Ask HN: Year old CS student. Should I continue?
The plan 6 months ago was - get good grades, go to a good uni, get a CS degree while doing programming to pay for uni/do a degree apprenticeship, and go into a CS career. Without LLMs messing the world up, I think it's a pretty good plan. I'm a good programmer for my age and I'm academic.
However, with recent developments, that seems borderline foolish. I'm interested in physics and engineering, I think I could enjoy a career in either, and I've still got time to pivot to either. I might be able to write better code than GPT-10 - might - but there would surely be increased competition. I also just don't find asking a LLM in the "correct" way and having it spit out code trained on millions of other programmers interesting in the slightest. I worry that actually knowing how to write code without a LLM will go the way of knowing assembly and having intimate knowledge of a computers internals.
I'm frankly tired of hearing about LLMs, but I know that I can't just pretend they don't exist. I have less than a year to make a decision on which uni courses I apply to and I don't know whether I should abandon a sinking ship or hope it's alright and it stops here, as an aid but not something which ruins programming.
I've seen people saying "But we'll still need to translate client requirements to prompts" - 1. I doubt that'll pay well, and while that's not my primary motivation it is something to consider before committing myself, 2. I'm not interested in being a human-bullshit to AI-bullshit translator.
And yes, there's more to CS than programming, but not in a way that pays the bills. If I pivoted I'd still write code "The old fashioned way" for fun.
So: What do I do? I'd appreciate some advice.
Thanks :)
99 comments
[ 2.8 ms ] story [ 155 ms ] threadI personally experienced the paranoia of the "end of programming " when VisuslBasic was first released, and when UML code generation ('member Rational Rose?) Tools arrived.
I'm still doing a good living with my programming skills.
oh boy I do. Had a whole course in engineering school on it. Never touched it since.
LLMs will just be another programming tool to help us be more productive, at most.
Different subject, but in some ways, I think what would flip the bit for me in believing in machine-learned code-working is systems that can do really great refactoring things on instruction. Taking some human instruction & spitting out some html or js doesn't really impress me, to be honest. But if machine learning can take my code, create a builder pattern based off the current constructor, make a config file that allow programmatic defaults, that's a huge time saver of manual labor, not at generating boilerplate, but at helping me shift around what I have. Rational Rose & it's ilk were never particularly great at helping folks refactor systems, but ideologically they suggested that good code had graphical representation & that there was a bi-directional relationship; bringing out that promise doesn't take machine learning, but if machine learning got good at helping juggle code, could open the door to those ideas again, that would be powerful.
Maybe actually say something, next time? There's literally nothing to agree or disagree with in your content-free position.
Ok, latter, bye!
I encourage you to try this in ChatGPT. Give it some code and ask it to refactor out a builder pattern. It won't be perfect, but if it's like anything I've tried, what it gives you will save you some time, and may offer you an insight or two.
---
On an entirely unrelated note, I also encourage you to keep your frustration in check -- an aggressive response to a perceived contentless message says more about you. Equally effective without being condescending or aggressive would have been "Could you say more?"
I chose to interpret your message in this way; others may not be so kind.
disagree, thanks again. if you just show up to blindly dunk on my actual-contentful talking without saying a single thing, it's you whose being aggressive. i'm fine with aggression too, but i expect it to actually present a form & not just live in shadows. like you did. ultimately i think i contributed to reason, shared perspective, and if i ruffle feathers in negative people who won't build their own stances, i have zero bones about it.
could i do more to coax them into saying something useful? perhaps. should i? meh. apologies, but vacuousness is it's own clear counter-argument & i don't lament pointing that out.
there's some legitemacy to telling me to fuck around and find out. frankly i'm not really sure how i'd get access, and i dont think these systems can actually understand & rewrite my code at all, right now, can't read a github repo. so it seems like an impossible ask to even try to do the very basic long-term co-partnership i'd asked after, at all. and again i don't really have access, as far as i can tell, cause they're all fairly guarded inaccessible experiments. i really don't think we'd find much of hope or promise, if we were somehow able to get these systems to try to start working on our existing code-bases. does anyone have any starting place at all to start re-believing against my objections?
I see some of the people involved in that are heavily dunking on GPTs.
While they have a valid point not to get carried away and ascribe capabilities that aren't really there, it seems to me GPTs have perhaps already, in a few months, had more positive impact on code and coding than those methodologies and tools have, ever.
But so far, to me, these are all cool amazing demos, but none of them at all mirror the real challenges of working significant & interesting systems. There's a real chance that means significant job loss. Maybe this is a "I never thought the leopards would eat my face" mistake, and it has some ego to it, but I personally cant imagine wanting to do any of the work these systems have done.
