Are Programmers Obsolete? Will AI Replace Them?
ฝัง
- เผยแพร่เมื่อ 30 ก.ย. 2024
- Should you still learn software engineering in 2024 when we are on the cusp of AI being able to write all the code? Dave answers! For my book on Asperger's and ASD, check out amzn.to/3tQQtDv
If entry-level jobs disappear, how will highly skilled engineers be trained to work on AI-assisted projects?
Maybe then people just have to spend more time in school training with projects
@@ManunKanava I don't think training in school will give you the same skills as experience in commercial, real-world projects
As far as I know, every engineer gets born with +10 years experience in some very specific areas. So I don't see the problem?
Silly you, entry level jobs don't exist in this field. They all demand years of experience and specialized skills, something that by definition no entry level worker has.
I don’t think entry jobs will disappear at all, what is sure is that they will not look like what they are today. The core of a software engineer is not coding is the capability of solving problems using a particular tool, what is changing now is how that tool looks. This is not new in itself. When we moved from assembly to higher level languages something similar happened. Think about how abstracted we’re now from the actual hardware that our code runs, think about how many hours you spend writing code that anyone can write, that is the part that will disappear. As we move forward new jobs will be created that do not exist now. If you have been in the industry for some time 30+ years the job you have now more than likely didn’t exist when you started. The same will happen and continue to happen. And that is why is so important that education prepares you to understand, approach and solve complex problems and not focus only on teaching a skill, skills have a lifecycle.
Before reading: no and yes.
Junior programmers are indeed obsolete, though, but this is just a big problem because you can't have senior programmers if you don't have junior ones first
Being a junior programmer is only a state in the natural process of building up your competence and learning how to handle much more complex problems.
I'm a software engineer with 30 yrs experience of designing telecoms systems all around the world. My primary job was the technical delivery of the solutions I designed, but my secondary job was the mentoring of new developers so they could progress into my role in the future. My worry with ai is that these junior roles will not be filled by people and nobody will get trained into the senior roles
I'm already worried about this eventuality. The last decade I've tried to bring in juniors or intermediate engineers to "pay it forward" like many did for my career. Every time, all I hear from managers is, "We only hire seniors who can 'hit the ground running'. Our project is too important to bring juniors in at this time."
We're creating a bad situation and the salaries of truly senior engineering just keeps rising as they become a sort of "dying breed".
I haven't seen enough people worried about that. If I need grunt work I can use chatgpt or a junior, the results kind of the same. that's not a big deal over 2 years, over 10 years that's going to be a travesty.
I understand the concern, however it is not actually worrisome. The computer science and software engineering you've experienced was just the beginning of the field, which is now called computer science. That's right it is a science. And when it is taught like a science the results are way different. We already often see that a well educated intern can surpass by skill both AI and many seniors. What intern does not have is experience of how things are planned, actually done, what happens when things go wrong afterwards and how it is managed. Well established company processed often mitigate those flaws. Thus the current seniors have to move to somewhat management/supervisor positions, where they manage and just point people in the right directions. The progress in the last 10 years is insane. Things I have to spend couple years on to master are now taught to undergrads in a half a year. The new gen is not software developers or engineers, they are scientists not limited to a set languages or tools - they actually have their pet project code generating AI on par with GPT-4, and they can tackle complex problems. They run our insanely complex product and say something like "hey, that initialization time seems longer that it should be, here is the graph of initialization order of modules and it is not optimal, here is what it should have been, and this is because in your custom code generation module and the custom compiler this feature/optimization is missing, according to compilers theory it should be like this...". Then that intern fixes something that he has found there and wow +20% performance somewhere. Also the language is no longer important for them, yeah we want a C++ dude, but we get an intern, who knows, C++, Java, Scala, Haskell, Python, JavaScript, SQL, Idris, Coq, Scheme, Julia, the list goes on. And when we interview this guy, because well, what's written seems like a lie, turns out yeah, he actually knows that to some depth and that's what he was taught by some very smart professors. I just have to reeducate myself to stay competitive despite being senior.
"well educated intern can surpass by skill both AI and many seniors" never saw that@@linuxgaminginfullhd60fps10
@@linuxgaminginfullhd60fps10 I'm a also a 30yr dev and I was using copilot and chat GPT recently and I can say honestly it could eventually replace devs. It saved me a week of work. Wasn't rocket science code but still it was mostly useable. Eventually a secretary will be able to describe an app to it and it will be created. There may be a few who can refine the app a bit or give better prompts to gpt. I mean, no one touches 0s and 1s today. At one time it was the only way to program. Why is it so crazy that natural language will replace devs?
You didn't mention the most important difference, from a business point of view, between humans and AI: accountability. Nothing mission critical can be entrusted to an entity that might hallucinate. Therefore, anything an AI creates will have to be thoroughly analyzed and understood by a human engineer. You can't just do black-box testing and assume it works in all cases, if you don't know how it works.
Depends on how critical the component is. There are definitely pieces of code that don’t need to be understood by a human if the goal isn’t 100% accuracy. If a piece of AI generated code works 99% of the time and that is acceptable then why waste the time sending a human in to analyze it?
@@reynoldskynaston9529 Depends on the size of the catastrophe that can result in those 1% of cases.
A company that wants to cut corners should know the risks. Low quality software might harm their reputation and a security breach could destroy it.
Humans are the weak link. Have 3 different AI's analyze the same data, without talking to each other. If they all come up with the same answer, you're in the clear. If not, then have them analyze the differences. No more humans needed.
@@koristrange9655Only if you can ensure that the three AIs don't all share the same faulty training data.
I’ll probably be the guy screaming *“I have people skills!”* to the AI-powered efficiency consultants. 😂
Hopefully I'll be the guy with braided suspenders and matching braided leather belt.
I'll be that guy who doesn't even bother to show up at that meeting and says "Go ahead and fire me if you want. Let's see what happens."
i cringe so much seeing this. no.. ai don't replace competent programmers in whatever project they work in. Its all about connection and human skills to work in a group.
@@somerandomchannel382 not where the extreme majority of code is written by AI. AI doesn't care about the people skills of other AI. Personally I don't care much either just as long as the people that I interact with aren't complete assholes. But then, I'm also the kind of person who will correct that behavior immediately upon encountering too.
AI will replace everybody except
the CEO. For some unknown reason, it will be crucial that the CEO be a human.
If all entry level programmers are going to be replaced by AI, then 30 years after, there will be no human developers.
It just means that definition of entry level will change.
"current types of entry level jobs" will become something different. Like instead of rudimentary C logic classes, we'll have AI to sort that and our entry level guys will start their learning at "how to train your LLM" or "finding buggy" or other favourite Disney or DreamWorks film names hastily shoved into a pun. "Machine-Learning Inc." okay i'm done...
I have a theory on why this idea is kinda shocking for people to mentally digest, but just think of the invention of the automatic switch-board for telephone systems. Do you think that ended all those careers?
na people dont understand the problems here probalistic models don't account for deterministic problems
I've been programming professionally since 1982. Every year since there have been predictions of the end of programmers. The technologies used to predict the decline have varied over the years, but the predictions have always been overly generous about the capabilities of the technologies. Last summer, I mentored several software interns and asked them if they were afraid AI would take their jobs - they just laughed. I do agree that if your job is easily automated with upcoming technologies, you should be worried. In general, if you are employed as a "spec to code generator," then you should be trying to upgrade your skill set.
