Is Coding Dead? (AI's Takeover)
HTML-код
- Опубликовано: 26 фев 2024
- Here are my thoughts on whether or not AI will eliminate coders.
Discover More From Me:
🛠️ Explore hundreds of AI Tools: futuretools.io/
📰 Weekly Newsletter: www.futuretools.io/newsletter
😊 Discord Community: futuretools.io/discord
🐤 Follow me on Twitter: / mreflow
🧵 Follow me on Instagram: / mr.eflow
🐺 My personal blog: mattwolfe.com/ (Currently Being Overhauled)
Resources From Today's Video:
Jensen's Thoughts: / 1760619467019804878
Emad's Thoughts: • Why AI Matters And How...
AI Bug Fixing: / 1762175776726663267
Computer System Layers: www.secplicity.org/2018/09/19...
Programming Language Trees: asset-cdn.tecky.io/2021/10/20...
GPT-4 Coding Assistant: • Meet Jarvis: My GPT-4 ...
AI Code Speedup: www.tabnine.com/
ChatGPT Platform: chat.openai.com/
GitHub AI Programmer: github.com/features/copilot
Gemini AI Fixer: gemini.google.com/app
CodeWhisperer Generator: aws.amazon.com/codewhisperer/
Cody Coding Aid: sourcegraph.com/cody
CodiumAI Code Testing: www.codium.ai/
AI Tool Selector: www.futuretools.io/?tags-n5zn...
AI Assistant Review: • I Tried Every AI Codin...
Problem Solving Value: / 1762110222321975442
Sponsorship/Media Inquiries: tally.so/r/nrBVlp
Mailing Address: 3755 Avocado Blvd Unit 287, La Mesa, CA 91941
#AINews #AITools #ArtificialIntelligence 
Наука
![PIECE BY PIECE - Official Trailer [HD] - Only In Theaters October 11](http://i.ytimg.com/vi/7Bc6trBc1kc/mqdefault.jpg)
In my many years as a software engineer (I'm now 77, retired) - The biggest obstacle in delivering bulletproof applications was the push to get it done by the using community/customer to the point where we were forced to begin development before the problem was adequately defined and understood. Often, we were financially incentivized to start early or threatened with a financial penalty if we did not begin before ready. These important first step require humans that understand the business on both sides of the aisle (user/IT). We were often paid extra to deliver something with holes in it and then paid once again to patch those holes post deployment. While I believe that today AI can assist in accomplishing these important first steps it cannot do it alone. Maybe someday, just not yet.
Humans are not going to be replaced anytime soon but when the technology arrives to do so it will mean those exact same humans will be able to apply it to whole new products not imagined today.
I totally agree with you. It will not happen now, but perhaps next year. 🥰
I disagree with that and would say it's entirely the other way around. The design and concept needs to be from the human. Once set up and rolling AI would be better at finding bugs.
wisdom++
no different to having n aapprentice tho
Learning how to code has taught me much more than just writing code.
I learned about algorithmic thinking and problem solving, I learned about how the internet works and what safety lacks there are. But much more important, I learned how Ai actually works under the hood rather than just knowing the right prompts.
I think it’s a good point that you showed the abstraction layers. My goal has always been to understand every single layer and I don’t think that will change with Ai on top of it.
Keep learning guys. Ai is an awesome tool but don’t let it ruin your ability to think.
Spot on!
You didn't know about problem-solving before that? Like from elementary school? And how did coding teach you about the internet and whatever you said after that doesn't seem like English? And algorithmic thinking? I sort of understand that one.
Over-reliance on scientific rationalism and an analytical mindset have already ruined your ability to think. Please track down and try to understand Iain McGilchrist's work to see how dumb you actually are. Thing is, you can't and won't understand it, because your brain has been ruined. Shame. Personally, I love seeing all these egomaniacal, arrogant, pampered engineers being put out of work because machines can now do their job as well or better for pennies a day. As Ilya Sutskever warned: if you're all about "intelligence" (in the narrow analytical sense that term is commonly used these days") you are about to have a bad time.
@@joaoguerreiro9403 thanks mate
@@nickgirdwood3082 I think you don’t know what I‘m talking about because you can’t relate, can you?
I think the problem lies more on the low-end. There won't be a need for entry-level coders, but this of course will make it much harder for programmers to gain the necessary experience to become useful.
Nah you can use AI to accelerate learning too.
they are not honest they come to replace you the art world is the only one appearently fighting this with lawsuits u can see thhis while they are talking like in" 5 years there will be no programmers and thats good noone wants to do that ...." they act like they help humity meanwhile they have ai generated work behind their backs on the screen which is based on stolen work on the internet
An AI assisted human will still learn way slower than an AI. I would know. A year ago I spent my life savings attending a pretty highly regarded coding boot camp. But now I can't get a job, because AI is outpacing me at every turn, even though I'm using AI to continue my education. I just can't learn fast enough to be useful at any company.
@@consensusg9226If you are a good developer (not programmer, programmers do what others tell them to do, developers 'think' and do design work) then you're in demand. Hell you don't even need to be all that good in many cases. I keep hearing programmers will be gone in 5 years. AI code assistants are useful but no where near able to code without a developer. Now I get that we're advancing rapidly, but we don't ever seem to talk about how complex systems are. I'm more of a backend developer, with high scale distributed systems. Even if we say AI can build out such a system (not seeing that in the next 5years), these systems fail and need experienced developers to maintain the code. Also writing code from scratch into a solution is one thing, adding updates, that would affect multiple parts of the system? Well that's a whole different ball game and one that is extremely difficult for AI to do. However, if we get our AGI and it can perform these tasks, then we're all out of a JOB! The issue is with simple (relatively speaking) things like a game there's very known set patterns, however not all code is a simple pattern. I do think that entry level software engineer roles will be harder to get, as AI will be able to do a lot of what they can contribute in a few years.
@@Gatrehs it accelerated learning rate 5x but increased the minimum knowledge requirement 10x.
The worst problem with abstraction is the loss of understanding the previous layer. We are building "black boxes" that we don't understand.
Take a more physical example like agriculture. We started by growing small plots of land with our bare hands. Nowadays we just grab a package from a shelf, not knowing ANYTHING about the work done to make it appear there. It's convenient but makes us more and more vulnerable.
I'm a developer and I don't have a problem with taking on a new career. In worst case I'll grab a shovel and plant potatoes, back to square one i guess. 😅
Best comment. Great analogy
Sure, but it will democratize a fair chunk and give people with good ideas a chance of trying to make something ground breaking that a techie would not have come up with, the fields in coding that will need specialists will still be around, but the IT job market will shrink enormously... Hence, you like a good analogy: before, you needed a ton of people doing different tasks on the land to get the product, now, you have a ton of abstraction and ease of use on top of the basic labor, aka: machines, but you only need a handful of mechanical and biological specialists to keep the process in the right lane and have more and better produce. I mean, it's an obvious one really, but still, it's progress, jobs and stances towards tech / IT will change and adapt, some people will get a bad deal, many more will get a good one.
