRs for me, there’s still more work for me to do than I have time for.

Agreed, but IMO saying that "AI can code" is too generous. If you point it to any moderately sized existing codebase and say "add feature X without breaking anything", it won't be able do it. What it can do is regurgitate pre-existing code snippets for trivial problems, which makes it a better stackoverflow.com (from which it's gathering its knowledge) - which is how professional programmers are mostly using it today, if they're using it at all.

[Slevin]

@zbigi agreed, I’m giving generous theoretical ground here in the case where AI actually gets better than I am at coding (which it is currently not).

[Henry]

So even if the coding is owned / unkept by AI, the AI still needs taken care of / upkept. OpenAI isn’t the current winner of the AI game, Nvidia is, and everyone selling and upkeeping compute is. Nvidia helps you build AIs, and building and upkeeping all the infrastructure will be the biggest goldmine there is. AIs can’t automate that until they get physical dexterous robots, which is nowhere near right now.

I disagree on the robots but I do agree with your point on the building and upkeeping of AI. The TSLA Dojo supercomputer that runs their FSD is a cluster of NVDA processors. This concept of still needing a computer, in both concept and actuality, seems to eliminate the possibility of AI completely wiping out tech jobs unless the bots are both adequately dexterous, and can not only code as well/better than humans, but can perform every aspect of creating and implementing AI better than humans. So even if AI comes up with a better idea than a supercomputer, it was still a human that had something to do with the ability of AI to come up with a better idea of a supercomputer. But I live on the side of The Matrix where humans created AI so I understand my opinion is subjective to those who live on the other side of things.

[Sclass]

I’ve already seen a lot of design automation in electronics. More recently automatic code generation for routine but painstaking work.

It made people move on if it was good. If it was bad it validated experts.

LLMs look really good for a B level college paper or pumping out a specific use case of a tired old algorithm. But for really novel or complex stuff it fails miserably. I amused myself for hours creating Star Wars fan fiction “what if” stories. Tired ideas that can be rejiggered into pseudo new content.

I remember seeing the first PCB auto routers lay out wiring and people would say “you’re gonna be out a job!” I don’t think anybody lost their job. Designs got more complex. Workloads got bigger. Look at some hand routed antique electronics from 1985 and compare it to modern tech. We just couldn’t produce the same content back then.

Same thing happened with logic synthesis. I remember a time when it was really important to learn Boolean algebra and do DeMorgan gate minimizations on paper to design digital systems. That(and more) is achieved by entering input and output requirements on a good logic synthesis package now. And we get to design something like a GPU instead of a 1980s child’s toy.

From an entrepreneur’s perspective I love the possibility of getting more automated design capabilities without hiring more skilled people. The less people I hire the less management and the less investment. So I guess that would be bad for some guy who just wants to lay down routine design for somebody else over and over and never pick up new skills. For me the democratization of engineering via computer is a gift. I got to be my own software, hardware and mechanical engineer and then I got to lay myself off after my product went to manufacturing and sales. Why would I want to stay on the design treadmill till death?

I think it’s all going to work out. There may be some shuffling around while we find out what these machines can and cannot do. But like times before things are always changing.

[Henry]

As AI grows, so does its power requirements. I'm not in tech but it seems to center around connectivity. Repurposing and connecting a decommissioned nuclear power plant to Mag 7 database center seems like an endeavor that would involve a lot of tech jobs.

[flying_pan]

AI will definitely streamline and improve efficiency in a lot of cases. What took you several hours of reading through documentation and experimentation, now can be done in like 30 minutes from a conversation with AI, and it will probably only improve. So it can eliminate a lot of jobs IF the market reached the saturation and that efficiency can only be used to reduce costs.

I have a feeling just that workloads will increase, and that configuration might be even more complicated now. Maybe we'll get a point where it will be somewhat competent, but it will need overseeing for some time, and we'll see in another 5 years.

[Scott 2]

The generative AI progress we've seen over the past few years, is brute forcing of a relatively simple model. It has a ceiling and IMO we are rapidly approaching it. The costs are currently hidden, as firms develop market share.