That was about when 4GLs were to put programmers out of work. That was about when my favourite programming languages were Natural (Software AG), and S/370 assembler which was largely the same as assembler on Fujitsu's OSIV/F4 (which was stolen from IBM) nd OSIV/X8 which wasn't, quite. I use assembler to interface Natural to AIM/DC.
@odabasim Of course anything is potentially possible in the future, but if this happens the current alternative jobs will be gone too anyway, so changing your course of action due to that potentially future doesn't really make sense.
@odabasim Maybe. But the same doom about AI was predicted in the late 80s. Having lived this for 40+ years, this angst feels a bit like the predictions from psychics saying the world is about to end, then postponing and postponing. If you feel strongly about AI, then you need to act. There's nothing wrong with being prepared. I'm near the end of my career, so I look at things with less urgency than you might feel is warranted. However, if I was early in my career I'd probably would be interested in the technology rather than scared.
@odabasim We were supposed to have flying cars decades ago. We've had code gens for decades. My experience is, technology adoption (not technology itself) is much, much slower than we predict. I agree, AI will work as an assistant to programmers (I've used it several times), but validating specs as a business analyst, creating and assembling an entire solution, even if sections of it are AI generated, will still be performed by a software developer for quite some time. AI code, while correct many times, does not understand the many niche business specifications. Business analysts have no interest in writing specs to generate a likely incomplete outcome. Developers will be the glue that holds specs, generated code and edge cases together.
@odabasim Nations? Chill with the movies, will ya!
Hope you feel better Dave... and yes, I think you are right on the money. The days of trivial programming are limited but the days of problem solving are here to stay. The only thing that changes are the complexity and sophistication of the problems.
Along with ride share drivers which will likely be replaced by self-driving cars in the next decade or so. Just because they did away with the horse and buggy a hundred years go, it created new industries along the way. It's called progress.
The best joke;
-Ai is writing programmes for free, companies don't need us anymore... What we do now!!??
- so costumers need to ask AI exactly what they need program to do?
-yes
-no worries mate we are safe 😂😂😂
I wonder why don't companies automate away the customers, and the workers, and the C-level suits, and the investors. Just make a lot of GPU run the stock market, isn't that what it always wanted, to be free from the real world to speculate to its "hearth".
Who needs humans anyway, is that the end game of corporations, all run by machines to make numbers go up infinitely ? instead of being a tool to satisfy the human needs...
Maybe we should close the global corporations and go back to local smaller economies of just humans. Maybe a butlerian jihad against the machines.
Every time someone invented a platform for applications without programmers.
Customer: "I want a button to do my Job"
But low code no code solutions are supposed to be easy and do everything for you. :)
Valid, customers very often don't even know what they need, let alone describe it correctly enough for an AI to do it.
From this pov, yep we safe.
I suppose it'll be a bit like when libraries and larger codebases came along; suddenly you didn't get your hands dirty flipping bits all day, you just imported a library or 2 and glued together a bunch of functions from within. Maybe the job of the dev for now will be breaking the task into sections that AI can build, then adding the glue, until the sections get large enough that no glue is required.
I feel like at that point, you become a glorified IT worker instead of a software engineer. Not to dog on what IT people do, their work is unique and takes a different form than engineering, as they implement solutions that engineers have built, for companies.
@@docmars probably the opposite will happen, the IT glue job is going to be done by the machine, meanwhile the engineering job which is problem solving can't be done by AI
I think this is the most accurate estimation of how it will play out 👍
You missed the fact that GPTs require data to be useful. Even now, ask any question about systems programming in depth, it will fail miserably. But JavaScript is written very well. And this will not improve with GPT 5, again, you need a lot of data and niche questions will fail miserably until AI is AGI.
That's 90% of JS dev work.... If course then you end up with massively bloated apps and need to trim down. ;-)
My experience of working in the industry is that people still can't write good specs-at least for "applications" that they want. I'm not talking about "take this number and shift it left" or whatever, but for high level apps, people always miss out what to them is obvious, but which they don't realise is specialised domain specific knowledge. So the functionality isn't as expected but they are genuinely surprised.
So I can see the new market Dave describes for "the AI operator and input refiner".
Yeah, a huge part of being a successful software dev is the ability to interpret what clients or managers "ask for" vs. "what they're *really* saying." Generative AI is not creative or insightful enough to do this.
Yes! Many customers that we deal with often struggle with their mother tongue, let alone actual algorithms and such 🤪
@@internetuser8922We will always need newer ideas generated by the best generator - the human brain. AI is about to replace all gap fill jobs. Like it or not, IT and computer programming has been gap fill with the majority of time spent debugging, which is an inhuman task. America has a totally broken supply chain and is about to starve yet they are graduating more coders than ever. What gives? Start creating a real job for yourself people...
People think AI can program, but writing code isn't programming, specially when the AI is basically just smushing things together from the internet. Writing a B-tree algorithm isn't programming. Programming is coming with the B-tree algorithm given the needs of the problem you're trying to solve. After that exists on the internet, AI is basically a latent space search algorithm.
People need to stop being afraid of AI like it could actually think. It doesn't think, and probably will never be able to think like a human, so they're basically useless for decision making, and we should have ethics and laws to prohibit using them instead of humans, even if they could reason logically (which they can't because they're basically Turing machines, that's what a deep neural network can do, and the halting problem applies to them but doesn't to biological intelligence, like humans).
@@monad_tcpThe AI is an excellent decision maker. Not fool proof though, so if its a critical application humans might still be involved as fail safe. If it fails in answer your question, you probably can shorten your question length and watch it do a better job than you could. Especially in the amount of time (instant).
When project managers and owners can’t even specify a single feature properly, how would they be able to specify an entire program to an AI?
One aspect I think is worth mentioning: Since the current form of AI draws "knowledge" from all available sources, but must bias moren frequent data more fhan rare findings, everything thats not easily available via Google search prompt ist also not available in AI. Yes, GPTs can easily give you working code that has often been written - even variations of that code. But once you get to problems that either haven't been solved a million times before or which solutions were not publicized - AI will struggle to give you meaningful insights. The benefit of AI over other information sources is the ability to do dialog-style searches (i.e. explain what you are acutally looking for) and adapting the answer to the given problem. Everything else seems out of reach of the current gen GPTs (which is to be expected, given their training methods). Newer AIs might be more flexible if their training contains fitness functions of autonomous problem solving - but this has yet to be proven viable.
Thank you Dave for another insightful videos!
Nope, not worth mentioning. Q* and similar approaches handle this - in the near future, the data will come from significant amounts of synthetic data. Generated from solving problems in a complex slower process to generate it. Solved years ago - just no one thought it makes sense to put it in before - well, 2023, which is possibly even too late for GPT5, though hints in interviews indicate it will mostly train on synthetic data.
wait until you see transfer clearing!
Hey coder, learn to drive a truck!! Now there's a twist!!!
I think there is an analogy to be made here with self driving cars. 10 years ago, the hype was that self driving cars will be here tomorrow driving us wherever we need to go without any human monitoring or intervention. It's the same hype today, and it will probably be the same hype 10 years from now. Maybe it will come true some day, but I'm not holding my breath. I think AI taking over programming is that same kind of hype.
