Ask HN: GPT4 Broke Me
I don't know where to start this. So let's just jump right into it.
I can't tell what will happen immediately next year. This has never happened to me before professionally. I was born a nerd, started programming early, and jobs just came to me.
AI scares me for job security. My entire pitch to get to top %5 of salary band has been that I have many years of experience in my specialty. Thus securing me high roles in my field along with high salary.
I have been already working nearly at a burn-out level for many years to reach this comp.
Rent is sky high. I don't feel comfortable taking on mortgage despite having multiples of needed deposit, because I am not sure how much longer I can maintain my comp. with AI automating everything.
Many people share the comforting stories, that there will be other jobs for engineers/programmers.
How am I supposed to retain my TC if I have to switch to another field from web?
I am unable to enjoy any content, movie, tv show anymore. Even the SciFi from last year, feels outdated. Reddit comments feel like they're all AI-generated. I made an AI reddit bot myself. None noticed.
I use ChatGPT on a daily basis to build projects, and the more I use it, the more scared I get. It is just too powerful. The more I use it, the less proud I feel for my output. It just feels like anyone could do it.
Where are we really going? Can we just stop the optimistic techie talk and accept UBI is not happening... They don't give you healthcare, do you think they will give you CASH like that???
I hope this doesn't get flagged, because I really need your inputs.
182 comments
[ 4.8 ms ] story [ 217 ms ] threadSelf-help Singh has the right attitude for you: https://www.youtube.com/watch?v=YHxwY3Fz2gU
I imagine it's like this for many people. And, personally, I'm trying not to think/dwell on it, and simply move on with life and projects and work. Whatever happens, I know for a fact I don't have control over it.
The seed of disruption has now been planted, so we have to wait and see in which direction it grows. But I think a big wave will eventually come that will displace a lot of jobs in one big swoop.
a human is still very necessary in the loop. there is no loop without the human. currently, there should be no loop without one. cost prohibitive if nothing else.
the better the instructions, the better the output. something must be done with the output, and it has to be done safely and consistently.
there will be new challenges. there will be new careers. (lol no prompt engineering is not it.) there will be new and old jobs.
content not so compelling? take a walk. touch a hand tool, or a crochet needle. escape the screen. blame the content before yourself. was it ever that good? or just good enough? have you already read the good stuff? maybe nothing good as of late. maybe chatgpt is simply causing you to question what makes content… good.
spend a little less. save a little more. your compensation is likely “a little high,” and may suffer a little bit. keep solving problems and it isn’t yet an existential crisis.
where are we going? too many variables, someone will successfully guess, plenty will unsuccessfully. somewhere new yet familiar.
if any of this wasn’t useful, didn’t resonate, or didn’t even make sense, throw it out. just like one should do with language model output.
Of course that's like telling someone who is depressed, to "just be happy".
This is great. It means you have more financial security than most in the event of a worst case scenario and the financial aspect appears to be underpinning a lot of your concerns so give yourself some credit for that.
The most valuable antidote to the concept of AI replacing skilled workers is depth of experience. Before your job becomes obsolete, the jobs of every engineer with half your experience need to become obsolete and we're a very long way away from that happening.
If you've reached the top 5% in your field, I'm sure you will be able to do this.
If AI is giving you anxiety, maybe take a break from reading and thinking about it for a while.
It sounds like you may also need to find meaning outside of work. You mentioned you are burned out and it sounds like a large part of your identity and self worth are tied to you work. This has helped you be successful, but it can be unhealthy for you as a person, a human being who is more than a job title.
If you have the willpower to do that I wouldn't worry that much. Maybe use the extra productivity ChatGPT gives you to learn some skills that LLM-AIs will take a long, long time to learn if ever: specialization. Industrial programming, scientific programming, etc. One of the reasons I'm not worried about AI is that I work with machines so specialized there's maybe half a dozen people on the planet with my skillset and almost none of it exists on the internet at all.
But the compensation is much lower in these fields in Europe/UK, compared to web :/
As far as everything else you've said... oof, you need a break. You seem focused on money and ego. Maybe it is time to simplify a bit, explore what else the world has to offer. Worry less about whether anyone else can do your job and more about whether or not you are enjoying your life. Make changes, have some fun. If you don't want a mortgage but have multiples of the deposit needed, buy a smaller, simpler place with cash. Then you don't have rent or a mortgage.
I'm curious, as I see quite a few people saying this. You might not be worried about GPT4, but aren't you at least a little concerned about GPT8 or whatever?
Just having a post like this a mere 5 years ago would've been unthinkable yet here we are.