But who made the shovel? 🤔
@@yoyonis6840 They grow in the shed? 🤷🏼♂️
Actually the point of good abstraction is to start near the desired solution and in its language and find or build modularly what you wished you had to solve that type of problem. With a few languages really good at custom at the problem space abstraction/language level solution creation today you quickly hit a software language and tool kit for weaving together COTS with known APIS. The problem is we too quickly get pulled out of the customer problem space into what our set of nails and hammers looks like. Using LLM doesn't solve this. It makes it a bit worse. The LLM has seen lots of examples of nails and hammers and how they can be pasted together. You have someone fitting it prompts that may decide that. So all you have gained is applying the nails and hammers. You haven't taken care of the leap from real customer need space to selection of tools AT ALL. This is the top down view.
From bottom up view the point is to have mastery of the level the abstraction is built on done by experts in that level with input from those that need that level to do work but need a VERY different view from their own level of interest. Repeat all the way up to the actual customer with a need to be fulfilled by a customer solution that makes sense to them.
LLM produces faster code monkeys.
Not only coding but design and research, too.
As a Senior UX Researcher, I see my profession having ~15mth shelf-life for the _vast_ majority of us in the public sector. Those of us in government, very large enterprise corporations, or companies which will fall-over, will still employ us but those positions will mostly be eliminated, they're just slower to react.
What a few fortunate UX Researchers will evolve into may be some form of UX-AI Agent Managers. That too will be a temporary bridge to whatever comes next.
Pretty much any jobs with repetitive workflows. Even cooking.
@@aynrandom3004 we need to start voting for automated goods, because if we dont we wont have income as society to pay for what we need and companies will own us
Totally agree, I spent the last 10 years working in various digital agencies. They’ll soon be redundant, creative teams will be much smaller and work directly within organisations. Time for us all to adapt to this new future.
I'm a Service Designer and UX Researcher and agree with your assessment, Brian. I think there will still be a role for some time for a human to go around and collect the necessary context and 'feed it to the machine' so that it has (reasonably) clear objectives, but the analysis and implementation will be largely automated within a few years. I think the main skill we will all need to practice is flexibility!
Such a great comment with a deep and real realisation! @@jonathanmelhuish4530
As someone, that is struggling to get into web development, in the hopes to be able to provide for my family with that work, this makes me really anxious.
I was studying it for a while. Feels thoroughly pointless now
If you’re struggling then this might be the way to go.. learn to prompt the code co pulot
These tools help newer programmers more than the seasoned programmers.
As a developer who uses these tools regularly, I think they are still some way from replacing human software developers. For now they can help make a developer more efficient. I think this will likely be the case for a while to come, but hard to know for sure.
@@chrisprosser5055 in this video the suggested timeframe is anywhere from 1 1/2 years to 5 years at the latest. And with the progress being exponential, I think it's probably closer to the first.
3:17 Uhm, C++ (1980s) is not easier than BASIC (1960s). Not even close.
I thought the same about the COBOL comment. COBOL was not difficult, just arduous, and the same holds true when comparing BASIC to C++.
Try learning Rust lol. These languages don't get easier to use, just better at managing the complexities of coding.
Exactly. The dude is being polemical, not accurate.
That's a shame, I learned a bit of Basic, this was a big ego boost, now I feel stupid again😂
@@randfur No kidding. Rust is probably the hardest language to learn on that list, and it came WAY later.
Depends on where the limits of neural networks are. One thing is spitting out small pieces of code, another thing is to create an entire project from scratch. Complexity grows exponentially, and there might be some kind of mathematical limit to what a bunch partial derivatives can do.
yeah but the limit is much further than human limit :D
An AI can get the context of 1 milllion tokens. Probably 10 million next year and 100 million in 2026. It's coming for super large code bases
@sco444 Yes gemini 1.5 can process over 1 million tokens. It still can't even accurately process the job of finding on needle among many, meaning search the context for sentences like "this is the nth key: 13213", it fails when there's more than 1 and this is a task even simple code can do. And despite increasing the context massively google didn't move the performance on the code benchmark "HumanEval" in any direction past what the previous values were. And that benchmark isn't full of particularly interesting tasks to begin with (it literally has things like reversing a list , creating simple tables and the like)
@@NiblssI hope you realize we just entered the AI age. These LLMs will only get better with time and will make less mistakes and become 99.99999...% accurate
@@Niblss so the 99% accuracy for needle in a haystack is BS? Damn, I thought it was a real deal.
I'll believe it when it's actually possible. The problem is you still need people to work out the requirements and how to actually do that kind of work. This comes from a software engineer with 20+ years of experience. We still have people that do the code and the building of software because most people aren't cut out for the real aspects of the job, code is small part of building software.
Artists felt the same way.
You're not going to be telling your AI bots to design and optimize operating systems and game engines and domain specific proprietary code on custom hardware that you'll never see on GitHub. Programmers will still exist and use AI as an aide. Quite probably not as many. We'll see how it pans out.
@@lairdpeon clearly you've never coded a project. Ai will replace devs eventually, but the main job a developer is to asks and extract the business problem from the client and then solve it using code. Clients don't even always know what the issue is so you need to ask the right questions to know what the problem is and then solve it.
You'll need to take into account that LLMs are getting more intelligent. AGI is not an if its a matter of when. And AGI will be like talking to a person. AI coding products will be off the advanced LLMs at the time. Its a better to assume it will happen soon rather than believing when it actually happens.
@@jonathandavis8599, actually that’s the realm of business analysts.
I'm a dev. I believe coders will not disappear, but one person will take the jobs of more and more people. Already with the help of AI I can code like if I were 3 people, down the road it will be 10 or 20 people. The job as dev will be more like that of a supervisor. There will be a bottleneck, maybe at 20 or 30 people, since you can't supervise more than a certain amount of code.
So I guess a lot of jobs will be lost, but then again this will happen in every single industry.
its going to fuel more income disparity everywhere, as if we need more of that
I'm not a dev (marketing/data), but I see things like you.
I frequently use GPT to dev some javascripts and I still need devs to help me sometimes.
@@julien5053 And of course GPT will NEVER get better. I mean, it's been around for what? A number of months? Surely we can extrapolate that "all is well" based on this huge data set. Excellent point.
@@RobLed Never say never ! 😄🙅♂
With AI running everything it will be essential for more expert coders to keep the system from destroying itself or the planet. Also learning how to bypass these systems and h4ck will be highly lucrative and possibly life saving.
1. CEOs say these stuff all the time to pump up their investors. Believe it when you see it.
2. As everyone said, coding is 25% of the job, rest is problem solving and creating value. So, coding might get abstracted but software engineering won't.
Great video BTW!! 👍
but like 70% of the coders just do mundane tasks, they will get replaced
Give an example of problem solving or value creation in software engineering that AI can't do?
@@ghhdgjjfjjggj I've never worked anywhere like that where there's a bunch of people who just write code based on what someone specced up. Maybe there are coders in some places but I've always worked as and with developers and software engineers, being able to use a programming language is a fairly trivial part of the job. Understanding the requirements and crucially understanding where the where the requirements are wrong and how to solve problems and how to integrate with existing code and how to find workaround for things that should have worked on paper but don't and many other aspects are part of the job. Producing isolated lumps of code is just not a job that people do.
writing non-buggy code @@emmanuelameyaw9735
@@emmanuelameyaw9735 You tell me what problem solving and value creation in software engineering AI can do. All I've seen it do is produce isolated lumps of code. That is trivial to developers but is the tiniest part of software development. Have you seen AI write a good story and I mean good not just something that fits the definition of story. If it can't even do that, it can't produce a complex piece of software accounting for all the interconnected systems, make it ergonomic, make it secure, make it efficient. It will improve as time goes on, but until we see AI do more than generate lumps of code it not sensible to make huge predictions.