I remember first articles on self-driving cars in the 90s and it was same bs hype just to get more investors money. Investors money is critical in majority of cases. Nobody wants to freeze money for 30 or even 50 years, they want results in 5 years so they cook 5 years predictions and naive people are still buying that.
@DestroyZionists How? You do realize that self-driving cars is AI are well?
@DestroyZionists whatever lets you sleep at night.
@@scmseanThere is a big difference between autonomous driving and programming. Driving is a task any able minded (sometimes not even) can preform at >95% proficiency. You can not say the same thing about programming. AI programming models are a different class of difficulty.
@@AJewFR0 You clearly don't understand AI or coding at all.
If you want to see AI disappear create AI CEO's. In case you couldn't hear it that was said with sarcasm and a nervous laugh.
Programming is about the last job I’d trust software to do.
Sorry bud if 3.5 million truck drivers lost there jobs to ai the us economy would be absolutely devestated..game over
I once was the main programmer on a system and trained 3 others to do what i did. The only problem was when the system had a major issue none of them had the ability to solve it. I was always called in, that's the indispensable part. My boss at the time asked me to write down the process I went through to solve issues. I told him I have no process and in order to do what I was doing you would need my experience also. I've never figured out how I did it, best I could come up with was stay clam and be persistence.
lol, if there WAS a process involved my boss would be the last person to know. Bit like a magician revealing his tricks.
There's always a way to explain how to achieve / fix something. If you're not able to in a way that more junior people can understand, it just means this is something you're lacking in your skillset. If it's only very recently that you trained these people, fair enough, but if it's still an issue after a couple of months, you're probably the problem.
@woet7891 no.
This is a talent. On 50s I still find errors in the "experienced" programmers, errors that I didn't do in my 20s.
It is about getting ideas about what and where the bug might be. The comes from experience only.
@@woet7891nope. Just because you can't explain something doesn't mean you don't understand it. It simply means that you're bad at formulating sentences
In the early 2000's the "Asian Invasion" brought cheap Indian contract programmers into the fold. As you likely know, wages took a dip, until employers found out that you do indeed get what you pay for.
I built and used many "code generators" thirty years ago, that greatly simplified my life by doing the boring bits.
But both AI as it currently exists and the "by the book" Indian programmers all have the same problem: zero actual creativity. No insight. No "light bulb" moments.
I don't know if it could work out, but AI for _project management_ sort of seems like it could be a thing.
But I retired two weeks ago, so I can sit back and watch. Any code I write is just for fun.
Is there any objective test for a "light bulb moment"?
Dealing with offshore developers I spent an inordinant amount of effort getting them to actually use source control and check source in.
I've been programming professionally for 38 years now and one thing I know for sure is that you have to adapt and change. I did programming in BASIC and even some database programming languages before I wound up programming in C and eventually C++. Still learning new things as over the years since I have done C# and Python and a few processors assembly languages. You have to stay curious and try new things if you want to stay competitive.
True but its does not take a person that creative to type in to a prompt to write code so I don't see how this applies really. Changing ones career can apply for example if web dev, go away, so I go to Embedded or Electronic Engineering but in the end, the AI can do all of those things as well. What would you suggest a person do in the up coming years on the Software Developer track?
Hi, Sir I'm complete beginner I've some questions regarding data structures and algorithms should I need to know this as a fresher , because I already knew - Arrays , Strings , Searching, Sorting, Recursion, Matrix , loops Pattern Printing , numbers should I proceed on other data structures ?
@@tapabratapalchowdhury507 Yes you should know all the common data structures and algorithms as they are the heart of computer science. Being strong on math is also important because it opens up more opportunities.
Is it possible that AI will start writing bad code the more code it writes? Or is it possible that AI will eventually become less creative if it keeps cannibalizing its own code?
socalled AI is not creative now. I won't accept AI exists until I see evidence of original thought, and so far I see none. I have been describing it as ML, Machine Learning. And then I saw someone who knows more than I that what we have now isn't even that. It's only DL, Deep Learning.
It's not going to help or hinder my photography because I'm autistic. I see things differently from every other photographer, my photographs are mostly quite different from anything anyone else makes.
I agree with Dave with his predictions. My career as a senior engineer who architects large changes for a system to handle complex user requirements isn't going anywhere anytime soon. How it gets written might change, but the up-front knowledge of knowing what can and cannot work hasn't changed. You still need someone who fundamentally understands the science and algorithms enough to apply them. Maybe AI just writes all the glue code, but it'll still be humans designing and verifying its input/output process every step of the way.
I'm in the same boat and even the glue code is going to be a long time before it's really replacing anyone, I think people are underestimating and undervaluing our profession and what we actually do. Look at all the products over the last 2 decades that were supposed to replace people or make building commercial systems easy by taking care of boilerplate... they've just not worked.
I don't know that programming will even be significantly impacted by AI at all. There's a real possibility that the LLMs are 80-90% as good as they're ever going to be. We went from 0 to 1 and now we are just going to see billions of dollars and man-hours spent going from 1 to 1.2. And, as things stand, exactly 0 jobs have been stolen by AI.
Given a huge codebase (like tens of thousands of source code files) of an app, sometimes a small thing does not work as expected. Good luck with the AI tools to figure out what the hell is going on! You need debugging skills and a deeper understanding of the codebase to be able to resolve it, and at the same time making sure something else doesn't break because of the resolution. That's where the software engineers secure their careers!
Dave is completely correct. I'm not a systems programmer, but I suspect I'm probably only around 10 years younger than he is, and I've been using computers for almost as long, as well as having used language models for the last 12 months now.
The one thing I can't get any language model to do, is write a tesselating hexagonal grid. They can do a square grid perfectly well. They can do a triangular grid moderately well, but they can't do a hex grid at all. The main reason why is because there isn't a lot of code to write hex grids already in existence, and all language models really do, is graft existing human-written code together. It's basically very sophisticated templating. The only LM in existence which is close to performing true, dynamic inference is GPT4, and even it doesn't do it very well yet.
Because language models are therefore not capable of novelty, a market will still exist for programmers that are capable of thinking and coming up with good original ideas. That means systems programming, and it basically means any area where there isn't already a lot of code for it on Github.
And there isn't a lot of code for most problems people actually solve on github or they wouldn't be hiring you to begin with, I think that's what a lot of younger people are missing arguing that AI will put us all out of work on here. It's not really true AI, it's a language model, a program for understanding and putting data together in a better way.
@@gentronseven What AI is really going to do, is raise the ground floor, in terms of the lowest level of abstraction that we all work on. Dave talks about having written assembly for instance, whereas hardly anyone still writes asm now. The world is even moving on from C to an extent, if you look at Linux possibly transitioning to Rust.
The question isn't so much whether there will still be programmers, but whether there will be technicians, or people who are naturally more inclined to interact with technology than the majority. Engineers will use AI to help them do their work, and the nature of their work will change as a result, but I think it will be a very long time (if ever) before human brains ever become completely redundant for dynamic problem solving.