Code tends to be kind of long journeys, iterative, and we've seen from ML what, so far to me, look mostly like boilerplate-generators and modest-tweakers. I cant imagine spending months building product with one, it being a coworker. We're in a new day, but I've been preaching a book that meant so much to me, After the Internet: Alien Intelligence (2001), which is pretty wild & course & miscellaneous & not super gripping, but sold an interesting core message to me of expert intelligences that are in their own realm. That have incredible powers, but which are unintelligble & vast & distant. Right now these coursely trained ML systems have a lot of embedded expertise, but whether they can show up again and again and become a useful ongoing part of workflow, can work with existing systems & ideas, is very much unknown & tbd.
So, yes, I agree:
> it will be a very long time until there is no need, or even declining need for computer science majors.
Think broader.
Right now our only interface to these powerful models is prompts, but this is an artifact of how we discovered them.
Future you will not be (paraphrasing) "finding the right prompt to have an LLM spit out the right code" nor "translating client requirements to prompts".
Instead you'll be letting an LLM fill in the basic details of code, syntax, systems, and propose algorithmic approaches that you pick and choose from, while you think bigger and write programs with with substantially more complexity and capability.
How this could become a less-remunerative activity I have no idea!
I fear that this is true, but that the complexity will be accidental rather than incidental. Humans are capable of understanding architecture and design patterns - as far as I've seen so far, Copilot and the like do not.
You program at the level of architecture and design patterns, and Copilot and the like fill in the details. Do you need to keep a watchful eye on the details? Of course.
The complexity will either be necessary for the great things we build, or unnecessary but time-saving, or just bad all around -- but that's true as software grows independent of LLMs.
(And my wrists will appreciate it :) )
Decide based on what interests you most.
You could also take some classes in AI in addition to the run of the mill CS stuff.
GTP3 was a breakthrough. We’ll figure out how to integrate this level of AI into our lives, and there will be incremental improvement, but who knows when the next breakthrough will be. What you’re predicting, with GTP10 being able to write large amounts of code is just not possible with incremental improvements. GTP has no internal representation of the world, and so it struggles with novel problem solving that would be trivial for a human.
I suspect once you know how the sausage is made, you'll understand how short sighted this question is.
Also if you want to make the most money, get the heck out of CS and into finance.
https://twitter.com/ID_AA_Carmack/status/1637087219591659520
nothing stops you from enhoying both industries though.
Now, if you want to go to CS just to get a programming job at some company and you are doing this for money, I would indeed ask you to reevaluate. If you like programming as a hobby, a good way to kill the passion is to turn it into a boring 9/5 job. Physics and engineering may use programming as a tool and that may be more interesting to you?
That said, LLMs will not make a dent anytime soon. Yes they will plagiarize code and spit it out in a way that's similar to what you have requested – and for this very reason, have been banned at my Fortune company (Intellectual property is tricky). But they have no reasoning capabilities, there's still demand for solving actual problems. These days, writing code is a fraction of what I do, a lot of work goes into syncing with other teams and figuring out _what to write_. When I actually sit down to write code, that usually takes just a few days. That pales in comparison to all the architecture and coordination work that had to happen to reach that point.
Even if we assume that computers will write the code for us and it's no problem, that would only save a few days for most of my tasks. And any task where it would save more than that probably isn't worth my salary at all.
This might heavily affect pure programming outsourcing companies that only write code and require detailed specifications. That might be a true concern for them. Everything else will just adapt to the use of a new tool.
Like you mentioned, few people program in assembly these days (although some still tweak and plenty more read assembly) but I don't really see anyone complaining about that.
Somehow I think having no computers on the planet might affect the funding for her CS research.
> Why are you interested in CS in the first place? Computer _Science_ will not be affected by LLMs and other technologies at all.
I haven't really had a chance to try out pure CS. In my own time I work on practical applications (though a current project involves lots of boolean algebra), and the curriculum at school is mostly things I already know (Sorting algorithms, and we'll do big-O at some point). The practical part of the school curriculum is heavily-OOP C#.
> Now, if you want to go to CS just to get a programming job at some company and you are doing this for money, I would indeed ask you to reevaluate. If you like programming as a hobby, a good way to kill the passion is to turn it into a boring 9/5 job. Physics and engineering may use programming as a tool and that may be more interesting to you?
I'm certainly not doing it for the money - I genuinely love it - but when I'm a 17 year old making pragmatic decisions about my future, "Can I get a job" is a pretty important one. I hope that programming as a job doesn't kill my passion, but that's certainly a concern. How frequent is that?
I'd be interested in programming within other industries, but I'm not sure on a path to doing that. I've had a lot of comments suggesting I take physics/engineering courses at uni in addition, but I'm not really sure how much of an option that is in the UK. I'll ask.