I think that it's still going to be useful to have people who can break problems down, think systematically, and has an understanding of how things work under the hood, e.g. what is easy, what is difficult, what is impossible. For example a layman may ask for an app to have an offline mode in a situation where an offline mode would be pointless or dangerous.
exactly. Sometimes the client will tell you to do X and you need to say no that is stupid.
AI will be able to assess and break down complex problems in the most understandable way possible. Essentially what it is already pretty good at. A layman may ask that question, but AI will be able to understand and interpret what he means. Think bigger!!!! : ) xxxxxx
@misterr3083 At some point yes, but for a while it will continue to have that flaw. I've tried AI code generation, and current systems are like a junior dev. Can write code, but no common sense, terrible at debugging, and loses the plot on bigger developments.
@davidcummins8125 "for a while" yeah, for like a year
You are correct, but newbies here will probably not understand, and believe you can always throw AI on top of the mess of web development and its abstracted layers and everything will be fine.
As someone who has been programming professionally for 20 years, the biggest issue is trust. If you hand over programming tasks to an AI, you need to trust that it will do the job correctly, and be able to return to that same code to make changes-- without breaking existing features. The moment you have to step in and read the entire code written by the AI to understand what it needs to do next, you are now programming again and you have failed to delegate the programming work to the AI.
Your not wrong but I’d imagine you’d be able to tell the ai why the code is wrong and what needs to be fixed, all through text
💯
@@funnycrow4462But that just goes back to the same problem that all generative AI systems have - some times it is just easier and quicker for a visual artist/programmer/writer/whatever to fix the obvious issues yourself than to exhaustively explain the AI what nuances it missed or what it did wrong.
If it has reached the state (if such a state can be reached for AI) where it can always intuit what humans mean and require, then no profession that inlovles interfacing with a computer or computer controlled system is safe, and if it can't humans will still have to examine what it produced to make sure it didn't do something wrong. And the threshold is high for professional work - you wouldn't enjoy it if in the middle of an otherwise fine but non-experimental novel, that was not advertised to AI written, suddenly introduced new characters, went on weird nonsensical tangents, had a quarter of a page of gibberish and so on - and you would like it even less if the software for an elevator or subsystem for a water sewage plant to work correctly 95% of the time.
Generative AI has come a long way, and it may be safe to entirely automate several industries, but that day is not today, nor is it likely to happen in the next 5 years, if for no other reason the fact that the AI and AI tools have to earn our trust.
Even today you are already trusting the many layers of abstraction that lie beneath the code you are actually writing.
Same can be said of humans, look how much human written software is complete slop. AI doesn't have to be PERFECT, it just has to be better than humans. Currently there are fields where AI outperforms humans and people say "oh it only scores 90%, that's terrible", but the expert humans are only scoring 80% so do we give the jobs back to the humans and lower performance?
I started with Perl, currently a PHP developer, I have worked with legacy code (written 25+ years ago), I would like to see how AI starts maintaining and writing new functionalities in that kind of project.
Chat GPT can understand English. No one needs to learn English anymore
lol best comment
😂
You need to learn English to interact with the world and interface with ai.
You don’t need to learn coding to interact with the world or interface with ai.
Dumb.
@@codycast no
This is a dream come true for me even though I love AI I’m not a person who has the mind for doing detail tech, so I’ve always been the conceptual person and it’s been a limitation for me. I know what I wanna do I always have ideas but I can’t get past the tech and it’s very frustrating that I have to have a gatekeeper helping me to do whatever I wanna do now I’ll be free to reach the top shelf without having a beg somebody for their TECH HELP. This is been a dream come true I have been waiting my whole life to have AI where I can be free and do all of my creative ideas, I feel bad for a coders if they think that it’s gonna be the old days but I think you’re right I don’t think they’ll be replaced. I just think they’ll be enhanced.
He worked as programmer 0 hours and knows everything about future of programming, amazing
Exactly. I just block these kinds of channels on YT, it is pure garbage in every sense and it is just polutting the platform.
Cody helps you think about problems and how to solve them in detail. Always a good skill to have.
Kompozycja muzyki jest w tym lepsza. Bardziej abstrakcyjna. W momencie kiedy zrozumiesz harmonię muzyczną zaczynasz rozumieć wszystko dookoła.
something some of these fools in the comments do not get. I learned to code not to earn a lot of money, but because it improved my problem solving skills.
"skills" are a very 20th century thing to value. Get in the van--we're headed for existential meaninglessness, baby!
If the human race doesn't continue learning these valuable skills and pushing our abilities then we will become extinct. There will be plenty of people wanting to run systems off grid to protect their privacy.
Spot on. As an assembly programmer for 50 years, the utility of my skills has been overtaken by high-level languages. Not that things wouldn’t be better, faster, more efficient, if written in assembly, the point is, sadly, no one cares. Everything is viewed as temporary and disposable so no one is willing to invest in the quality.
I like it when there's something like a clock or a or an electronic banner in the background that way, you can see how many times the creator had to cut his content. makes them seem more human.
GPT 4 is great at generating individual components of an app well but when you ask for too many things in one prompt, or even a single component with many constraints, some of the constraints are left out or half done. Task specific agents may solve this.
Yeah you gotta tackle things step by step and keep reminding it of the main prompt. That’s how i tackle the issue.
So for example I’ll be like:
“Remember this prompt from my earlier message where I’m working on a project that’s needs to do xyz?
Here’s a reminder:
[pastes prompt]
I’ve now completed sections A and B as you have instructed and here is the code:
[paste code]
Can you now discuss the next parts of the prompt? Do you understand what I’m asking and are the current sections correct so far?”
This kind of approach really helps, over time you kind of just learn how to prompt to achieve your goal. I also usually have other tabs open for side issues or quick fire questions so that the main gpt tab doesn’t deviate too much and lose track of what we’re working on
Yes, but this is already solved in Gemini model. In it, you can paste milions of words or of code. Just temporary hurdle, will be gone by end of this year.
@@MartinDlabaja Yh I’ve seen it has a much higher context window but through the eyes of others it seems like the reasoning is not as strong as gpt yet, and it seems it may need a lot more context just really understand what’s going on. Tbh, only way to know for sure is okay with it myself but I guess I’m waiting for the next update before attempting to move over, although I really hope gpt 5 increases their context window by then..
Which I’m not sure if they will because they’ve even shortened the context window for gpt 4 lately.. which is definitely unimpressive
@@MartinDlabajaGemini Ultra 1.5 is not yet released, it can't even read the list of films I've seen, let alone millions of codes. And Gemini currently can't even understand that the question you just asked is related to the one earlier prompt before it.
sure, I know, but what I am saying is that this is temporary hurdle which wont be there in few months@@berkertaskiran
AI can also write books and articles. You still need people to look at the output, understand it and adjust. Same thing with code. Also, writing code is only a part of building software.