So there will still be a hustle, of some type; it's just that your skillset might need updating. The good news is that if you've already got a background in programming, you're ahead of the curve for learning prompt engineering.
On a whim I tried your tessellating hexagonal grid Github Copilot with html/css. The code produced wasn't a million miles away from target but it still needed fixing...
>a market will still exist for programmers that are capable of thinking and coming up with good original ideas.
Which is pretty much all programming - from new tech stacks to mature code bases that require highly specific edits and additions, the LMs are of marginal use. They are just the new kind of search engine.
As someone getting a late start into learning programming, I'm finding AI to be more of a boon than a barrier. Having the new GPT4-based Github Copilot chat mode built right into VScode is a pretty amazing learning tool. It's pretty 'smart', can see changes to my code in real-time, and is inhumanly patient with my questions lol. I feel like I've learned a lot relatively quickly just by being able to back-and-forth reason with it about what I'm doing and why. Awesome video, btw.
That is super cool! Great way to learn quickly.
@@dallasroberts3206 It really is. It makes coding accessible in a pretty unprecedented way.
Same here, inhumanly patient
I think AI is great for Senior Devs also, where a Jr can converse with AI and try to work out the problem before "bothering" a senior dev.
You're welcome! 😊
AI learnt by reading decades of our work on repos and IDEs!
I just turned 18 and will soon be going to university. I always thought about programming as my future job. But with this AI stuff I'm not sure anymore.
I mean you say someone with high skills and deep understanding of code is going to be safe for the next 10 years but what about someone who just wants to start learning and is going to need a job still in 30 years?
I don't want to be jobless due to AI
We need government to regulate AI.
AI can't laugh at farts so no, it's not taking over the world.
I think you do make the point I am trying to make to my fellow junior colleagues. Coding is about to change in a very big way, but the coding is a small portion of what a competent software engineers does in any given day.
Boring coding job will disappear, boiler plate will be generated, perhaps even fully functionning program, but they still need to de document, debugged, optimized, maintained, deployed, supported.
Not to mention requirement gathering / analysis.
The next few years will be transformative, but no less interesting.
However, and I am starting to see it right now, is the MBA-type logic that you can a few high-paying engineers in the US that verifies / debug the code AI-assisted code made by incredibly offshore cheap cheap engineers.
"document, debugged, optimized, maintained, deployed, supported." - just feed it back into the transformer until it is all done and well.
@aresaurelian in my experience once you end up in this situation AI keeps speening in circles and proposes worse solutions or forgets some early requirements in order to "solve" most recent prompt.
@@zolniu Same here, whenever It is caught not knowing, it just start to make stuff up, and changing the original hypothesis.
Before AI take over my job, it will have to be much much much better.
AI is just a huge database of data already available publicly like from reddit/stackoverflow that has the ability to reword/recode its source, its a great tool for people who already know what they are doing but kind of useless for those who don't.
An example, I came from PHP, Javascript and Lua for more than a decade and I've been learning Rust before ChatGPT was release. When it was release, I use it to code Rust and my experience was just horrible, sure it can answer basic things like what you can easily find in stackoverflow. but when you dig on a bit complex logic and topic, its becomes horrible, it creates random fucntion that dont even exist, it gives you wrong syntax, it invents things, just pure horrible sh!t. The more you asked to fix the code, the more it gives you stupidiest sh!t, I even have similar experience with complex PHP object setup. the worst part is it does not know its providing wrong information and just keep inventing things.
I think I'll become more stupid learning Rust through chatGPT compare to reading books and watching videos from those kind-hearted people who's providing tutorials for free and digging/reading the code from opensource library.
People who believe AI can replace even entry level programmer surely never use it to do basic coding.
we are digging our own grave
Ai assisted programming is years and years away. GPT model 4 is OK for very small sections of code, but i think it will be a long time before you can tell it you want a project and it will write it.
Another problem is, people rarely communicate what they want in the first place. Numerous times I've had to rewrite the core operation of systems as they've changed their minds, or omitted details. You need people to interrogate other people to pull out the details.
AI assisted programming is here now. In a couple of short years all programming jobs will be gone. There will be no competition. We have to ask, what about the real jobs that supply us with real stuff, like food on the table, new fuels to run in generators and cars? Maybe people need to wake up... and start tooling up their garage.
@@togowackOk
@@togowacklol
@@togowack AI auto completion. AI written code for a whole project? You're living in a dream land if you think that.
@@thedeathcakeThey've got AI droids walking around doing household chores. The reason you can't get it to perform, is you aren't asking it to do a big enough job. Micro managing the small tasks is not what it does best. It solves the whole problem.
I’m a mechanical engineer. All I can say is when you’re past the entry level of coding (Python, C) after 3 months, the real issue is to come up with a creative solution for a given problem. I am still at entry level. What ChatGPT does for me is to answer questions about syntax and basic software patterns. But it couldn’t really solve any more sophisticated mid-level Leetcode without bugs and efficiency issues. I imagine that if I was not just a hobbyist and even with AI evolving I’d still stay ahead of current AI development. And now that many are anxious to choose a CS career it’s probably the best time to do just that. Problem solving and communication is something that a guessing machine is still far from excelling at. Just look at all the AI generated content out in YT. It is so generic that it makes you vomit most of the time. I am more busy now blocking such content from my timeline than anything else on YT. A bike is a bike but it still needs a rider
Similar to you, I've got 40+ years as a software developer, beginning in Basic and moving towards COBOL and other languages on the applications development side. I have been simultaneously learning how to develop AND developing my own AI app for the past year (in PHP, of all things!) Thank you for this very insightful and balanced take on the future of coding. You are the first person I've heard make the point that this "no-code" future everyone expects is still going to have to be created and maintained -- by coders. I have no idea how smarter the AI is going to get as we move forward, but working with today's iteration (all day, every day) I can tell you that anybody relying solely on AI's problem-solving capabilities in code development is committing professional suicide.
The dumbing down of society , I think this is a grim stupid future indeed. Next they will wonder why dementia is on the rise lol. People will use their brain less because they like the easy way out. If I created a Ai to do math, I guess nobody will learn math anymore. Ai Art is just as stupid for the same reason. I wish the people who made this stuff would just use the technology to as a tool instead, but In the end, none of this matters. I think the true path to freedom is to work for oneself. I hate the feeling of begging for a job.
Agreed. I do think its usefull for mundane stuff and for teaching. The smart guys are safe!
Hi, Sir I'm complete beginner I've some questions regarding data structures and algorithms should I need to know this as a fresher , because I already knew - Arrays , Strings , Searching, Sorting, Recursion, Matrix , loops Pattern Printing , numbers should I proceed on other data structures ?
@@tapabratapalchowdhury507 just go with it man
I hope the AI's can read some source, then remind me what I was thinking 10 years ago when I originally wrote the code. Nothing worse than having to pull out and debug code that is doing something that seemed really clever at the time, but now seems undecipherable.
10 years ago, I'd be happy if it could remind me what I was thinking 10 minutes ago.
The great thing I found with females (the 100,000+ years old definition of females) who worked in IT was they were great at documentation and testing. Autistic 'tard like me did document but only more complex innovations and not the mundane. They also interfaced with the business side in a corporate structure far better than me.
Unfortunately I think most women are going to be side-lined by AI which is a shame.