And I think situation with self driving cars and LLM is different. For self driving car you need it to be at least 99.99% good to be useful and initial investment is high.
For LLM is enough to be just 90% good and it already scales to millions of inferences at the same time. Investment for user is either free or 20$ per month.
One thing I will add is that many people are considering it a given that chatGPT will keep progressing at the same rate
Even if it didn't, the capacity for a developer to learn all the frameworks just got much much greater, which is a bad thing for developer salaries.
But 20 years of experience vs. some schmuck who is right out of the coding bootcamp, vs. guy with gpt4 prompt.
<toughchoice.jpg>
(of course it depends on the situation, but the point stands)
I'm at the 20 year mark as well, in terms of developing software professionally. I've always felt like with new technology, I could grok at a high level how things worked. But LLMs like GPT seem like magic and I went through stages of initial astonishment -> despair realizing the potential impact it would have on the industry -> acceptance.
While I still feel uncertainty and fear about the future, as others have echoed, I'm realizing it's a tool for developers to use. We can either choose to accept it and understand how to work with it, or reject it. The things GPT can generate amazes me, but I'm finding that it's a good starting point or reference to build on... not a final solution. It will generate things that are sometimes completely wrong, and it's your own experience and judgement that has to be used to determine that. GPT cannot do that... at least not yet.
I think back 20 years ago and remember reading through a lot of physical books with occasional web searches landing on experts-exchange or random forums. Then came Stack Overflow and that became in invaluable tool, along with the ubiquity of free tutorials on YouTube and elsewhere. And now we have GPT which I'll ask if I really get stuck on something and it gives me new ideas to try. Perhaps in the near future, GPT is the tool that I'll use first.
I found this podcast episode helpful for me to process what I’ve felt: [Lex Fridman Podcast #376 – Stephen Wolfram: ChatGPT and the Nature of Truth, Reality & Computation][1].
It's an unsettling feeling (in general) to feel like a foundation you've built and live on could potentially be made quickly irrelevant. I'd like to say I have words of wisdom to get rid of that feeling but I don't. What has helped me is to acknowledge these feelings as valid, and then try to get clarity in what direction to move. It's not the foundation itself that's important per se, but it's the skills you've acquired in building the foundation that's more important.
[1]: https://lexfridman.com/stephen-wolfram-4/
This is exactly how I feel. I felt so out of my depth looking at the ML architectures and I could not make any sense of it. I thought perhaps, they get inspired by neuroscience for the layers etc.
But a friend who works on LLMs mentioned, the architecture of large ML models, are mostly experimentally discovered, not designed. If that's the case, that's even worse... it means an entire field which perhaps could replace me in future, doesn't even have a knowledge foundation for its breakthroughs, but just goes by experiment... I thought it was only the weights inside the model that evolves, not the architecture itself.
Which body of knowledge do I study then, and is it even engineering anymore? That's something else, which I am not sure if my programming experience applies.
The amount of GPU/Capital it takes to evolve such architectures, run such experiments has to be prohibitively expensive.
If you held a gun to my head and asked me to tell you even at a sky-high architectural level (let alone in any detail) how ChatGPT worked, well... tell my family I love them. This is the first time in my 20+ year career I have felt like some computing thing is total unexplainable black magic.
Its slow erosion of our responsibilities. If AI can do some of your work with minimal supervision then sooner or later managers will figure it out and reduce you job scope. Get an intern to do it.
We are certainly not at level where you can talk to chatGTP and give it requirements to generate code, but who knows where we will be in 5,10 years.
As a reason for my 'doomerism' consider digital artists industry right now.
I can imagine that if you are concept artist at game studio, you are probably seriously worried. AI will not replace all artists - you need specific and consistent art assets to be created, AI doesnt understand fingers etc - but some of the workload can be done by pretty much anyone. Or have artists take AI generated images and touch it up.
Was a compiled language a "slow erosion of our responsibilities" compared to writing assembler?
Was writing in a garbage-collected language a "slow erosion of our responsibilities" compared to manual memory management?
The better tools let us do more, faster. They let us waste less time on the trivial, and spend more time on figuring out how to actually build what we were trying to build. They didn't reduce the need for programmers - far from it.
GPT will probably be the same. It's a force multiplier. You can write more in less time. That will make people want software that they couldn't dream of before, because it was too expensive to build. Net programmer employment will probably go up, not down.
No.
Quite the reverse.
AI is a super intern.
Both super productive, and super clueless.
It's the interns (and possibly their managers) who are at risk.
AI is a liability in any area where you can't afford to slip up.