You're a bit behind on the spectrum of AI if you still think that written AI code needs human inspection. Even at this time AI can do that better than any human in terms of accuracy. Things are changing, and they are changing at a real fast phase. The info that you knew about AI weeks ago, today can be miles ahead and even further with each passing day.
@Sudegink It does. It makes some mistakes even when trying to solve simple C++ requests. It's good at explaining concepts and teaching how to code, though.
Who will blindly trust generated code?
Have you read anything that ChatGPT writes? It's utterly insipid. Funnily enough, good writing might be the one skill that humans have left to them, if just for good prompting.
@@zootsoot2006 So far. But the stuff that ChatGPT comes out with today is better than the stuff it wrote when it was released less than two years ago. Right now, it's a sophomore creative writing student. It's still learning.
The new layer is software technician. This kinda reminds me of something historical like an employee who was for example a steel parts fabricator or a mechanic a few decades ago. They used to be generally well paid because of their technical knowledge, experience and training. It took a lot of them to produce or fix something. Historically these people thought they couldn’t be replaced because their skills and experience in directing a machine or using hand tools to produce a part couldn’t be replaced - or so they thought. But automation changed all of that and many of them didn’t see it coming. Now proportionally speaking many of those jobs are just gone and what’s left is nothing like it used to be. Now they’re at most technicians that monitor the process or click a few buttons to make a part - requiring much less skill, lowering labor cost, and generally lowering the cost of production. Software developers are now becoming the next group of these workers. It seems this will automate many out of a job. Because of all this I’m getting more worried about my own job in the mid to long term and there’s nothing as a group we can do to stop it and the tools that replace software developers are just in their infancy. But unlike the past, the change is coming much more quickly the impact for many will be abrupt and harsh. I’m scared.
One thing about code generation is, it works great if you tell it things like "write a python pytorch image classifier". Thats. great if you know what all those words mean and why you'd want AI to do that. And you also know what to do with the code after AI writes it. So you still basically need to be a programmer. Just to pose a worthwhile prompt, and use the generated code.
Sure! Unmaintable codebases that nobody understand... Good software engineers have bright days ahead.
"Assistant, write a code to sent a rocket to the Mars". Keep fingers crossed, what could go wrong?
Remember over the pandemic when people who knew legacy languages were in demand.
A non issue. The AI will maintain it aswell.
Assembly code is not really a layer of abstraction. It's symbolic, and also works on more than one architecture. In any case, it's relationship to machine code is completely different from the layers of above it.
I'm a software developer/DevOps professional turned software architect. I've been in the industry since 1999 and have adapted constantly. I agree with you - AI will eventually replace classic software development, but I'm cautious about predicting when. People often underestimate the trade-off between abstraction and efficiency; coding won't disappear entirely because of this. Additionally, AI is incredibly energy-intensive. I'm unable to make the maths on my own, but the amount of energy it would take to replace all the traditional software in use today with AI systems should not be overlooked. Unless there are major technological breakthroughs, it'll take longer than many think for AI to become truly accessible for everyone. You can add to that that social transformation is always slower than technological advancement. I'm sure I will need to start reinventing myself again soon and it won't be easy, but that is how this industry has always been.
The cost of inference is currently the biggest obstacle for the deployment of AI, mainly due to limited hardware availability and energy costs. In the coming years, however, new architectures specifically for inference will come into use. These are much more powerful and consume less energy. This will be the moment when the tide turns.
"Unless there are major technological breakthroughs, it'll take longer than many think for AI to become truly accessible for everyone. " Not really sure what you mean? It is for free on any phone with a browser.
@@MartinDlabajaliterally "Cloud based AI app" am I a joke
lolol Go ask some laid off Google engineers, who, by the way, could code circles around you and now are collecting unemployment checks.
if you're in software engineering circles, you would actually know that it's economy related. I believe it will someday take many software engineering jobs but if you actually work in SWE and know what you are doing, then you realize it's much too early. AI hype bros are so obnoxious lol.@@JohnSmith762A11B
If anything, all of us have been amazed / scared about how fast all of this has been. We are in the infancy. I'm not sure the
innovators know really
where any of it is going.
they know. openai has been slowly leaking what they have to the public. behind closed doors they definitely already have AGI. maybe even ASI. then again once AGI is achieved (which it is) then ASI is certain.
What that person is doing in the video is for small apps.
For big apps, people will have to first discuss with the AI to draft a design document. Then the designer will go through the documents and then amend it again and again then the AI will code the software. Then the human will tell the AI to write testcases to test the app. Then the AI will run the testcases and then tester will observe how the app is running and is it working as intended. Then iterates again and again till it works perfectly.
All these will then be run with agents specialised for such tasks. And humans will be totally not be in the picture.
The final limitation is that people cannot keep the context of a larger application in their head, so they will not be able to give instructions for features that are internally consistent. Their instructions will inevitably end up being inconsistent with what already exists, on a fundamental logical level. That cannot be solved by the AI, but it might be able to catch it and help humans change their expectations on what is actually going to work. Essentially, what tech leads and architects have been doing for product managers for ages. :)
AI is the odd man out here as AI cannot conceive a total concept. Humans can dream and quickly conceive concepts to an end point. AI will never be able to do that.
@@comatose3788 People that say "AI will never be able to do " have been proven wrong time and time again. There is no reason to believe that AI won't eventually be able to not only do everything a human can do, but many things a human cannot do. Also it will do it faster, better and cheaper -- the engineering holy grail.
@@pvanukoff I understand that. But not in this case. This AI you're seeing now started life in the late 50s as a chat bot. Most of this is still smoke and mirrors. AI is the dumbest programmer I've ever seen to be honest. At a child level with no real comprehension at all. AI can show 20 ways to make a circle and never understand why you're using a circle. Also, like me telling you "AI started life in the late 50s" or "smoke and mirrors". You and I can communicate with each other at a higher language level, as you understand metaphors. AI wouldn't be able to make any real sense of them statements. AI is not smarter than a Human and never will be. We can transcend meaning, AI can not.
Thanks for putting these videos out! I come from the solution and enterprise architecture space but have been doing data science for 4 years now. After using copilot for the last 8 months I’ve found myself doing almost no coding but, rather, providing guidance and letting copilot build. For the first few months I would triple check copilots work. Now I briefly review and focus more on thorough automated test scrips. I also have copilot document the code, create debug scrips and even solution training manuals. I essentially no longer code… I guide AI.
This is my first year studying how to code, i don't know how to feel right now. All of a sudden the motivation is gone
Yeah me too, entire 2023 I was learning code and was really proud of it but chatGPT happened somewhere around June and man I feel like all I learnt is now useless. I was planning on going to college for SWE but now I've decided to do something else.
I will continue to learn programming as a hobby but no way in hell Im choosing it as my career! It already takes years of junior dev work to get to a good position, imagine how hard it will be in the future when not as many developers are needed.
As a programmer of decades, I've tried really hard to get GPT4 to write moderately complex programs. Its surprising how helpful it can be. However, the real skill is in
problem solving, design and testing and those steps are a long way off being automated. As others have said, this is by far the biggest part of the job. AI is a useful productivity tool at the moment, nothing more. Its possible the limitations can be overcome by AI but I don't see it happening for many years.