I also found female project managers far more risk averse than males. This is just my observation.
Or you could just comment.
@@Drew-Dastardly Sounds right. An Australian farmer was having trouble recruiting farmhands, so he decide to actively seek women. He found they were better workers, and his maintenance costs went down a lot.
@@oneeyedphotographer can you site any proofs without just talking about it.
Will there is another problem Dave, By the time I teach the AI what I want exactly, I could've done it myself. I'm no AI expert but I think AI will be just a glorified library and somebody has to read it.
Thanks for the video. I mainly work on the firmware side designing safety critical systems for medical and aerospace. Coding is just a tool that I use to solve real world problems. It's so much more than coding so I wouldn't mind ai taking over that part but i doubt it would anytime soon.
Even if it gets fully replaced by AI, it doesn't mean the hobby will die.
I just love programming just as much as doing brain exercises like Chess.
The specification for an AI to create a complex app could be just as complex as the code.
After all, code is the ultimate specification!
I don't really mind. I'll pick up a job waiting tables if I have to. I love programming.
I’ve been a self taught TSQL coder for 30+ years. I am hoping that this pays my bills till I retire in about 8 years. Your video is very humbling to me I feel clueless in an industry Ive been in half my life. 😊
with all due respect, how did you manage to survive in the job market by just knowing TSQL only and TSQL only ?
@@ldandco wow thanks for asking. Back in the mid 90s I was an out of work apprentice electrician who couldn’t find work. I took a job at a warehouse house for Walgreens. I taught myself MS Access and fell in love with database and making forms. Walgreens needed someone up in corporate offices to answer phones to dispatch the repair technicians out. I was sent up there. And through networking people I met up there. I was offered a position in a different department that was using MS Access. That took me into setting up their GIS system with ESRI. Somehow I got it all to work. So I went from loading trucks to demoing my work to Walgreens CIO in about 18 months. They moved me to the IT department and spent 20 years building spatial ETL process to load data for their maps. I was very productive with VBA/VB6. But the move to IT was working in moving TSQL to Oracle. About 7 years ago they restructured the whole area and was let go. Since then I’ve been doing mostly ETL work in SQL Server. It’s a crazy story that I still can’t believe happened to me.
@@ldandcoi know right? prolly a bot. Shit made no sense.
@@dispatch-indirect9206 For instance, large beverage manufacturers use really old SQL databases cause the machinery runs on really old tech.
Hi, Sir I'm complete beginner I've some questions regarding data structures and algorithms should I need to know this as a fresher , because I already knew - Arrays , Strings , Searching, Sorting, Recursion, Matrix , loops Pattern Printing , numbers should I proceed on other data structures ?
Without discounting the impact that AI will increasingly have on software development and coding. But assuming many of the layoffs currently taking place in the tech industry are the result of tech companies dealing with the economic ramifications of higher interest rates.
What is you perspective on the current capacity of AI to replace the workforce being laid off?
Lots of layoffs of programmers lately + economy not so good yet + too many competition, there is more ppl looking for a job than available jobs in programmming now + AI that can help a developer do the work of a small team alone, developers will still exists for sure, but jobs will be very hard to get from now on.
Ok today but not for long. Dave you have zero concept how fast things are moving. It's laughable you mention engineers will be tweaking the results from AI to match the requirements No the business will be revising the specifications. It's like your knowledge of AI ends in September 2021.
Some six months ago I asked ChatGTP to write a class in Python to control stepper motors (rotation, speed etc...). It was a real disaster, more like a joke. So.... I did it myself. Works best after all.
It's actually getting worse at generating coffee due to the scaling back that seems to be happening.
@@thedeathcake As long that it doesn't start generating beer... we're ok. At least for now....
You are asking it the wrong questions. It is well capable of the entire job scope. Could be that your answer is the wrong answer.
@@togowack Don't be so sure. I'm writing a DirectX 12 race car simulation engine and have found ChatGPT next to useless in that process. For DX12 code, more often than not it hallucinates methods and often entire classes that don't exist in the API, meaning almost none of it even works, let alone is optimal. Correcting it with different prompting doesn't make it go away in my experience.
About all I've found it useful for is writing hlsl pixel shaders and so forth. It's pretty good at that, but not as good as a human. I had it write an hlsl Gauss-Jordan compute shader solver once. It worked but was a terrible design. Within five minutes of manual rewriting, I had a version that ran 70 times faster which went up from there as I threw it out and rewrote my own versions. Often it's good as a starting point to get the gears turning, but for me it's quicker just to write stuff myself than to try to get ChatGPT to do it most of the time.
On the C++ side I don't even bother outside of very simple questions about specific functions which can be found in tutorials anyway. I treat it as a "quick Google" more than anything. I've found it helpful at times, mostly in debugging small sections of code when I'm tired and making dumb mistakes, but that's about it.
I think what we'll see more than anything in the near term is a whole new industry akin to life coaching who teach prompting, endlessly responding to criticism with statements like "you are asking it the wrong questions, for only $x/month I'll show you how to get the results you want." Courses were popping up only a couple of weeks after GPT 3.5 was released promising exactly that. People sure became experts fast. 🙄
@@toddwasson3355In the evangelical church realm, by prophecy, we are hearing the AI and androids are going to replace most middle class jobs, even CEO jobs, far more jobs than people expect. I still think a few high end jobs will exist, but only because the government will stipend some of that and require a small human presence. The 'doing' jobs (trades, manufacturing) I think will be much harder or impractical to replace.
So basically unless you're in the top 10% of the industry experience/skill wise, programming will be akin to a crappy $50k a year job you could get anywhere. sucks.
For anything serious you'll probably want accountability. What do you do if the AI does not produce what you want? What do you do if you explain your problem to the AI and it doesn't understand because it just is unable to understand you? With a human you can take time and try to explain the problem, but the AI either gets it or doesnt. If a human developer failed in this sense, you could fire him and get a new one who can do it. With the AI you're screwed. You get what you got.
Why can't you explain the problem to an AI?
"Even if ChatGPT were 3x better at blasting out boiler plate Python that's not what those folks are doing all day anyway". Well well said... I use ChatGPT and Gemini for code gen daily, it's my little helper. There's little it generates I don't have to use my skill to refine... It let's me focus on the challenging stuff.
I am not a programmer, but I love the way I can find answers to any question, without shifting through many web sites and digging down until I get an answer. AI will open many doors for people getting instant answers, but digging into many pages and reading different answer gives you much needed incite to the overall picture that is missed with AI. just my 2cents...
Yes, it can speed up what you’d normally do with a series of manual searches, or just present the results in a more readable manner, with comparisons and so on.
Yeah, it's really great for searching, google is right to be afraid of it. LLMs are basically better search engines not a replacement for humans.
I'm learning and writing cpp code and language, so give me a solution, what should I do, in fact what juniors should we do?
while we are paying our time for programming and learning programming, should we learn the programming ai, or learn the ai development? I'm learning c++
I think one of the most important aspects of being a successful software engineer is being able to adapt to the ever-changing tech landscape and the passion to learn new things. I started out as a rather successful Adobe Flash developer, making interactive web apps for businesses. Imagine what would happen if I didn't branch out to other web technologies.