> That said, LLMs will not make a dent anytime soon. Yes they will plagiarize code and spit it out in a way that's similar to what you have requested – and for this very reason, have been banned at my Fortune company (Intellectual property is tricky). But they have no reasoning capabilities, there's still demand for solving actual problems. These days, writing code is a fraction of what I do, a lot of work goes into syncing with other teams and figuring out _what to write_. When I actually sit down to write code, that usually takes just a few days. That pales in comparison to all the architecture and coordination work that had to happen to reach that point.
Yeah, I did a week's work experience with Microsoft research about a year ago. I definitely got that impression - I, the work experience kid with no internal access, given some small projects they hadn't had time to do, programmed far more than the actual programmers on the team because I was in less meetings. I enjoyed the meetings and thinking about how to make the tool practically useful, but I'm not sure if that would continue into a decades-long career. I'll probably be doing it again, so I'll have another opportunity to consider that.
Part of my motivation behind going to uni is that I've had a pretty bad time at school due to bullies, violence, and a serious, ongoing concussion from an attack on me. Going to university and finding a good tech job appeals to me because it would (in theory) mean being surrounded by intelligent, motivated, decent people.
Go broad, things are changing fast it’s better to hedge your bets.
AI, computers science and robotics will completely change our world in this decade.
IMO, taking programming classes in college is a waste. It’s nearly all the type of thing you can learn on the job or in an online course. The real reason to take CS is if you’re interested in applied math, because most math departments lead you into pure theory, but this depends heavily on the school.
there's been other threads on this site where people talk alot about how much they learned in either or both their compilers or operating systems course. This rings true for me -- the value of my first few semesters of CS was that it prepared me fully to tackle a short list of really helpful courses. Honestly finishing the major wasn't all that much more helpful (in terms of what I learned), but the degree itself has worked out fine.
One last thing, but for me there was a sort of retrospective signal -- I did have a second major and other areas of study, but eventually I realized that I had always saved my computer science homework for last because i found it the most fun, and gosh, maybe that was a hint of where to build my career. So to make it more general, the other "use" of a bunch of mediocre CS coursework is that you can gauge your own interest level. We're all built differently, but honestly taking a wide variety of courses to learn what you're interested in while in your late teens / early 20s isn't such a terrible thing (if you can do it without too much debt).
my algorithms class definitely helped me pass interviews like Microsoft. I know many people learn this stuff on their own, and many people don't, but i can't call it a waste by any stretch of my imagination.
Thanks, I'll consider this.
> IMO, taking programming classes in college is a waste. It’s nearly all the type of thing you can learn on the job or in an online course. The real reason to take CS is if you’re interested in applied math, because most math departments lead you into pure theory, but this depends heavily on the school.
Yeah, I'm near to a few good unis that run what seem to be high quality courses with lots of optional courses on ML/maths/etc. I would avoid doing CS at an average uni due to its reputation.
I'm alright at maths, but I really have to work at it - it doesn't come as naturally as physics and compsci. I'm not sure I could succeed in a pure maths degree.
1. There has always been competition in CS. If not LLMs, there are international developers and bootcamp people and just regular engineers. However, good software engineers from universities are still finding good jobs. Generalists are common, specialists are hard to find and companies will pay the specialists what they are worth.
2. Modern software engineers reply on networks similar to LLMs all the time. Think StackOverFlow or google. LLMs will just be able to generate better responses that will be quicker to find.
3. If there is a LLM that can program as good as a well educated SWE, many more jobs will be replaced besides programming. I reckon a LLM will have a much easier time understanding physics, which changes a lot less than CS.
4. Don't concentrate all learning in one place! You could do a double major with physics or a minor. You could do computer engineering or you could do a business minor with CS. Nobody knows the future, we can only guess. However, in the case that CS becomes totally replaceable by LLMs, it will be good to have other skills.
5. Remember that your major does not define you! I know many people who studied physics in college and ended up in the banking industry. Many others who studied engineering ended up in a software development role. Just because you chose a major now does not mean it defines you for the rest of your life. You can always pivot careers, go to grad school, or shift departments within your company!
It is more about becoming a specialist in a subset of CS where
think you may be thinking about it
1. Programmers will always be needed for maintaining and upgrading LLMs.
If you want to make money, the picture is less clear - hopefully LLMs will reduce the demand for code monkeys which is skewing dev salaries to such absurdly high levels (though admittedly this is less of a problem in the UK).
If you want to feel some sense of pride or accomplishment in your working life, I'd advise engineering instead. I work for a FAANG company but my friends all have real jobs like "teacher" and "engineer". I'm much wealthier, but they are much happier.
For now, don't worry about it and just go to a good uni and enjoy the 4 years (definitely do an undergrad-masters course, not just a BSc). Keep an eye on the economic climate in years 3 and 4 - if you think you have it in you and no obvious better options are available, do a PhD with the intent of ducking out of academia immediately afterwards.
Good luck!