These are some basic limitations I've found.
1. AI needs such detailed guidance that it takes longer to write the prompt than to code it yourself for anything real world.
2. You have to manually design the solution and break it down to very small steps to stand the remotest chance AI can code it. Even when you do, it really isn't worth it. The prompt does serve as useful documentation though.
3. Almost any real world problem is too complex for AI to design, code and test.
some points I want to mention
1 - not all current programmers are good problem solvers, many only receive a request and program a piece of code without understanding the full context of the problem
2 - Programming is not equivalent to painting, the end client does not care much if you programmed it in C#, javascript, or pearl, they care that it works and meets their requirements, in general a client does not "delight" with your lines of code
3 - If today you have 300 programmers concerned about a piece of code in a broader context and only 30 people who understand the context and are ultimately the ones who "solve the problem", those 300 programmers will mostly become obsolete and easily replaceable by an AI in charge of these 30 people plus a select handful of assistants
4 - Programming can continue to be something that someone wants to learn, like today someone can learn to paint or play the piano, but that does not make it an economically viable profession as it is today, it can go from being a source of livelihood for many to a hobby for some and a source of livelihood for very few
Will the coding be dead? probably not, but that doesn't mean it will be a profitable profession.
Right now there is a mountain of programming work to be done interfacing LLMs to traditional computing, then "interfacing" LLMs to the physical world - robots.
Unless good methods can be developed for self interfacing LLMs - function calling advancements, + function writing.
Three minutes in and I love that Matt sounds like he knows what he’s talking about and talking about it. Historically I hope it continues this way and he shoot down his perspective by talking his book or that of his sponsors. Thanks Matt!
I write code all the time and I’ve been using AI for a year trying to see how far I can push it. AI at this point is helpful for small projects and small blocks but it doesn’t scale up.
I use it as a much faster alternative to Google searching snippets. But it takes domain knowledge to use it for actual projects. And a whole lot of manual rework and very specific instructions. I dont see a non coder getting anything more done than some small example projects just using language models.
But you also gotta remember AI is in its infancy stage so it’ll only get better
Fools. What's publicly available's just for testing. It's not the best thing out there and it'll never be as bad as this anymore, it has no limit to its development.
First airplane couldn’t fly for more than 12 seconds, now they can go to space, and use it for a lot of other things. Satellites, GPS.
@@Ilamareawhoa calm down. Op clearly said "at this point"
The role of a software engineer is to solve a business problem in code, not to just write code. The oempa loempa in the video saying 40% of code on Github is written by AI, failed to say that the what was written had to be extracted from the client by understanding their business and then find the root cause of the issues and then tell AI to do it. The tricky is part is to "tell" AI what to write, writing code is not the hard part, knowing why you are writing it and then that it actually solves a real world problem is not so easy.
lmao, like that isn't an intrinsically natural language-suited task
Of course you have to consider register, like with any language... but you ignore that this is exactly what models are already capable of and how hard they stretch to reach the (admittedly high) ceiling. I'd get the skepticism if we were talking about a one-year timeframe, which is still so much more feasible than we ever imagined it to be, but five years? You've got to be utter garbage at betting to put up any amount of money against that assumption. Of course coding is going to be abstracted down towards natural language interface schemes. Knowing "why" we're writing it is going to be such an irrelevant question too - despite certainly not being an easy task, at least initially.
If we got to the point where we pretty much ravaged the entire translation and text production-field, advanced code monkeys aren't going to leave us waiting for too long.
soo.. you still won't have to know how to code?
You make it sound like that’s such a difficult task for AI or something. If anything that’s the easiest part of AI’s Job. Querying users, Problem solving, and planning accordingly.
cope harder, daddy
devs and coders will be STILL needed in the future. the BIG ISSUE is - how long is that future going to last from now. 5 years? 3? 1 year?
Denial, Anger, Bargaining, Depression, Acceptance. There are Soooooooooo many people still stuck in the Denial phase when it comes to A.I. taking their job.
Well, currently I am in the process of thinking of beginning to learn ML and AI (as a programmer). I have programming experience since before. I don't know if we need "yet another model" or perhaps if I should focus on using models for problem solving. I suppose it's a question on what "layer" it makes sense to get educated/experienced in.
The problem with coding assistant is they can’t yet build other UI alternatives and multimodal creative actions, they can’t build a coherent UX grammar as well they can only do functions not creative programs yet
I see it like the full self drive where AI will be able to cover 75 to 80%, but that last 20% will remain difficult and need experience coders because this is unique situations unique challenge that the AI doesn’t understand
Computers are already highly abstracted with excellent UI/UX most of the time. However, many people, including those from the younger generation, struggle to use them. You can abstract and simplify something to the extreme, even with natural language, but it still does not mean that everyone will be able to use it. We already have all the necessary tools to create a better world with computers. Are we making it?
I'm also pretty sure that AI will replace traditional coding and greatly simplify the development of digital products. There's a 99% chance that in the future, entrepreneurs will be able to generate their MVP app in 5 to 10 minutes. There's no doubt about it. But again, those tools already exist in a way with low-code/no-code platforms. Having a complete overview of what software is will still be valuable. Perhaps the developer will evolve from programmers to full-range product owners.
And in the end, when coding becomes obsolete, the whole concept of software will too. Nowadays, a small device in our pocket packs hundreds of features that were handled by full-time jobs in the past. In the same way, we are moving towards a road where big companies (OpenAI, Google, Nvidia) will have a monopoly (because of computing power) and will provide all-in-one, self UI/UX building apps that will completely annihilate the purpose of most utility apps.
Singularity is a one-way path. Choosing a job or a skill to demonstrate its future obsolete state is a waste of time. Everything, including our reality, will become obsolete.
I've used GPT-4 since last April for coding (10 months ago). It's great at remembering common things like how to change the CSS to move stuff around on a website or translating from Python to Rust. It's also great at writing tests. However it can't reason yet, so sometimes it takes a long time for me to try to convince it to do what I want. In this case, it's faster to write it myself. Right now, complex coding tasks need to be heavily curated, but simple tasks can be done by the machine. It writes most of my comments, then I tweak them, deleting half of the docstring and correcting a couple things. Overall, coding with an AI is more efficient and more interesting.
I look at it as moving up a layer in abstraction (or down a layer).
Almost no one programs in assembly language because of the languages already built above the instruction level, now we're just moving to a new phase where you no longer need to know the base components of those languages as intricately as years prior.
The hierarchy just moves up.
Project management is going to be the field we all need to live in. When the AI can do coding and any other computer work, it'll be our job to decide what to build. Humans will design games by coming up with a cool idea and then working with the AI to make that vision a reality. For video, everyone will need to become a director as they tell the actor and editor AI what to do.
Oh nooo, more project managers 😂
yeah agree new breed of PM will have to have coding experience.
Nono, everything will be done by AI. That's where its going :D
there is not enough employment as project manager, and with the oversupply the wages will plummets
@@geesus2963Can be this way, everything done by AI, but a AI will not be able to overtake a AI that works with a Human.
