Well, I learned how to program back in the 70s and 80s, so a lot of these newfangled programming concepts are tricky for me to implement (list comprehensions, lambda functions, functional programming, all of that). Claude has no such limitations, so it does things quite efficiently from what I can tell. Even when it’s putting together some Excel functions, it uses all of those cool new Excel features (lambda functions, other neat array features) that I’m only slowly getting familiar with.
Maybe 6 months ago, but the gap continues to close. Goes back to what I was saying about everyone thinking they’ll be the last one standing, that AI might take the tedious jobs or the entry level jobs but surely my job requires my unique human skills. In some industries we’re past that point, in every other industries it’s just a matter of time.
We all want to picture ourselves as Geordi having the ship’s computer run complex diagnostics and then handing over the results to him for his clever analysis. That’s just not what’s happening now.
I’ll ask again for anyone who cares to answer. I have a senior in high school. He wants to go to college. What should he study?
I think things that require a human touch will continue to be valuable – sales, nursing, doctors (especially surgeons). Or, just ask Claude what roles will continue to be valuable in the future – dusting and sweeping data centers?
Construction management. Electrical engineering. Welding.
This is what it comes down to for me. To use a car analogy, the executives are at the “trolly car to private roads” point of adoption where they’re rapidly tearing up the existing public infrastructure to force everyone into their private profit models. This doesn’t mean the concept of a car replacing a horse is a bad one, but we have the right to decide if that’s the overall balance of public good that we want in the future.
I’ll bite. Why is he going to college?
- He wants a high-paying job
- He wants a robust education
- He wants to build his life in a specific location
- He is passionately interested in a particular area (arts, sciences) or field (film, literature)
The answer is going to depend on the mix of all of the above. If it’s just that he wants a job, go to a community college to get the first couple of years out of the way, and then transfer to whatever school gives him the best networking opportunities for the industry he’s interested in.
My main objection to AI are the overwhelming majority of people I work with who use AI without sufficient knowledge of the subject matter to know when their AI generated output is bullshit.
The struggle is real. Another reason why I’m not overwhelmed with the general knowledge of LLMs.
What I’m seeing is that people trust the AI implicitly, without actually you know, reading what it produces.
Some may be getting out of their lanes like you describe, and creating documents they weren’t qualified to produce in the first place, much less actually thoughtfully vet for content accuracy and relevance.
Others I suspect struggle a lot to just read and write at the level a lot of jobs require, and producing a coherent, organized document is a massive step up for them, and they don’t have the motivation to vet them for content.
It’s a problem; I used Copilot to help make me a vendor guide, since we require a set of technical artifacts for technology purchases, and vendors are frequently confused/unclear about what we’re asking for.
Copilot made a fine skeleton, but it does need a fair amount of tweaking- it’s got a sort of too corporate-speak kind of writing style for my comfort. But it was great for actually organizing what I was trying to ask into a coherent document.
The thing is, a whole lot of people would just be perfectly fine with releasing it as-is, and that’s what I think we’re seeing more and more. The AI output is “good enough”, and that’s what people go with.
Nope, AT&T. But as far as I can tell IBM consulting had been a big part of their profits for years, in the form of contracts to help integrate their hardware into the workflow. It did increase later, true.
Probably depends on the problem class and also the specific algorithms. It probably knows all the sorting algorithms and can pick a good-enough one, and often the best-known one, for any particular use case. If you asked it to solve P=NP it would probably struggle, though I think it’s also helping with some recent research in that area (I don’t remember details, sorry). In between, I would say it’s generally 80-90% better than me (a working programmer) at most tasks, but once in a while I can still give it a useful direction to make something more efficient. Its outputs are never really deterministic, so the same problem and prompt run 10,000 times will probably give you something like 3000 similar results and and maybe a few hundred truly unique ones, some of which will be better than others.
Part of the job of the harnesses, classical testing, etc. is to help rein in some of this model-level uncertainty. Human review too, but that’s less and less frequent, and only for the most critical parts of the code.
(Just one programmer’s experience. But it’s certainly “good enough” now that none of the programmers on my team code by themselves without AI help anymore. I think it’s like 90% AI and 10% human input at this point (the inverse of what it was maybe 2 years ago).