I see a parallel here with generative AI. People with our skills are always going to be in demand. The only major consideration is how willing are you to adapt to the current environment? I'm on a constant quest to learn the latest relevant technologies. I'm just lucky, I guess, that I thoroughly enjoy that process.
I’m not so fortunate. I hate the rigor of constantly having to adapt. So exhausting and irritating. Compound that with a sleep disorder that makes it very difficult to think clearly.
@@andywest5773 Is enjoying constant change a learned trait? I accept that it’s necessary but the acceptance of the truth doesn’t make it appealing to execute. If you’re always having to change, then what can we take comfort in?
@@boingaon I think the only reason I don't mind the constant change with this specific field is because I'm really passionate about it. Like, I do it for fun, as a hobby and as a career. I'd want to constantly learn more about it, even if it wasn't necessary. Constant change in other aspects of life would be significantly more challenging for me. That being said, adaptability absolutely can be learned & improved.
If AI was really so smart, it wouldn't WANT my job ... LOL. I've been programming since I was 11 years old, and now in my early 50's, I am not worried about AI at this point. It's been a helpful tool in reducing some redundant, repetitive code and GitHub co-pilot certainly saves me $10 a month's worth of time. But, it doesn't actually solve any real problems for me. Most of the code is garbage, but it looks pretty.
Sounds like the same thing that happened to manufacturing when robots came on the scene. All the small easy jobs went away but robot programmers and advanced maintenance techs took off.
But to replace manual jobs you need to install super expensive hardware, human jobs are much cheaper than those hardware.
That's the reason their jobs still exist but with ai these Google and Microsoft will launch subscriptions for few bucks as they control the Operations
@@rajnijuneja251 lmaooo u really thinks robots and AI is the same, yall programmers will be replaced by 2040 .ONLY DOCTORS ARE SAFE
Well Dave, lets change history a bit and say Visual Basic and Visual C were just introduced in 2024.
"Look, it's AI! It takes care of all the low level drudgery and geek stuff so you can just write a program that does what you need. No need to learn the Windows OS structure and calls. Our AI does that for you"
See my point? This AI craze is way over-hyped and rich with FUD by investors who have no clue of the incredible gap between a machine and the human mind. AI is simply the next evolution on software engineering. New CS students will have to study this new TOOL. But is hardly a substitute for human reasoning and learning ability.
i agree with you but, AI needs to become AI not a large language model that uses probability, if all it knows is what we know now it cant solve future problems that havent been seen before. i use copilot to assist but it gets things so horribly wrong even given code that only has 1 error(i do it intentionally to see) i dont work on bleeding edge stuff im mainly keeping old crap alive until i no longer can. if AI could replace even half a developer i would be worried but it cant touch mine and i am far from a savant i just figure it out as i go which has made me invaluable. i joke about having a magic hat i extract solutions from. i was diagnosed with Asperger's in 3rd grade and have obsessed over technology for the better part of 50 years, ai worries me 0. if i were young i would obsess about AI but now im to disorganized with my time to get anywhere learning a new trick.
As I always say, 100% automation will not happen. But a significant reduction of work force for Software will certainly happen.
AI will not take our job, at least not yet. It is giving the same bad answers as a code that you can find on StackOverflow. It is like working with a worst of a kind junior dev who can't even use his brain. Not helpful at all, rather upsetting. And has complete Amnesia. Once in a hundreds attempt, it gave an approximately nice answer, the tab got closed and then the who discussion became unrecoverable. Trust me, I tried to ask the same questions, it gave me only wrong answers. (We are talking about Azure SDK, which is Microsoft's own product.)
I recently asked ChatGPT a question regarding a technology in our cloud stack, that's publicly documented, and it confidently gave me incorrect information. The problem with gen AI is that a user builds confidence in asking it questions but is then lead astray when it suddenly provides completely off the wall answers when asked other questions. I've enabled Google's Gemini search feature that will summarize search results for you but I mostly ignore it now because you cannot trust the information it's telling you is up to date or even relevant. Who knows where the information is actually coming from or if the sources most prominent in its models are even trustworthy.
For me, the latest LLMs are more akin to a narcissist that always has an answer to any of your questions but oftentimes has zero expertise on the subject matter.
It's so funny listening to all those "at least not yet". There is no "strong" AI in the world. Not only has it not been developed, but there is almost no research in that area today. What you call AI is a "weak" AI, which is essentially a large equation system, called a neural network. Through it does not have anything to do with biological neurons - the most advanced "AI" is countless orders of magnitude less complicated than nervous system of an ant, and I haven't seen ants programming anything so far. NN can to some extent do stuff like preparing speeches, painting pictures, writing phishing e-mails, i.e. work in areas where one could get away with bullshit. It won't be a brilliant painter, but it could produce tons of cheap amaterous daub for low-demanding customers. But it won't design a bridge or write a life support control system where accurate answers are required.
Majority of today's software production staff is a cheap workforce combining third party components into poorly working unimportant software and by some terrible misconception calling themselves programmers. Bundling Spring with Hibernate, copy-pasting trivial DTOs and REST entry points is not programming and it never was. It is automation of routine which threatens them. One way or another, with or without AI, it will kill their jobs. It happens everywhere: production, agriculture, document management, legal consulting, fining traffic violators, building social relationship graphs, etc. So no reason to expect it not to hit IT. Where are all those folks who were punching cards 40 years ago? So it's not AI but an inevitable automation trend that they should be afraid of.
AI will just be a tool to make programming easier/more efficient. I use AI to generate scaffolding/simple and generic blocks of code.
It can't replace system architects, and it can't understand complex, esoteric business logic
It won't do the first either.
Just ask it to write a dual needle stopwatch class you can use to benchmark your code and you'll see what I mean.
Then there's the copyright problem, because given that copyright lasts 95 years, there's no code out there that could have safely been used to train those AI's, without the guarantees that the AI is just reproducing it's training data.
@@scsftCannot tell what you are trying to say.
@@scsft I use Copilot already in my job, as I said for generating scaffolding and simple blocks.
I can't speak to your example, but it's a tool for me.
Also open source exists? Not sure why you're talking about copyright.
@@eadweard. I think it's more a case of dunning-kruger.
@@scsftWhat would be the problem with opensource on the mit and bsd licenses?
First, there are going to be loss of jobs in software development. The only question is how far are we away from rapid jobs losses in the industry. Besides the fact that AI abilities will increase significantly over the next 5 years, I think the next stage is using that AI to create new ways to build and manage software. Then it will take some time to start implementing these tools, which likely would be better when starting a new product or solution from scratch.
Libraries, building blocks, C++ Builder / Delphi, Code Completion, AI. It always seems that there is something that is going to "Get rid of my job," since the 90s. Still seem to be coding, or as you pointed out, fixing and properly redesigning other people's code.
My personal favorite/annoyance is when I properly design something. Get asked why I didn't do it this other way, explain I though of that, but these are the problems with that solution and these are the problems I can't find a solution for in that solution. Get asked to do the broken solution because it is faster to implement, spend twice as long as the proper solution and it is still broken and no one can offer solution to the problems I initially found. Then have to go back and implement my original recommended solution.
I have tried shortcutting to just the proper solution, but then I am "taking too long" to get work done, or "not being a team player." Oddly when I take 4 times how long it should take to play the game, I don't get complaints about things taking too long. I don't think I'll ever understand how most executives seem to think.