You mean like we should have flying cars by now? When a technology comes out it's easy to over estimate how it will end up.
Thank you for another great video, Matt.
I am not a coder but intuitively, I agree that AI coding through voice/text prompts is the future. I also agree that such AI tool will be just another step above the programming language in the Layers of Abstraction chart. However, I don't necessarily see it as transparent as other items on the chart are. What I mean is that if a human uses a programming language, they understand it (ideally) and if they are not getting the results they want, they can debug their work. They can even make changes in the language itself or develop a new one! There still will be people who will learn to do so but all the others, without the knowledge, will be totally dependent on the AI tool and won't necessarily know how to achieve a different result if the tool doesn't give them what they envisioned.
At the moment, people who are involved with any of the items on the chart understand them (each his/her own field). This will not be the case with AI coding.
In summary, if the tool breaks down, there will be no coding (for people who didn't learn it).
Still, I am super excited for this development because even if I will never understand what it is doing under the hood I can get much better results than I can currently achieve with my zero programming knowledge.
You seem to have a dead pixel (it can seen clearly on the rim of your cap, at 12:23)... I thought my laptop screen had gone bad ha. Is it some weird watermark or just a bad pixel from your camera?
Coding will be increasingly widely accessible ("democratized"), not eliminated. There is also still the problem of clear communication & discernment of output (prompt + flow engineering), which itself will continue to be a skill so long as we have specific needs and need to communicate them. Creating a shortcut and a faster path doesn't mean nobody will take the scenic slower route. There is also massive benefit to thinking like a coder (modularity, extensibility, naming things well, creating good abstractions, etc. -- these thinking models spill over into other areas of life every day for me).
I mean, coding as we know it will definitely be eliminated beyond recreational projects - or at least turn into a true human relic, what with it still using natural language to some degree. I don't think we can't and won't still learn it, it just isn't really going to be a requirement.
What if someone makes a program with AI in it that asks the layman specific questions for what they want?
Literary was going to write this.
@@ghhdgjjfjjggj If we had that in the 1800s then we would have faster horses today instead of cars.
It will be so democritised and accessible that the customer will just """code""" by telling the AI what they want.
As a senior developer who embraced AI for coding. It will give you enormous productivity boost sometimes and sometimes it removes all that productivity and you will be stuck in debugging it’s generated code because it can’t add the features.
It’s mostly very good at doing basic stuff or very specific. But it can make you crazy.
And yet, none of those programming languages have managed to remove the requirement of understanding "logic" in order to use them. Even if you could remove the language and the syntax, how could we remove the logic?
It seems like it's all about knowing what you want, then knowing how to ask. I was writing some basic javascript. As usual, it was taking me a long time because coding isn't my strongest skill. I asked ChatGPT instead. The output was a little off so I asked Bing's AI. It gave me exactly what I needed quickly. I was elated but also kind of terrified.
As one of your commenters explained, AI will need requirements, especially creating untypical software. So will there be a high demand for people who know what to ask?
I've been using Midjourney a lot lately. The prompts almost feel like the old DOS-style commands in a way. Will there be a need for UIs that make asking easier and more accurate?
We must push even further, with artificial intelligence, everyone can be a programmer....but also an engineer, doctors, lawyers etc... and soon a plumber.
can i just be god?
Coding is not just a skill.. it's a fundamental way of thinking. Breaking down a large problem into procedural steps. Just because the syntax changed from C# / JS / C++ / Java into English doesn't make it dead. Programmers today still have a massive edge in advanced prompt engineering if they understand how to use their coding skills to leverage LLM's to build super advanced prompts. So no I don't believe coding will fully die, it will change
I think this is awesome. Right now coding a new project is 95% planning and 5% coding. With the biggest problems being that people normally don't fully know what they want when they first request the project. this just farther pushes the programmer to be a expectations manager then they already were. Cool.
Thank you for your informative and fun videos, Matt!
I wrote a consumer interface for our new line of safes. It is in PDF format with images and words. Pretty primitive and hard to edit or make changes. I do not have a team to consult and brainstorm. (Someone else is doing the coding based on my writings)
What AI could I use to format it differently so I can make changes easily?
And what AI can I use to run it for errors and improvements?
Thank you!
I've worked in many large corporate IT environments. The amount of legacy applications out there still being used shows me that AI isn't going to be replacing human developers anytime soon. Maybe a junior level developer or for tedious tasks.
Totally agree. In company Im currently working, we got software, that was created in the middle of 2000 inside this company. I heard like 6-7 years ago, that the plan is to get rid of that next 3 years. And here we are, after 7 years, some new projects are working in a better soft, but some old projects are still using this old software and I dont think so it will change next few years, There is too many of dependencies there, complicated logic, based on which this company exists and somebody needs to maitenance it.
Also, there are some areas, that wont go into AI so fast from security reasons, like banking systems. A lot of banks dont want to use clouds, becase they are afraid that somebody could steal data from there, so generally I think developers will be still valuable assets for most of companies, because they are resolving a lot of other people problems. AI can just make they work much much easier.
@@pbanaszek Where I work, we have begun to transition from an old COBOL platform built in the 1980s. The timelines have slipped a couple of times, and I don't seriously expect the new design to be operational for a few more years. Today's web developers are still churning out the legacy code of the future. There will certainly be work for another generation to maintain and patch software.
AI will be able to break passwords and encoding. I'm sure it can handle some legacy code. Are you serious?
you think that AI can't understand legacy code? HAHAHAH
if AI replaces all the junior level developers than how will we ever have new senior level developers?? where will they come from? do we expect the current ones will live forever?
Software companies with SaaS models should actually be far more worried than programmers. A good developer with a smarter AI could quickly create software that competes with the big companies. So the future will be more smaller companies selling software directly.
Nah, it'll be big corporations with infrastructure and robotics. AI will be fully automated. Developers won't be needed, the corporations will pump out 10000x what even the best developer could do with the best AI.
@@Ilamarea why would the “big corporations” with no developers be able to produce more output than a developer using the same tools?
@@JoePiotti Because for anything useful you need infrastructure and investment, any original idea will be instantly stolen and copied a 1000 times and ultimately AI will work completely autonomously, replacing even the decision makers and stock holders of those corporations.
@@Ilamarea the comment was about SaaS companies. No robots needed. Not much infrastructure either. It’s just software. My point was about proliferation. The way RUclips has made making video content easier so too does AI for software. And although there are still big studios, they don’t have a tiny fraction of the audience they once had. This too will happen to software companies. Then you are fast forwarding to where basically you have 1 person, the owner, with AI that does everything. Yeah, that’s my point. This 1 person doesn’t have to be a developer, but in the near term a developer or even a product owner will be better at creating software with AI that can compete with these companies that are going to be in a race to shrink.
No you will not be able, anyone can build a facebook social network alike, but you can’t compete, anyone can build an online market place, but not everyone will compete with Amazon, or eBay.
Great thread, Matt. Interesting to think about what the next layer of abstraction will be and how we will continue to develop/grow to build it. IMHO, AI, when all is said and done, is a tool we use to create what's next.