The common refrain in our industry (programming) is “woodworking”.
That, or he network like crazy and try to get into one of the big AI firms in time for their IPO, work for a few years, then become generationally wealthy.
People mention all these positive things first. I’d also say law enforcement.
Some AI experts say there will hardly be any jobs left to do. Some say it will have no effect on unemployment. My prediction is that predictions of future job demand will prove to be inaccurate.
If he already had a subject he loves, you wouldn’t be asking, right?
If so, it is a perennial question long predating AI. If they allow it, I would try some subjects he didn’t study in high school. I remember long ago being at a college visitation with one of my kids where they had presentations on different majors, and geography was sold as a good one, with varied career options, for students who didn’t like what they first tried.
Maybe the best advice is to always go to office hours, and chat about career paths there, not that teenagers listen to parents.
Ooh, bold prediction! ![]()
I don’t begrudge people using the tool the way you describe. I do the same. It often provides a useful starting point and it’s great at reformatting/re-writing and generating info-graphics for punching up presentations.
It often serves me well in my technical IT duties.
I don’t hate the technology itself.
Most sales are already being done by algorithms right now anyway – it’s why every single social media app exists, after all. In person sales, sure, but that’s always been a cutthroat industry and will only get worse as other jobs dry up.
Doctors in many fields are rapidly becoming “the human in the loop”, approving AI diagnoses and treatments. Surgeons will be replaced by robots at some point but may be one of the last holdouts of the human era, but that’s sort of an elite field anyway.
Nursing already has a high barrier to entry and requires a considerable amount of hours relative to earnings, but may be one of the few things we can recommend for the next 5-10 years.
Management and engineering jobs are going to fall to AI quickly. Robots have been taking welding jobs for decades and I think the pace is only going to increase as AI/Vision based welding systems become affordable.
Option E, he just wants a job that won’t get taken over by AI so that he can pay his bills. If we’re already conceding that the sort of middle-class white-collar jobs that people typically went to college for aren’t going to be there when he graduates, then, well… oof. I mean, jobs that pay well but aren’t “high-paying,” jobs that can be done in generally any location, jobs that don’t require being a starving artist…
The common refrain in our industry (programming) is “woodworking”.
Yeah but programmers only say that as an excuse to buy 10s of thousands of dollars in woodworking equipment without their wives complaining too much.
If he already had a subject he loves, you wouldn’t be asking, right?
He does! But I can’t look at him with a straight face and tell him he’ll get paid for it. And yes, that’s always been a problem with artistic kids who we have to be honest with. But now it feels like we have to treat every degree like a potential art history degree.
Aside: This does not answer the immediate question, but if he loves art, let him major in art. And something else that might actually get him a job. No I don’t know what that is. Sorry. I just think a double major is a good option for creative types who also want to find reliable work.
Management and engineering jobs are going to fall to AI quickly. Robots have been taking welding jobs for decades and I think the pace is only going to increase as AI/Vision based welding systems become affordable.
Dr.Drake:
steronz:
Construction management is (I think) different in that you have to be on site at least some of the time and it is very client-facing. People building $3000/ft homes are less interested in saving 2% than having their hands held. Also, things physically go wrong all the time in construction and need to be dealt with in person.
Automated welding will certainly be optimized with AI, but field welding? Is it really going to happen that bipedal robots take these jobs?
For what it’s worth, he wants to get a job in the agricultural sciences. Two years ago that felt like a mature direction with lots of applications in major industries. Now…
He’s going to keep going in that direction because a) momentum and b) we’re doing well enough that we can support him indefinitely if we need to. But I don’t envy anyone who doesn’t have a safety net right now.
Eventually, yes; at first a bipedal robot will walk around a construction site recording video, an AI system will distill what it finds into recommended actions, and there will be a human in the loop. My prediction, at least.
On-site welding can be done with a robotic arm with cameras attached and someone making peanuts to roll it over to the application. Eventually it will roll itself over. I don’t know that it will ever be bipedal because it doesn’t need to be, but maybe?
eta: For now, the AI on-site supervisors are not-bipedal. Seriously, anything you can think of, some AI startup is on it.