Tell HN: I stopped caring about personal development in tech after seeing GPT-4
As a C++/Networking enthusiast religiously following any new features and quirks in the language and also following several rfcs I have lost complete motivation for self improvement in terms of keeping my skills sharp due to the rapid AI takeover. I had an opportunity to see gpt4 in action on an inhouse product and I'm taken aback, it architects, generates tickets and starts writing code using feature / bug / spike branches for an embedded device a company is working on.
It can do almost everything I can do a bit better and I have years and years of domain knowledge, keep ontop of rfc changes, new languages, c++ standards etc, side projects and even occasional leetcode.
Oh well, this gold run did run on long enough - Im glad I made a bit of money from the industry but I think all these students going into CS are in for a rude awakening and we're in for a huge shift in this industry.
177 comments
[ 2.4 ms ] story [ 234 ms ] threadWill we ever get to the point where it does everything independently? No idea. But right now your reaction is premature.
I'm a Pascal programmer, not a Python programmer... but I'm hoping that I can leverage CoPilot to help me navigate the nitty gritty boilerplate that would otherwise take days to sort through, and get to the heart of the refactoring/patching necessary to get WikidPad up to date and fix the breakage.
I see GPT4 and kin as tools to allow more freedom of action, and less worry about the stuff I always hated anyway, the minutiae of coding.
--
>years of domain knowledge
Usually the term "domain knowledge" applies to real world non-programming knowledge such as chemistry, manufacturing, etc. This is the first time I've seen it applied to programming. Programming is just a means to an end. I've never considered programming to be an industry. We produce a product with zero marginal cost.
I suspect you are in the same emotional place that accountants were, the first time they saw spreadsheets in use. It must have seemed like the end of the world to them, but it wasn't.
>all these students going into CS are in for a rude awakening
As long as they know that computers are a tool, not the end result, they'll be fine
[1] https://wikidpad.sourceforge.net/
C++ versus Pascal mindset. :)
> This is the first time I've seen it applied to programming. Programming is just a means to an end. I've never considered programming to be an industry.
Yes and no. It did not replace the good accountants who actually orchestrated the whole department, but it got rid of a lot of the low-level grunt work.
AI-supported coding autopilots seem to go in the same direction: The junior devs whose whole job it is to translate architectural and design specifications into excessively verbose boilerplate code will struggle to survive, but the software architects above them will find a new means through which to express the analytical thinking, process planning and understanding of complex dependencies that they're paid for.
Artists didn't stop because the camera made it possible to do what otherwise took hours with a canvas and oils.
The nature of programming may change, but it's still about using computers to solve problems.
--- Putting on the old guy hat ---
You kids didn't learn programming from magazine articles typing in your favorite little game into a computer that went away when you powered it off.
You didn't have to toggle in a boot loader before you could use the computer.
You didn't dream of one day having a modem and being able to call BBSs, then dream of your own phone line that you didn't have to share.
Yet, ya'll turned out ok, despite all those changes.
It'll be ok, kid... it'll be OK.
Computer programs will always be complex beasties with bugs hiding in the corners. There will always be a class of people willing to find those bugs and make things easier to use. You are that type of person, right? Good!
I personally do very well for myself, being in the top 1% of earners in my generation. It deeply concerns me that I can barely afford to buy a house despite this fact. Most of my peers can't even dream of doing something like buying a house right now. And why is this happening? Because computers are being used to extract vast amounts of value from the system to be captured by a tiny fraction of people who are increasingly owning everything. AI will only accelerate this trend.
And perhaps I'm one of the lucky ones to have carved out a good career for myself, but everywhere I look, I see folks struggling more and more. And stuff like AI is about to make life a whole lot harder for them. Yuval Harari talked about the rise of the "useless class" [0], and the challenges with mass unemployment in the future when AI has replaced many jobs.
You can repeat the "it'll be OK" mantra as much as you want, that doesn't make it true.
0: https://www.youtube.com/watch?v=94o-9zR2bew
This was at the beginning of the software revolution. Growth of the industry was huge and fast. Entrants into the field were virtually guaranteed a decent career with little competition and almost no automation.
Those days are over. We are dealing with a relatively mature industry now with a new ai tool that is limited by hardware processing power instead of more biological beings. Low level / beginning programming skills will no longer be needed and experienced programmers will evolve into roles that will drive the ai tools at a high level to create solutions, occasionally debug tricky issues that the ai created, and implement code that the ai is unable to generate properly. The last task is the one that will determine the remaining role of the human developer.
And so far it only increased the demand for software, because for anyone outside software development, cheaper software just means you can leverage your budget to get better faster.
My guess is ai’s will largely replace the benefits of rote memorization learned in these bootcamps and make these skills less marketable to companies.
I actually feel my university education prepared me well for what’s going on. I studied a mix of theoretical comp science, physics, philosophy, and mathematics (esp calculus and number theory). Very little of what I learned is obsoleted, and especially probability theory and NLP are turning out to be very valuable to know when working with even CoPilot.
But man, I feel bad for those studying “programming” as a trade rather than computer science as a scientific / engineering discipline.