Not even a good one imo, I ask it to cite its sources and it makes up the URLs 95% of the time.
Part of using a tool well is understanding what it is good for and what it is not. If you are .looking for citable references... or even full factual accuracy, it is the wrong tool.
yep that's where I'm at too. If you hand my boss Chat GPT and Copilot and tell him "okay there you go, make a website" - you're gonna come back the next day to find a mess of completely disconnected chunks of code which maybe kinda sorta work on their own, but haven't been tied together at all into any kind of viable thing. You'd have better luck sitting him down with Squarespace.
Get something by Disney included in the training data by mistake and just watch what happens.
Your best hope is LLM’s in the current state aren’t enough, and we need to wait for another algorithmic leap or a lot more computational power improvements before we get to human level.
Other things that are soon to come - trade secrets no longer being trade secrets because company staff are feeding internal documents and source code into a cloud model (a practice that doesn't really align with "reasonable efforts to keep something secret"). Or even just the basic question of, who gets the blame when something generated by an LLM causes a significant problem? It doesn't take much in the way of judgments or settlements before an org realizes the risk of un- or under- reviewed output simply isn't worth it.
You need to be more concerned about boring, usual tech industry ageism doing you in than an AI taking your job.
- when the code the model spat out was wrong, how did you fix it? Did you identify it was wrong before you ran it?
- what level of complexity was there in the code, in terms of "business logic" or complexity of the requirements you fed into it?
I ask these two questions because I am not sure an AI will get to the level of experience you have in the near future in _generating_ complex applications, let alone being able to reflect on why its own creations are wrong, fixing them, deploying some output, and then explaining the changes?
A human is going to be in the loop in these cases for a long time, and I'm assuming part or all of your decades of experience has been spent understanding quite how poor people are at explaining their requirements. What if you thought of generative AI as a tool you can learn to utilise to do your job more effectively?
My productivity definitely went up at least by a factor of 5 by a safe estimate.
This is amazing, all good, yeah. But doesnt that mean companies need 5x fewer people?
Over the course of a few years, dentists began using 3D scanners to digitally scan patient impressions, which allowed them to email the scans to my dad's dental lab (previously he'd have to drive to each dentist and pick it up in person).
Then, 1-2 years later, they started emailing the 3D scan to a company in China who would do the same work as my dad, and then mail back the finished product directly to the dentist from China. And of course the Chinese dental labs did this at half the price my dad was charging.
He went out of business and ended up retiring early at 55. The "retiring early" explanation was a great one, but the real reason he retired was innovation hit his industry and made it easy for dentists to outsource crown/bridge manufacturing to China. He works for a property maintenance company now for $20/hr, mostly as a way to fill his time (he's doing fine financially).
Most of the comments here are telling you that you'll be ok, and I hope you will be. But the reality is innovation causes disruption in many fields/industries. If you're in the path of disruption, you have to be willing to quickly adapt and learn to live alongside it rather than fight it.
Also remember you're not alone. Technology has been displacing jobs for decades. Life is about learning and adapting. As long as you commit to adapting quickly, learning new skills as necessary and being open to different types of jobs, you should be fine.
That seems to be a contradiction of your story. Your dad was lucky to be able to retire. He did not adapt. He had to quit and he got low end job.
Its easy to say adapt re-skill, innovate but to a lot of people it might not be an option.
It's probably more of an example of how not to handle this sort of situation. Instead of being afraid of or avoiding the technology coming for your job, embrace it and figure out how to thrive with/alongside it.
My first takeaway is if the dentists didn’t give the Dad an opportunity to compete on price and went straight to China. One day you’re getting orders and the next day you are not. Tough situation. A pivot would need to be done quickly. Once the dad caught wind of the China situation, maybe he could outsource his stuff to China as well and lower his prices. Is there any margin left though?
Same thing with GPT. One day you’re getting contracts for work and the next day you’re not. You later find out they are using GPT.
One day people are picking cotton and the next day a machine is doing it.
The OP‘s concern is real. One good aspect is that everyone is aware GPT is here. The time is now to pivot or adapt. Those that wait to find out what will happen are usually at a disadvantage to those that act more quickly.
Here is a tip: take no platitudes, and change your life in a way that works around your worst fears. The healthcare bit you can solve by emigrating to a country with free healthcare, for example. If you have money for a deposit in the place where "rent is sky high," maybe you should consider a different place to live, maybe even a rural location.
Nobody really knows how things are going to develop. But if our brains have gotten us this far, maybe our brains will get us out of a potential future crisis caused by AIs. Just hedge your risks.