This has been at multiple companies over the years.
Anybody that accepts being AI debugger is not smart person and i fully enjoy his torment talking to chatbot till rest of his life ;) What you will end up is doing 10 ppls work by reading random code 24/7 and that is fate worse than coal miner. You will go insane and be paid zero money. DONT allow them to do that for your own sanity. "Prompt engineering" is other way of saying now u are educated coal miner.
Let's not forget where the code that the AI is trained on came from. Without humans writing (or at least revising) code, AI generated code is unsustainable long term because it will end up having too much of it's own output as input. I'm inclined to think that now is the best time ever to get into software, because there'll be a big mess to clean up sooner or later.
You make a very good point
What happens if it uses copyrighted code
@@NatetheNintendofan The coding AI can talk to a lawyer AI and "make a deal" with a bussiness AI that have copyright authority, then fix payment with the banking AI...
Maybe for non overly complex applications. If you don't want AI to take over programming jobs DON'T feed it all the input it needs to thteaten your careers.
I've been hearing "computers will write their own programs, and there will be no need for programmers" since the TRS-80 days. And ChatGPT's frequent lies would lead me to NOT use it for programming.
No matter how much complicated the machine, they will never surpassed human because the creator is human himself. Machine has no capabilities of becoming real human, so machine always need human to become 'alive'. Machine can't do anything without human.
I suppose if an AI wil create an entire app or program, that it will also have the ability of debugging and fixing itself.
Whenever I think about programs writing programs I always end up thinking about the "halting problem". I do believe that we will never have "perfect" AI, and there will always be room for the human touch. In fact, history has so far proven that technology always creates more jobs than it eliminates.
Exactly, the kind of AI that the singularity peeps talk about; recursively self-improving is dangerously close if not exactly in undecidable territory
But this technology is designed to replace people. Like self-checkout registers and ATMs have been doing for years if not decades.
Have you seen the new robots coming out that can cook? A lot of fast food jobs are going bye-bye.
AI will reduce the number of developers needed, especially junior developers. Mid and senior developers will also get replaced eventually too but not for a while yet. My guess is there's probably a good 10-15 years left in the industry but after that I think there will be very few if any developer jobs around. It could happen sooner than that though.
To be honest I don't know what to think about the whole thing. The last thing I want to do is become a documenter who just fixes AI bugs but I do see the value in potentially custom tweaking a LLM to boilerplate specific stuff that a human has already decided is the right path/concept for a project. All that aside, entry level jobs have already disappeared in recent years I can only imagine how much worse it will get.
I agree.
Entry level jobs didn't disappear due to AI, they disappeared due to the economic cycle playing out. There are literally still fields of business that aren't computerized at all, even simple things for coding like payment management. The future is incredibly bright for young engineers now.
I’m early career. However, I think in my lifetime AI will advance tremendously and AGI will be here.
I think we will always need humans to manage complex systems and the models. Automation of the shallow work will be a no brainer.
You will see a lot of freelance P2P work due to AI and automation within companies who are up to date with these systems and/or technologies
Insightful views Dave. "AI" is some ways is little different to the various low-code tools that have been presented over the decade. I recall Logical Systems (circa early 80's) had a language called "English" we could train someone in a week to write programs, but it was anything but efficient. Yes they could do something simple, more complex that did work for the majority of folks we taught.
100% on the problem solving/creativity.. I have always maintained good tight code is somewhat an art form than engineering.
Looking forward to more of your insights through 2024, and great to se you looking much better so I suspect your back surgery has been a success.
It's sort of like a low code tool combined with a google search, so it can solve maybe 1/100th the problem, I just feel like only inexperienced people are going crazy about it. Maybe I don't get much exposure to that outside of youtube comments 😄
Oh gosh, those stupid scammer reply-bots trying to lure people onto private chats to scam people out of money. I hate them soo much.
Thanks for deleting that scammers reply to me.
As long as AI is giving answers that are often wrong or obsolete we can be sure that asking the right questions will be a task for professionals
wont happen because writing new code is only 10% of the job. 90% are changing *existing* code and communicating with the customer or other teams. I dont see how AI could do that.
Onus is on you, been a truck driver, been that software architect/engineer debugging multithreading timing issues in java, going in as deep into 3rd party libs or OS...
And I only have a high school degree because i suck at fitting in anywhere...
AI can't replace me... it's not bad enough lol. AI is assisting me but still has many areas that it can't. I was asking for help once with WPF and it was giving me answers as if WPF was truly object oriented. The whole dialog was like:
AI: here's the code
Me: that does not work because there is no property X
AI: I'm sorry, you're right try this...
That repeated until I figured it out myself.
I remember when the same argument was made about CASE tools. I've even worked with some of these tools. They made doing simple things trivial. But that didn't mean that you needed fewer programmers, though the marketing materials claimed it was so. It really meant you could tackle harder problems. I suspect it will be the same with AI. Rather than making programmers go away, I expect it to free up programmers to tackle increasingly harder problems.
Here is different.
Probably on couple years you just describe what you want to build and AGI just make it and implement to production.
@@mirek190 There's ALWAYS ALWAYS ALWAYS a reply to these types of comments stating "This time is different" or, "It's only getting started" in regards to new technologies like generative AI. Sure, it will change a lot of things, but there's always a group of people over exaggerating the impact of things like this, almost like they're wanting to FINALLY be proven right.
@@robbyyant6213 Within 7 months from my last comment AI got in programming huge step forward and getting very fast progress to became fully automatic programmer... is insane what can do now even opensource models.
Devs will fix ai code; I am not even exited to fix my own code, lol. No way it will be the case.
Devs will do more and more prompts and less and less raw coding
I imagine a lot of jobs will start appearing for "prompting engineers." You already see some of that with Stable Diffusion and AI art. You have to spend hours and hours trying to figure out what kinds of work through trial and error, and that's a skill some people are willing to pay for. The tools aren't powerful and straightforward enough for the average person to still use, and still require a certain level of skill to use effectively and get what you want.
agree Stable Diffusion has ton of extensions such Control net, CivitAI +, animateDif and many more.
Doesn’t AI and Machine Leaning need a lot of data and processing power? Would that mean that in the future there would need more data centers and more processing power to drive the extremely large data needed to power AI of the future. Compare that to software engineers of the past who could make entire programs that fit on a floppy disk.
BING says: "AI won't replace programmers, but it will change their roles and create new opportunities." But, if they were going to take all the jobs, then, that's exactly what they would say wouldn't they! :)
What are your thoughts on Embedded Software/Systems Engineers and Robotics Engineers being impacted with AI? Will they be able to write code to control hardware still?
The problem with this whole "anyone can code" movement is thst it's not true, and usually championed by those who can't either. . I spent most of my childhood failing at programming for the same reason i failed algebra, and the vast majority of kids have no interest in the time and patience it takes. My cousin's best friend's dad you probably know, I forget his last name but he invented Encarta. So in 1995 right after windows 95 came out, I got to run around Microsoft unsupervised with him and my cousin and even met the very first Xbox employees in the back of the library with every game system known to man. In fact the only place we couldn't go was the windows section :-)
As AI gains usefulness, I think that software will become cheaper to produce. However, I think that the demand for software will only increase as the cost falls - to the point of requiring more, not fewer developers.