As they shift into the profit phase, we're going to learn the true price. Hint - it is substantially higher than $10/month for GitHub copilot. Organizations also have unlimited appetite for technology. The ability to work at higher level of abstractions only makes even more ambitious projects feasible.

Tech continues growing, fueled by AI eats the world. Tech jobs are augmented by AI tools, but remain a highly lucrative path. Possibly one of the least likely knowledge professions to be compressed by AI.

[Henry]

I am aware of three tech companies receiving economic benefits that can be directly correlated to NVDA build out: TSLA, META, CRM. PLTR and AVGO have also increased significantly but I do not know enough to draw the same line.

[AnalyticalEngine]

As someone who works in tech and whose job got replaced, no, AI is not a threat to tech jobs. What is a threat to tech jobs is what happened to me: offshoring. Which, incidentally, once you account for the expensive electricity cost in training and running these models, paying low wages in other countries might still be cheaper than trying to get AI to code (which it cannot do well when you account for maintaining the large and complex domain knowledge that is a developer's actual job).

I'll also add I was trying to use it to give me questions and answers for tech interview questions while looking for a new job, and it actually kept largely giving me factually incorrect answers.

[Sclass]

Our educational system may be quite vulnerable to being picked off by AI. We’ve kind of set ourselves up by making a system that creates low performance meat based robots.

I feel that we have taken intelligent humans and depleted their potential by forcing them to pursue exam results and useless certifications in an attempt to making them into poor facsimiles of robots. Real robots may be better that humans dumbed down to be poor robots no matter how well certified or validated they are.

In particular academia may be under threat. You take a young, intelligent and impressionable human and program them with a lot of well documented data. Then they rejigger it and replay it for money till retirement. This sounds a lot like a crappy version of AI if you take out passion, creativity and vision.

I think if you act like a robot and try to make yourself into a robot be ready to be picked off by robots. The key is don’t make yourself into a machine that competes with the new technology . It’s like using your fingernail as a screwdriver. Stupid people do that with their human capital. Those who want to live that way should prepare themselves for heartbreak.

There is a small group of people plotting the control and exploitation of workers with new machines. I think a group has always been angling for this since humanity first picked up a tool. Bottom line if you optimize yourself into a stupid tool you’re going to get played by people who develop better and cheaper tools to enrich themselves. The whole digital revolution has been a progression of this and AI is going to be a big chapter in the story.

A programmer who survives on low value code development that can be generated by an LLM will have their lunch eaten. As bad as that sounds it likely isn’t too hard to do higher value work and create more value than the machine.

I’m having this flashback to grade school where we had races at the chalkboard on long division problems. It was a traumatic experience brought on by a brain dead fifth grade math teacher. She’d write the divisor and dividend on the board and I’d have to race another kid on long division. I sucked at it. There was this amazing girl who just speed ran the problems and solved them 100x faster. The class would roar with laughter as she blew past me and the teacher would give this approving smile. I’d be asked to stop working and sit down even though I hadn’t finished which curtailed my learning process. The girl was a machine. I looked her up and what do you know…she’s a grade school math teacher in exactly the same district of Los Angeles (different campus). I read a review of her on Rate My Teacher and it said she was good at explaining the arithmetic procedures. it’s kind of ridiculous when you think about it. A human robot perpetuating the educational process to make more human robots. Useless ones in the age of computers with Goldschmidt divider circuits. I’d like to see my old classmate race one of those. An example of a human soon to be picked off by a machine.

Want to beat the machine? Do something more valuable. Like with the invention of a sewing machine you don’t want to be the person who can just sew two pieces of cloth together. You have to make something wonderful by sewing cloth together. There is still a job for those creative people. Design a garment and set up a sweatshop. I think that’s why a small group of humans is going hog wild investing in AI tools now. There is always a monkey who eats the other monkeys’ lunches.

[ducknald_don]

A programmer who survives on low value code development that can be generated by an LLM will have their lunch eaten. As bad as that sounds it likely isn’t too hard to do higher value work and create more value than the machine.

I don't know, back when I was doing contract work it was almost universal that in a team of ten people you would find two that were doing nearly all the work, seven or eight barely producing anything and sometimes someone who had a negative impact on the project.