Domain knowledge about specific industries and what is required by the software seems the more valuable skillset going forward. How does physics and computer science help with that? Communication skills and ability to learn new, more unstructured domains is likely to be more critical like those learned by strong liberal arts disciplines, for example.
Which will, in turn, further drive shortages of good developer talent long term. The industry thinks there's a tech labor shortage now? Just wait until the ladder has been completely kicked over for young folks trying to find their career footing in tech.
We were all once juniors, doing work that could primarily/completely be replaced by AI. We had to do that grunt work to learn the lessons that make us architects today.
The number of non-STEM folks who make it into high-level tech, though certainly greater than zero, is still trivial in comparison to those who go that way early.
I figured this was going to happen long before ChatGPT, but sadly I failed to find an good exit, because well… most everything else that doesn’t suck is already like that and the golden handcuffs were too tight.
GPT-4 can do all of that stuff. Probably as well or better than you. If you think you still are better, can you do it literally for 24 hours a day? How about 3-6 months from now, are you going to become 50-500% faster/better at your job?
I would love to see a few samples of some real world "architecting" that you have done in the last few months along with the calendar schedule you did that on and then compare you head-to-head with GPT4 and a junior dev. And then we show your executives how that went. Sound good?
They're hardly given a chance to do so when all they're getting is tickets to turn design X or API spec Y into code with no allowance given to make it better. Yes, that's not the workflow in all orgs, but there's a lot of organisations where most people aren't expected or allowed to be creative until they're far up the foodchain. And those will be the first to lay off what they consider meatbag machines. Think Oracle, Accenture, IBM, etc.
With regards to your personal attacks, have a nicer day than you seem to have had until now.
So far, I find it a pretty extreme statement, but it appears true here in the Netherlands. Most employees that I talk to at various industries can immediately point out one or two things which they would like to see automated. In most cases, software would replace data transfers which now occur via spreadsheets or paper.
But in my experience, there's some hard to solve bottlenecks for those sorts of organizations that still insist on largely paper-based workflows:
1. Their workflows are such a byzantine mess that either nobody understands them sufficiently to explain them (be it to a programmer, or to a no-code platform, or to an AI), or they're fundamentally broken and only fudged along by people who shouldn't be allowed to do their job, if the process was implemented "correctly"… or both.
2. It takes a special breed of people who have the necessary analytical skills to really pull off architecting automated workflows that work in practice. It comes natural to some managers and consultants, a lot of programmers, and a bunch of others, but it's not a widespread skill, and those individuals know their worth. People who already can't conceptualize the underlying problems won't be able to do so with the help of AI any time soon.
3. It still takes budget to implement such workflows, AI or not, and the affected orgs usually don't want to spend any money on improving themselves.
Personally, I think we will do more and more complicated things instead of just being done with programming.
Try a few times and ask for more complex example.
This is the problem IMO. The model needs to somehow learn that out of its entire training set, the single sentence in the AWS docs saying not to use the assumed role ARN takes precedence over any patterns it may have learnt elsewhere in this specific situation.
Yes, GPT-4 came up with an incorrect answer, but it's an incorrect answer an experienced programmer could legitimately have come up with, and one they probably would have come up with before actually testing their code against the AWS endpoints. GPT-4 sometimes gets hard questions wrong. GPT-3 and GPT-3.5 make up nonsense.
If a coworker told you GPT-4's answer, you'd say they were wrong but you wouldn't say they were hallucinating. If a co-worker gave you GPT-3 or GPT-3.5's answer you'd definitely doubt their sanity.
Agreed. GPT-3 and GPT-3.5 commonly hallucinate. GPT-4 can certainly be made to behave badly, but on real questions I've put to GPT-4 it has a 0% hallucination rate. The few wrong answers it has given have been "sensibly wrong" in that it's highly likely an experienced human programmer would have made the same mistake (eg lots of Stack Overflow answers are wrong in the same way), and even its wrong answers have been helpful in guiding me towards the correct solution.
These occasional, "sensibly wrong" GPT-4 answers are fundamentally different from the correctly formatted academic bibliography citations for technical papers that never existed, by authors that never existed, in journals that never existed hallucinated "answers" I've received from GPT-3 and GPT-3.5.
A non-code example: Some days ago I asked it about "Searle's Wall" [0]. It gave me a mashup of the correct description and the Chinese Room experiment. So it clearly had the right answer somewhere in its data, but it mixed it up with the much more famous thought experiment.
[0]: https://www.researchgate.net/publication/260138925_Searle's_...
> Me: how to only run data "archive_file" if a path exists?
> GPT4: <blah blah blah> add: depends_on = [fileexists("/path/to/file")]
This is nonsense. Terraform tells me:
> A single static variable reference is required: only attribute access and indexing with constant keys. No calculations, function calls, template expressions, etc are allowed here
I just get this rubbish all too often to be afraid for my job.
There will still be a source of income in fixing the AI generated code. It won't be as fun though.
And maintenance by the way.
Interesting. I agree it is a relatively decent boilerplate generator, but it is quite useless for anything else I've tried. How did you measure its performance?