But personally, if CS is your passion and top choice, go for it and don’t doubt yourself. It’s true the field could be changing, but someone is still going to need to understand something, even if the LLMs do a lot of the lifting. And if you think programming is at risk, there are 100 other roles in the office that are much worse off. The Office 365 demo from this week has got to scare the pants off middle management and anyone who’s job is responding to emails, attending meetings, taking notes, interpreting simple data in excel and summarizing a project status in a PowerPoint deck. And by the way, that tends to be most people in most offices. Actual knowledge work is rare and usually rests on a few shoulders.
Personally, I am betting that the impact is large, but will be spread out over 15 years and create a ton of work for technical folks in the meantime.
My two cents: If you’re interested in engineering and physics, you should explore that. Does your university have a CompE or EE program? At mine a lot of the classes between different eng programs overlapped so it wasn’t difficult to switch.
But you could also just pivot because no one in the tech industry really cares if you have a CS degree, and it sounds like you’d still enjoy coding anyway so you’ll keep learning on your own. The CS curriculum is going to feel painfully old fashioned compared to what you’re learning on your own, though there are some useful fundamentals in there.
My university has a Compsci+EE program that I'm going to look into.
If you're already competent, are CS programs less time consuming, or are they easier but still as time consuming? I'm wondering how much time at uni I could spend working on difficult problems in my own time.
In my experience, in both business and academia, bullshit intolerance is a severe career limiting trait.
With your aptitude for physics and mathematics, have you considered some other field of engineering? In all fields of engineering strong programming skills are a strong form leverage for your core work. A friend of mine was in a similar situation to yours many years ago. He choose surveying and he is now the owner of a substantial company, making a comfortable living.
How very well said.
I've craved a book describing the alienation of the engineer, how strange it is to be so well connected to such self-speaking details, to be such a hyper-critical beast on the hunt for issues, a monster at scoping out problems & understanding their bounds, at evaluating possibility. And being trapped in a world where so much around us can feel like bullshit, like a removal of information, only making things vague. Where so many goals & intents seem detached & decoupled.
This simple phrase rather puts a pin on some of the difficulty of life as a developer.
I don't have many recommendations to explain or explore The Most Important Life Aspect Of Being A Programmer, which you've so very really captured imo. Ellen Ullman's Close To the Machine (`97) is a relatively-early memoire that has some very nice capturances of this weird world, of the life of being a machine-speaker, a truth-finder, apart from the rest of the adrift weird world. Haven't picked it up but her new book (Life In Code, `17) yet. I really crave a lot more discourse on this. This is the most undersung reality. Programmers & would-be/could-be programmers should know. The world should know.
I have considered other engineering fields. My school also specialises in engineering, many of my friends are studying it, and many areas of it interest me. Nuclear, electrical, and aerospace all interest me. I'd like to learn more about them, and I'm planning to talk to some of the engineering teachers about it.
This isn't the first comment saying that programming can be useful as part of another field. What would the route be into that kind of work? Would it be a normal engineering route and adding "Proficient at xyz" to your CV?
During your university education you will be required to solve many problems and some might be better done with some programming skills. You might need to use MatLab, etc for simulations. During this time you will gain experience with the sorts of problems that programming / CS / data science even AI could assist with. These days every field of engineering is a heavy user of computers and requires a wide variety of programming skills.
As for your question about what to include on your CV. I recommend the structure of: have solved xyz problem for abc using plx. That is, the CS aspect is the least important. The most important is the problems you have solved and the benefit / value generated. This is a far better way to demonstrate proficiency than listing a bunch of acronyms.
Good luck with your university experiences.
A lot of the articles you see floating around about "the end of programmers" are written by people who think software is designed by managers, which is only true in workplaces that you would not want to be in. If you can independently design and implement software, then you can work at a company that treats programmers like craftsmen, which is more rewarding both mentally and monetarily.
That said, the tech industry tends to move faster than undergrad currriculae. It's likely that any specific technologies you learn in school are already obsolete[0]. Focus on the fundamentals (CS as a subdomain of mathematics) and you'll find your knowledge has a much longer shelf life.
[0] When I was in university, the more "vocational" classes offered by the CS department tended to teach libraries and languages that had already fallen into disuse outside. Things like OpenGL immediate mode when the world had moved to shaders, or CVS when everyone used SVN and Git had been just released.
If you don't do CS what would you do?
If for some hypothetical reason you were forced to do something for the rest of your life for 8 hours a day without getting paid, what would that be?
Schmidhuber says that between one key innovation to the next, the time interval is divided by 4. We have entered the most noticeable part of the exponential growth where each year from now to 2040, you will be seeing very noticeable progress.
By 2030 your life as a software engineer will be very impacted.
People used to say that computers themselves were going to cause mass unemployment.
And, to be fair, they did end a bunch of jobs but also created more new ones so people didn’t really notice all that much. How many companies still have typing pools?