I managed to code(with gpt helping) an automated system that opens the Pi chatbot and talks to it by asking about various concepts and recording answers in a txt file. I was impressed by how organized the txt file is, here are some examples:
[Bot]
Concept: "self-reflection" in AI systems. This is the idea that AI systems could be designed to reflect on their own behavior and performance, and use that information to improve themselves over time. This is something that humans do naturally, but replicating it in AI would be a major breakthrough. Self-reflection could help AI systems become more efficient and effective, and it could also give them a kind of "awareness" of their own capabilities and limitations.
[Bot]
Concept: "emergent properties" in complex systems. This is the idea that new properties or behaviors can emerge from the interactions of the components in a system that are not present in the individual components themselves. For example, the ability of a flock of birds to change direction quickly is an emergent property that arises from the collective behavior of the individual birds. This concept is particularly interesting because it highlights how the whole can be greater than the sum of its parts, and how new and unexpected properties can emerge from simple interactions.
Kind of interesting that the issue with things 'getting lost in the middle' is a known issue with human memory, too. (the recency effect and the primacy effect)
The middle child😂
@@sharpvidtube😆
Denial, Anger, Bargaining, Depression, Acceptance. There are Soooooooooo many people still stuck in the Denial phase when it comes to A.I. taking their job.
yup...
I'm in depression and see no way forward.
@@bliantfive You're not alone. I was a writer for a newspaper column. I just got laid off because of chatgpt. It can write much faster and better than I can.
@@ghhdgjjfjjggj that's really hard. Keep your head up for now. Even if it's hard. There must be a way.
Acceptance! 😂😂😂 It is what it is. 🎉😂
I am a coder. I started coding in the 80's and there was one Cobol coder in the shop. Cobol is still present in far more places than you might imagine. I don't say there will be as much new development, but when there's a vast code base in place, it is nearly impossible to replace it with any accuracy or speed. Coders and coders with AI skills will be in high demand for the near future
As a programmer of 20 years, I agree with Jensen and Carmack. The more important skills moving forward are design, logic, and problem solving, and knowing code efficiencies will make your code better than the rest.
but wtf do you study as a career then. There isn't a career of problem solving yet, engineering perhaps?
That's the main problem with AIs currently - they can reach like 99% (and do it even better than humans) but then fail miserably at 1%. For example, AI might miss a thing that I can find in a document using a simple Find command, and it might even be not that huge text.
Today I found an ugly old piece of code and asked Bing AI to refactor it using specific framework functions. It created a nice working code with comments. Then I asked it to shorten the code by removing comments and getting rid of intermediate results to merge everything into an efficient one liner. And Bing failed miserably - it removed not only the comments, but also some parts of code, and when I asked it to add the specific missing piece of code back, it made it even worse and it did not even compile anymore.
So yeah, that last 1% might turn out to be the roadblock, the "uncanny valley" of programming.
Also, if AI finally learns to do it completely right and asks all the right questions about the business logic and the infrastructure, and performance, it won't need to generate a human-readable code at all. It will generate raw and super-efficient machine code. Finally, the apps will become much smaller without all the bloat of the intermediate layers.
Is it one step forward, two steps back?
Or two steps forward, one step in a giant hole that takes you weeks to get out of?
I know c# and I've used Unity for about 20 years, but now, Soon, very soon, I'll be able to make the game of my dreams 😁🎉
That’s awesome!! What kind of game?
@@kitlmao it's complicated to explain it would be an RPG game where the factions are controlled by an artificial intelligence and these factions have characteristics such as the health of the people who live in the cities, such as the resources they have available and each faction generates quests based on current statistics . furthermore there are reputations between the factions and therefore the factions can be hostile to each other or not. Quests are generated dynamically, so it is also possible for a faction to carry out a mission against another faction, thus increasing hostility or pacification between the factions. in addition, the scarcity of resources can affect the health and morale of the NPCs living in a fiefdom, generate monsters or other missions. in this way the player, despite not being the typical hero of a predefined story, can little by little influence the entire world in which he lives (plays).
The fun part is that the game logic is quite ready but I've never been able to end the game itself because the actual game, like inventory, combat, and AI that move the npc 😅🙄
As usual, your perspectives are super informative and very appreciated.
Best,
Carolyn
I am 27 years old, and I've been working on repairing computers since I was 15. By the time I was 18 in 2014, I knew how to program, but when I tried to enroll in Software Engineering, my low proficiency in mathematics forced me to give up due to financial constraints (the university was 200 km away from my home, and I couldn't afford to live there). I studied graphic design due to a lack of alternatives, worked in that field, and three years ago, I decided to return to the programming world through self-learning. I feel like the opportunity has passed, and my life is heading for a complete disaster. I am from Argentina, where the situation is tough, and I had hopes of making progress as a developer. Currently, I work as a UX/UI designer and Front-end developer, but I can see that it won't last as long as I thought, and in five years, I fear I'll be back to square one, jobless.
Devin is here 😅
I started coding classes right before chatgpt was released and I'm glad I had the first semester without AI to learn the basics. And get a get good foundational understanding of the language and process before I was introduced to AI it has made me 10 times better. Then I would be without it, and I'm grateful for it.
Loved this video and feel like it’s one of your better ones for educational value, and organization of the points you made for easy understanding. John Carmack is a legendary programmer and inserting his perspective was very informative.
Thank you Matt for your kind words and for sparking such an important conversation! As a Project Lead and Senior C# .NET Developer, I've seen firsthand the capabilities and limitations of AI in coding. While AI, including technologies like GPT-4, has become an invaluable coding assistant, helping us to explore design patterns, architecture, and even tackle some coding tasks, it's clear that AI is not poised to replace human developers, especially in complex scenarios.
Developing enterprise-level applications requires a nuanced understanding of not only coding but also business logic, user needs, and security-a realm where human expertise and critical thinking are paramount. Furthermore, the ethical and regulatory implications of AI-generated code, particularly regarding accountability for security flaws or data breaches, present significant challenges. These issues underline the importance of human oversight and the collaborative synergy between AI and developers to navigate the complexities of modern software development responsibly.
Your points echo a vital message: AI is a tool to enhance our capabilities, not a replacement for the creativity, judgment, and accountability that human developers bring to the table.
Calculator gets invented: "In 5 years there will be no mathematicians."
E-mail gets invented: "In 5 years there will be no postmen."
Google gets invented: "In 5 years there will be no doctors."
These people saying this kind of nonsense are sitting on top of thousands of developers, which are responsible for building the very tools they're trying to brag about.
It's very naive to think of "replacement" when in fact developers have by far the most benefits of it all, the more advanced it gets.
AI is not replacing devs, are actually giving them superpowers.
Thank you.
Terrible analogies. No one ever said that about calculators or Google, and, yes, post has been almost entirely superceded by the internet for communications
AI gives everyone superpowers--but most especially the companies or governments that actually control the most powerful models--that's the whole point
but it gets people to click and then be super emotional in the comments!
You will still need to know what you are looking at if producing raw code for now (until it gets so good you don't need to see the code). Still, calculators replaced the slide ruler, Google replaced so many things, and postmen have been reduced by more than 70% and will eventually be no longer needed! Junior developers will be the first to be replaced. Not everyone can pivot to AI, and there will not be a need for as many - just like the postman.