Obviously, ChatGPT can't do this today, but I don't see any fundamental reason AI won't be able to do this in the decade or so.
Even in this context I don't know that there won't be demand for humans that have the mental rigor to program computers. I do think we will need to adapt.
This abstraction, compiled away, into machine code that does one thing. Compilers also tend to take all kinds of shortcuts to make programs a lot faster, and therefore changing them might not be as straightforward as you think.
If you think I'm wrong, I'd like to see what "fundamental reasons" you've considered and how you've reasoned that they aren't an issue for this sort of system.
Abstractions like assembly code or C or Python are for the benefit of us humans. We can't reason about even fairly trivial x86 binaries because our minds aren't designed for that much complexity. An AI will have a completely different set of limitations.
This isn't going to happen tomorrow, but you'd be foolish to bet against it happening within the next decade. This advanced, but it isn't any more "magic bullshit" than ChatGPT or Stable Diffusion.
People expecting that their jobs never change would have a rough time surviving for decades in IT anyway. Automate all the drudgery away, new drudgery always comes back in its place.
This requires subject matter experts, like yourself, to use and implement.
These LLM are tools. They are not sentient.
However, as far as agency, these language models have none.
It's the same as IKEA, it's not as good quality as a handcrafted table from the 1800s. But it works well for most situations and most people.
A calculator didn't make mathematicians obsolete. It aided in the creation of more complex mathematicians.
My mom was a computer in a bank. That was the name of her job.
Society evolves just at the available jobs do, due to technology.
> If you are an expert in some area and suddenly everyone else catches up almost, it will surely impact your wages.
Find something that compels you and don't be a wage slave. That used to be what programmers did.
If all the things that compel you don't make money, you should introspect instead of thinking about the glory days of what programmers did.
The OP is not going to be jobless anytime soon if they have the skills that they say they do. Hell. Someone with less knowledge who just leans on a LLM isn’t as capable. HUMAN experience is valuable.
As an analogy, people in robot (I use the term in a loose way for machines) assisted warehouse find the work far worse than one without robots, because the job becomes soulless and centred around the robots, making it much less fulfilling.
I go back to what I said before. If your fulfillment in life is centered around work, you need to re-evaluate. If the OP has no motivation to improve themselves for the sake of improving themselves, that's an OP problem that existed WAY before LLMs. Whether it's weightlifting, programing, or basketweaving, the OP needs to find some benchmarks for life that come from self motivation.
That being said, the OP will continue to be employed. Rather than miring in the existential crisis of "not being able to continue operating in the manner they do today", the OP needs to adapt and align their skills with this new tool. LLMs are tools.
Is this everyone's experience?
To those saying it will enable them to solve more problems, yes that is correct. It will give everyone "wings", but once everyone has wings the industry will be so different in terms of wage and employment.
To people saying GPT gives incorrect code, please try GPT4.
If your age and circumstance allow, you should think whether a career change is possible. Not a hard change right now, but atleast explore what options might be available. I am exploring the same myself.
To those talking of chess, that is not a correct comparison since people want to watch (and connect with) human players playing chess (thus the pro scene survives), and play it for their own joy. Due to tools like stockfish, it has become far easier for people to explore moves. If the aim in chess was to finish more and more games from random given positions, and people were paid per game (and some value was created finishing it), stockfish would easily drive it to 0. Chess survives not because humans do better than AI, but because nobody is interested in playing against AI or watching Stockfish v Stockfish (By nobody I mean a very small number). Most people want to play against real people and watch real people play.
user: crop_rotation
created: 4 days ago
karma: 305
I’ve been doing this for years, and I have the least confidence I’ve ever had that I’ll ever pull it off. Credentials are expensive as hell and time consuming to obtain, and if massive wage loss is a concern, there aren’t many careers you’ll be safe starting over in. Anything “interesting”, but low paying will likely be affected in the same way software is.
Meanwhile Stable Diffusion managed to motivate me more than anything to learn drawing. I always gave up in the past because it takes so much practice to get good results. Now I can draw something, throw it into Stable Diffusion as input (the only way to semi-reliably get what I want) and get a more satisfying result, and it's still bad/inconsistent enough that I'm motivated to do it better.
TL;DR: I think your job is safe.
> What you're seeing now is as good as it gets (in terms of big breakthroughs, there will still be lots of small and medium ones).
Unless you have a strong source for this, I find it hard to believe. Also, GPT3 didn't have many big breakthroughs over GPT2, other than massive parameter size.
> Unless you have a strong source for this, I find it hard to believe
Like I said, this is just my opinion. I do have a world-expert level understanding of the technology, but at the same time I'm a strong believer that even experts are bad at predicting the future, so make of this what you will. Also, my impression of what constitutes a breakthrough and what doesn't might vary a lot from yours.
Personal development should be about what you like to do. I like solving problems and building the solution myself, for which I need to learn new things, which is also enjoyable.
There were always other c++ devs out there, why did you choose to do it anyway?
I spent 2 hours trying to fix an SSL issue on my webserver, neither stack overflow or my devops friends could figure it out.
GPT4 gave me the answer in 3 minutes. and without all the ego and gatekeeping.