Those seven or eight people just cruising are the ones at risk. Ironically they seem the most enthusiastic about AI. Probably because it can lift their mediocre output.

However one thing I have noticed is that as technology advances we seem to respond by asking more of it. Perhaps the AI revolution will just result in us creating ten times as much code. Jevon's paradox strikes again.

[jacob]

I don't know, back when I was doing contract work it was almost universal that in a team of ten people you would find two that were doing nearly all the work, seven or eight barely producing anything and sometimes someone who had a negative impact on the project.

This has also been my experience which is why I try to avoid "lets all work together to ..."-situations as much as possible. It will be interesting to see if AI can come up with a better solution to this "human problem" cf. the mechanistic "up or out"-hierarchies that humans have been able to come up with. After all, AI did come up with a better method for multiplying matrices and it also makes some smooth chess and go moves. Maybe these something equivalent in terms of organizational allocation. For a fun example, Amazon workers (meat robots) are already being sent around their warehouse by AI in ways that make no apparent sense to the individual worker.

[zbigi]

Those seven or eight people just cruising are the ones at risk. Ironically they seem the most enthusiastic about AI. Probably because it can lift their mediocre output.

I don't think they're particularly threatened by AI. These people are low performers, meaning they have low output per unit of time, but they can still do things that AI can't do. And, if the company cared about their slowness, they'd be kicked out a long time ago. Most managers I spoke to seem to think "if I could hire five high performers, I would - but I can't find five of them, so I'll take the one I could find, and four mediocres - and the team will do more vs if I only hired the single high performer". In other words, the (perceived) margin on tech work is so high, that even hiring duds is seen as profitable.

On the other hand, in industries with low margins, e.g. gamedev, management works people hard, anyone's low performance is immediately apparent and people get fired for low performance. So, the biggest threat to slow coders is an economic downturn which will make companies budgets tighter - just like in any other line of work (this has in fact already happened around 2023, with tech companies laying off thousands of programmers).

[Sclass]

@ducknald oh yes. I recall those people. We called them NOPs - instruction codes that took up code space but didn’t do anything. The negative work was a thing.

Not quite what I was thinking about…though yeah those idiots seem to be everywhere in 90/10 systems.

I was actually referring to people who did functional work but work that could be done with a machine due to its simple nature. Repetitive simple tasks seem to get automated out first.

In the spirit of ERE I personally would want to be the worker who thrives on change or the one who is FI to opt out of employment. A software engineer who acts like a US steel worker is a sad thing.

[Henry]

I'm not a historian of technology, but as I can recall in my lifetime, Bitcoin seemed to be the first piece of technology made available to the general public that created a troubled reaction similar to a significant advancement in military technology. You have your first telephonic mouth breather, you develop a movie rating system, Lucy had to keep one leg off the bed, they slapped some stickers on music, they said don't put a combustible engine in the back of the car, a child friendly google search. Really the technological equivalents of don't look at the sun too long. AI seems to be accompanied by a much deeper existential worry. Every day, planes are blowing the fuck up all over the place, but people keep flying. But when a Tesla bangs a shopping cart, or a machine can write a better English paper than a stoned, first year, community college student, it's the apocalypse. I think of my grandfather, who lived through the depression, never getting past grade school, toiling away his life in a hardware store, re-entering a world where a robot can give him a hand job and it makes me weepy.

[Ego]

I am not a programmer, so I have zero knowledge in this field, but I have been following https://www.gauntletai.com/ and I find their process interesting.

Those seven or eight people just cruising are the ones at risk. Ironically they seem the most enthusiastic about AI. Probably because it can lift their mediocre output.

Gauntlet is offering a $200k job to anyone who completes their 12-week training program. It seems like the program is designed to eliminate the cruisers. Yesterday was the first day. Four hours after the first lesson, one of the participants used their AI tools to get a full-stack multi-channel chat application up and running.

1. Does this mean that one highly skilled person can do the work of ten or a hundred?
2. As the tools improve and creates composable templates, is it possible that a slackers using it could get good enough to do 95% of the high-skilled person?