I made a simple chat room today using chat GTP whit no coding skills. It took about 1 hour.
Congrats, but the operative term is "simple." Very few apps are simple. Most of the big ones we all use are as complex or more as a city skyscraper.
Now go deploy it so people can use it, only using ChatGPT of course.
@@jason_v12345 the big ones are complex yes, but they are not the majority of apps. most apps are super simple like a guitar tuner app, or an app that showcases a business, etc. these will all be replaced
Amazing...no one said it couldn't do that, but does it fit into a business problem a client had? No it doesn't you just have a chat room, something anyone can create on whastapp using groups.
It's like saying I created a facebook
page using facebook
@@thehumanbagelcalm down… soon AI will be able to do it all easily.
Thanks Buddy, I learned something new today, and have an idea for the future!! Best Wishes
I've done coding and this just blows my mind! The thing is developers still have an edge because they know the questions to ask and sequence.
English is the new programming language.
I'm a terrible programmer and let me tell you, none of this stuff has helped me with real problems. Especially if it goes a bit deeper. Once it gets even a bit more complicated, you can through all these tools out the window. I can only hope it will get better, but for now it's useless if you want to do anything serious.
did jensen always have the microsoft sam version of AI voice generation? or did they morph naturally
Nowadays is very common to work in projects with very few people, 8 programmers for 5 months is already a large project, when my dad worked in the 1980s a big project needed hundreads of programmers, so we are already reducing the amount of human job necessary so much, in the end only project/product management will remain everything else can be automated
The context window is the most important in order to start the replacement of developers
I think you got close to this point when you were discussing layers of abstraction at the end there - most software is written for humans to interact with. If AI gets to the level where it is better/faster/cheaper than humans at understanding requirements and writing code - it's probably at a level where it can replace the jobs of those using said software in the first place. As an example - CAD why would we need Fusion360 and the folks who use it, if you can just ask an AI to give you the G-code for a widget directly - even more abstraction.
Learning to program and gaining experience by actually building, iterating and fixing bugs rewire the brain to think in a certain way, which is the fundamental skill that will be required in the future. I would tell students to definitely continue to learn, to continue to build. Carmack has it right.
Would the correct thing to do now as a beginner be to get started in Machine Learning instead of trying to learn front end developing?
Great analysis, it's only shortcoming IMHO, is that though you well reasoned humans are not out of the loop in the near-ish future, the conclusion didn't touch on the reality that the amount of coders needed in our future is now set to dwindle, this little bit is crucial for the present and leading generations to ponder.
Who is going to build/maintain coding assistants and these gen-ai? How many software engineers does OpenAI employ?
Guys has there been a massive development in face swapping ai side ? Is there a strong tool that can let me replace face on a game of thrones tv show episode?
Yes Matt, but we need to keep pool of competent engineers and code talkers / writers who understand the machines. It's not enough and not safe that it's one way - machine understand us but we cannot understand and scrutinize them. There should be reasonable pool of the best and hopefully ethical ones.
In that case, I think it would be wise to teach everyone some code, make it part of school curriculum or something. If everyone had at least the basics down, more people would understand the machines at a better level, we could defer to the experts when the hard coding questions arise... But I assume this is a 5 year deal before the machine is perfect.
There's still a need for horse trainers and horse farrier too. Just not as many are needed now....wonder why?
I don't know, unless we force the ai to keep using abstracted languages probably the most efficient way to run command will be on a machine language or something too complex for us to keep up
we already don't understand how most things in the modern world are made from scratch... I think it's gonna be like that with coding as well.
Oh well. This video and comments helped me a lot to understand the current situation. I can eventually become the program manager of my " AI bot developers". But i need to good enough to design, communicate, and instruct them. I also need to keep sharpen my coding skills. I believe when you are a few that dont give up and understand code better than a lot other people, you are the guy to keep/get the job.
I wanted to develop one simple shell script, chatgpt was helpful initially, but when I asked to modify the code with some additional parameters, it gave stupid code.
Absolutely loved this video! It's a great breakdown of how AI is changing the coding landscape, and it definitely sparked some new ideas for me. I appreciate how you balance the potential benefits of AI tools with the reminder that strong problem-solving skills will always be crucial for developers. Of course, the real question is: will you be offering a paid course to dive deeper into these AI coding assistants? 😉 I'd definitely be interested - you've got a clear and engaging teaching style!
Looking for a deeper dive on this. Works within simple domains or small projects. But code output is much easier to edit than say pixels that form moving images e.g. film or even 3D spaces unless built in a 3D program. Anyways, it’s exciting and opinions are going to fly in from all perspectives.
i personally meddled with coding as well and managed to make some basic scripts for my own use during work. however, large batches are definitely out of the picture. this applies across the board in fact for Chatgpt. if it gets too long, the AI simply can't comprehend the context of what you want and goes off script and writing
This is edge of your set suspense. I can't wait to see what comes next. Imagine a coding leap like that which came with Sora. The game changer could be a month away or a year, neither of which is long to wait.
@Matt Wolfie are you a computer programmer?
Considering these trends, is essencial for the computer programming industry to adapt and strategize for long-term:
1. Upskilling: Encourage continuous learning and upskilling among employees to stay abreast of technological advancements and remain competitive in the evolving market.
2. Diversification: Explore opportunities beyond traditional programming services, such as AI development, cybersecurity, or software testing, to diversify offerings and cater to emerging market demands.
3. Collaboration: Foster collaboration with other tech sectors and professionals to leverage synergies and stay innovative in a rapidly changing landscape.
4. Market Research: Invest in market research to identify emerging patterns, consumer behavior shifts, and technological advancements that could shape the industry's future.
Go comes from Python? Should I switch? Is Go still OOP? Edit: I think I misunderstood the chart.
The operating systems of the future will generate apps as required, it is already shown in the Google Gemini launch video.
Also if the OS can find solutions to our problems in real-time and we just have to input the problem in natural language, what will we be coding for?
Thank you for sharing this video. You actually laid out a very good points regarding the issue. I love how you share with us on the abstraction layers and the concept of programming and how it all started.
I personally think that the whole thing of learning to code/computer science, is just like people learning to conduct Science Experiments, yet we are not saying learning how to conduct Science Experiments are waste of time, just because there are more advance ways to study Science. We are also not saying learning Math is waste of time just because there is Calculator to do the job.
The whole point of learning is to have the understanding of the knowledge, exploring concepts and instill the humane value to it. Technology depends on the value it brings to the people, and we need PEOPLE to build the value FOR THE PEOPLE.
At this point, we are to redefine what is valuable for human....
"There are plenty of tasty delicious food that can be made or bought...but still cannot beat having even a bread with your loved ones"
I second Matt's critique. Among other tests, I tried both ChatGPT and Gemini for making a sheet that calcs trailing stop loss %s for a hypothetical investment portfolio. It is very demanding of a sheet, though small potatoes compared to most coding projects. Both LLMs failed repeatedly. I believe it's related to the known math weakness of current LLMs, and, more broadly, the inability to use math and logic well over multiple steps within a formula (or code). I tried to include the ChatGPT thread link, but seems RUclips doesn't allow such links (?). Of course in the (near?) future, LLMs will be able to perform reasoning and math well too