I think we are a long way off from AI making applications, I think we will see more AI in games.
Depends what applications we’re talking about. I can see it being able to pump out crud-type applications with relative ease in a few years because a lot of those applications are just boilerplate with some business logic weaved into to it.
Yes , programmers are obsolete, stop learning it, go do something else and leave the market open just for me mwahahahaha. j/k
I've only been working as a dev for 3 years now, but just given what I get up to on a daily basis, it just seems like AI will eventually become akin to new libraries and packages.
Also, a huge elephant in the room with AI adoption is how all the copyright stuff shakes out, I'd bet there's a lot of software companies out there watching the nytimes v. openAI case very closely.
There are two big open questions right now:
- If an AI is trained on some non-commercial use only code, is its output also restricted to non-commercial use? You can have an *opinion* on the question, but legally the answer right now is "no one knows" which for a corporation building software, that's scary as hell.
- Is work authored by an AI even copyrightable at all?
That second question is getting very little discussion right now, but is potentially a huge setback waiting to happen with AI generated content adoption.
AI generated code is not optimized only humans can generate the code that does the job properly.
I had a long career in IT. Mostly in system and network admin. Doing that I wrote a lot of code mostly to make things easier, to convert from one system to another. I had a developer role in a few large projects. My feeling is that the most time is spent understanding the problem that you have to write code for. Writing specs and testing the results that AI produces is going to be the future of coding. On the other hand, now that I'm retired, I write code for the fun of it - just started playing with Sudoku.
This is probably a dumb question, but since you had dev roles, I would assume you are at least decent at programming. Am I under the wrong impression when I think devs would be able to make classic puzzle games pretty quickly?
I just see these programmers on youtube sometimes and it seems like that just make everything really easily.. and it takes me a while to make stuff.
It depends. How much time do you want to spend on it and what you're trying to get out if it. I just started learning python so when I thought about sudoku, I thought I can draw it using Pygame and represent the values of the grid in a dictionary. I'm at the stage of writing bits of code to do those things when I have few minutes here and there. Right now I don't have much insight into how the program would solve a puzzle short of brute force.
Most experts concerns about AI impact on the workforce will start and end with technical jobs.
Another area it will have similar effects on is business management.
More VPs and CEOs will assume wrongly that AI by itself can do most of the work and won't consider the quality or reliability of the product.
I enjoy programming as a now-retired hobbyist. Ai will never replace me. I'm in the process of writing firmware for a new project as I write this. In assembly language (still). Programming real-time state machines running with pre-emptive interrupts IS fun! Much more fun and a better use of time than doing cryptic crosswords.
We are going to see more and more specialized processing and chips. It will take specialized firmware because they will have unique instruction sets. For example, my Subaru has over 30 processors running in it that watch everything. It even yells at me when the car ahead of me in the fast food line has moved. And how about the firmware in the latest cameras? It knows how to acquire focus for a human versus a dog, for example. It marries what it detects with fast, sophisticated firmware to set the lens properly. And so it goes.
I found the truck joke in very poor taste. The people on the lower skill level of programming can improve their skills and become better programmers. Sending them off to be truckers is kind of offensive and elitist.
In short its over.
No matter how skilled and 'indispensable' you are, you''ll be replace in a day.
I honestly cant think of something that humans can do that AI cant be taught
I think because AI is coming out at the same time as management is cutting back staff after the post-pandemic panic overhire people are getting the wrong idea that AI is taking over.
LLMs aren't replacing even crappy programmers anytime soon. Someone has to describe what needs built, and coming up with that description is programming, even if you can do it in natural language. Even at that point (NLP -> code), you need to be a skilled engineer to recognize when the LLM gets things wrong. Right now, and for the foreseeable future, LLMs are basically assistants that are really good at googling for your answer and slightly adapting it to your circumstances. Picking the closest Stack Overflow answer is really helpful, but it's not effective programming by a long shot.
I say this all as a professional that LOVES my LLMs. They really really make me a lot faster, but I still need to be good at (in the best case) code review to make it work.
Nope? First, developer already get fired, second... "Someone has to describe what needs built" - which is generally not the crappy programmer. Specs and requirements come down from customers.
@@ThomasTomiczekwow you must have some very literate customers because as someone who is working in consulting for clients whose core business isn’t technology they really need programmers/IT consultants to translate their ideas into specs and designs
You're looking on current llms .... That's really bad assumptions for the future ....
Look how fast llms are improving itself even in coding.
9 moths ago were hardly write simple code right now are able make whole application and put into production!
Imagine how good will be in 2 years.
@@mirek190 Apply that same thinking to speech recognition. There's a plateau somewhere, and progress after that might be impossible until you find a new paradigm shift.
What I feel is not everything will be completely replaced by AI or viceversa.... It will always be a combo of humans and AI agents so that the best of both worlds is generated... Therefore fast and accountable product development and delivery can be ensured..... But my worry is earlier teams that had 10 members can be cut short to 5 with the help of AI.... That means the lower end 5 people lost their jobs..... Enhance your skills.... Thats the only way🤷🏼♂️🤷🏼♂️
Coding, just like everything else, is evolving. Shame on those that can’t, or more sadly won’t take a peak around the next corner and adjust to the coming realities. As usual you’re spot-on with your observations and advice. Keep up the good work it’s always fun to hear what you have to say.
The problem isn’t AI coding, it’s the owning class replacing people with AI, so now no one has money.
People refusing to use AI to their advantage and claiming it's useless because they wrote shitty prompts with GPT 3.5 a year ago and it wasn't 100% correct are holding themselves back or coping. I learned so much and got an unprecedented amount of work done last year thanks to AI assistance, and I used it to help me land a better software dev position and automate various aspects of my business.
@@DingoYabuki imagine getting promoted while you do nothing in the job while AI does everything.
You read ALL the comments?
Well i applaud your tenacity.
I love your content and often find myself with it open on my 2nd, 3rd or 4th screen while working.
My analogy for AI in the development scene is: I am able to hammer a nail into a wood board using a rock, but a hammer will do that job a lot better.
AI, to me in my daily job, will be another great tool to enhance my productivity and take on the grunt work that i don't feel like doing (creating for loops, code docs, logic optimization, spelling checks, etc.). Much like Linting, intellisense, predictive text, heck... Even line numbers, AI (depending on the developer skill) should not replace me, it should make me a more efficient dev with fewer chances of making silly mistakes like mistypes or accidentally inverting !(!a && b) ever again.
Working as a senior dev, partly took on the role of tech lead for a bit as took a while to find one that matched the role, I spent about three weeks of the last 6 months coding, and I probably wrote over half the existing code base. The entire rest of that time was requirements scoping, design, and trying to handle integrating with other teams/existing products.
Management has this idea of what they need, but doesn't exactly understand how to properly define, and draw boundaries for that need, my (and other seniors and intermediates) jobs will be fine.
The question is really more how juniors will get into this space, since writing the building blocks often helps one in thinking of how it fits in the overall system design, think in the beginning it will probably be prompts with then analyzing the output to see if it fully fits, with seniors giving feedback on that and juniors picking it up from there.