The vast majority of programming is about facilitating, automating and improving business processes. The program artifact only has value as a tool to enable some business processes.
A website facilitates customer communication.
A complete webshop/eCommerce site facilitates selling products; discovery, ordering, invoice generation, logistics, reporting, returns, feedback etc.
The problem is charging for (mostly) time spent implementing the system, rather than the value added by understanding the desired business processes and drawing up an architecture for the system.
It is similar to outsourcing development - if you can solve the hard problem of gathering requirements and designing the data model and system architecture - you might be able to successfully outsource the less valuable parts - and still end up with a decent solution.
Now, with LLMs, a single system architect/senior developer might be able to do the work of a five person consultancy alone. You might not be able to charge five times as much - but perhaps three times for the same or fewer hours worked?
Ed: The money that pays for the software system still comes out of the value added for the customer buying the system. It doesn't really matter how long it takes to build - you need some senior resources to do the "hard part" - and the value added for the customer is the same - they get to stay in/improve their business.
I recently had ChatGPT "build" me a simple ruby graphql hello world-app. That was useful to me since I could easily read and understand the 20 or so lines of ruby, recognized the imported gems etc.
It wouldn't have helped the people in my company that I build software for one with. Many of them can create complex spreadsheets to solve certain problems - but they can't go from a sample api to something connected to a database and a web front-end. And especially not a system with continuous delivery, continuous improvement (features/bugfixes) or something remotely secure.
This is the point that underscores why the OP is stressing out, and it really is underselling the value GPT adds.
Ignore a consultancy. Consider a team of 5 you might be on. Esp. for greenfields, consider that a team of 1 or 2 could probably be as or more productive.
I'm working on a startup... I was shocked at how much progress GPT afforded me to build out a solution in a day, that likely would have taken over a week of research. It took like 30 iterations to get to a working solution. Unlike a mid-level consultant, I don't ask it to do a thing and it gets back to me the next day... it gets back to me within a minute. Rinse/repeat, incredible progress. Better than pairing with another senior dev, which would still likely take 2-3 days.
Boom, less engineers needed to produce novel products or features, more engineers on the market, depressed wages.
What's absurd is not connecting those dots. It's pretty basic. A business is always looking to increase margins, and with tech, it's almost entirely in human capital. Maybe some will want to move at 2-3x the speed, that's fair. Probably only in good times though.
In short, it doesn't replace all developers... it needs humans guiding it to work, absolutely. But it can certainly replace teammates, and that's the issue if you're working for a company, esp. one that is publicly traded.
Oh yeah, I have 24 years of experience myself professionally, been programming since '82.
Possibly. But software is funny - increased software productivity tends increase demand for software. Perhaps the job market will be saturated, but perhaps the market will simply grow.
In my experience most CRUD apps are ultra-trivial for the most part in a language with sane ecosystem, you literally just glue together things. That’s one day either with ChatGPT or without - the real benefit are the ecosystem (as per Brooks), that’s not the bottleneck (besides the occasional “this doesn’t work together with that because..”, to which chatgpt is just as susceptible if not more)
Also, why is there some notion it can only do simple things? My understanding is it's trained on a large portion of existing opensource work beyond just SO posts and the like. It seems too many folks saw it produce code w/ some bug or hallucination. The correct response to this is not "welp, my job is safe". It's to feed that error back in have it correct what is broken. If it builds the wrong thing, explain what is wrong, provide multi-shot examples.
You could conceivably create an agent that takes the generated output w/ tests, runs them locally, then feeds the error back in until there is fully working output. Right now the context window limit would be an issue compared to using the web interface. Perhaps OpenAI or Github will provide that capability as a feature.
It's still going to be subject to all the constraints of all ML techniques. Namely, it's really only capable of interpolating across its training set, not extrapolating from it. And it's going to fall on it's face whenever the distribution of the input data doesn't mirror the distribution of the training data. If it doesn't have lots of close-enough analogs to what you're trying to do in its training set such that the "right" code is tangibly represented in the distribution, you can feed errors back into it for infinity and it will never give you working code.
This might be controversial, but my belief is that if you find ChatGPT to be really good at doing your work for you, your work is likely closely mirroring a lot of code that exists publicly on the internet. Which would beg the question, is what you're doing not easily handled by just importing an existing library or something? Your application sounds relatively complex/niche, so maybe you are actually still doing a whole lot of higher level engineering and distilling your work down into trivial tasks that the model can handle.
While it did create quite a few bugs and a couple random hallucinations... and sometimes it would update some piece of code without telling me other parts of the solution were updated (ie. choosing to change protobuf def without being asked to), it was able to fix everything even if I needed to give it a bit of help.
I don't think there is anything I'm doing that is truly novel in isolation, and it's not a lot of code (just a few hundred lines), but I doubt there's anything quite like it in totality out there.
For something completely novel, or as someone with little experience, it’s going to be harder to get that kind of result in terms of understanding the requirements, writing the prompt, and comprehending the output.
I'm not a doctor, but I've suffered with mental health issues in the past. You strike me as someone in the midst of clinical depression and anxiety, and (relating to the idea that nothing is real in particular) like you might have early signs of psychosis.
Please get help. It is not the big deal you think it is: SSRIs, therapy, and lifestyle changes (relax a bit!) will help you see this. Things will be OK, but you're clearly struggling right now. Making a post here was a good idea, but now it is time to take stock and consult a professional or two.
Good luck, and look after yourself.
Your total comp might go down or it might go up. Every time you make a career move that’s the risk you take. But if you’re healthy and smart and diligent enough to be highly paid today you are in a better position than 95% of your countrymen and 99.9% of the rest of the world, no matter what happens.
Enjoy the time you have on this planet. Don’t let fear of the unknown get in the way of that.
Truck drivers will get automated away eventually, but not today and not next year. Self-driving systems get better every year. It’s been 30 years. We’re getting close but we’re still not there yet.
I expect the LLM revolution to be somewhat similar. The models will get better but they won’t have the kind of superintelligence that makes meat humans obsolete. Someday, perhaps. But GPT 5 will be an incremental improvement, and GPT 6 moreso.
To be honest the far greater threat to your job is an increase in the ammount of programmers plus an increase in individual productivity, with an industry declining in growth. There is your threat.
That productivity increase by a level of magnitude is from AI tools or ChatGpt. The field is gonna get a whole lot more competitive, the salaries will suffer as well but maybe it will be for the better routing jobs to other fields other than programming.
What Im worring most is increasing inequality and the class who will wield most power from AI
I am convinced that ChatGPT works multiplicatively. You need to ask the right questions and be able to quickly understand the output. Those skills come from experienced developers.
ChatGPT, if anything, will greatly reduce the amount of banalities as those are obviously easier to automate. The people writing banalities now are the ones which the least experience and skills.
The more senior you are the more broader you control the code, deciding on abstractions, interfaces and features. This is obviously something ChatGPT is horrible at and giving someone with 6 months of experience and ChatGPT the ability to decide on that will lead to disaster.
the same field who has the most imposter syndrome rants, also has most people who think they can't be replaced by someone with less experience equipped with AI.
Everyone is talking about the junior, mid-tier programmer who will replaced, but not themselves. It just sounds like willful ignorance.
Because in a shrinking market experiece + AI is vastly superior to little experience + AI.
Experience is becoming more important, since experience is required for the tasks AI is very bad at. AI is drastically increasing the barrier to entry, since it can do many of the things entry level people are/were useful for, but it can't do any of the things experienced people get payed for.
ChatGPT can not tell you which API changes to your 1M SLOC code base are helpful for the future interests of your corporation. It won't talk to management about technical challenges in demanded features or which hirings are neccesarry to be on time with future projects.
Rent? Fuck that. Your life is on the line and this is for all the marbles. You find a way to go rent free. Maybe do like the bums are doing all over the place, live in a van, etc. Nobody seems to care. Save that money and then get out.
I did this, ending my career in IT back in 2006. Best move ever. When they fired everyone on the last contract I was on, I can't tell how good it felt.
By the way, the house I lived in at the end... Remember the house in Fight Club? That would have been an improvement.
I think you nailed it. My current fear arises from the fact, I am at one of the highest HCOL cities in the world in comparison to wages. This makes the fear more imminent, as I basically would have to move if my salary goes significantly lower.
That wouldn't be the case where I am originally from, in Scandinavia.
1. AI is a tool, not a replacement: AI is designed to assist, not replace, human tasks and decisions.
2. AI needs human supervision: AI technologies require expert human supervision for their creation, maintenance, and evolution.
3. AI creates jobs: AI often creates more jobs than it eliminates by automating routine tasks.
4. Continual learning: By continuously learning and adapting, you can ensure your skills remain relevant in the AI era.
5. Human creativity is irreplaceable: AI can't replicate human creativity, which is vital for problem-solving and innovation in programming.
6. AI ethics: Ethical considerations in AI deployment require human judgement, which AI lacks.
7. Emotional Intelligence: AI can't replace the emotional intelligence humans bring to their work, including empathy, understanding, and interpersonal communication.
Hope this helps.