GREAT DEVELOPER HABITS PDF GUIDE: Tips on writing clearer code, managing complexity, sharing your ideas with others, maintaining the freedom to change your code, reducing future rework, making mistakes and being happy when you find code you can delete! Get your FREE download HERE ➡ www.subscribepage.com/great-programmer
Software engineering is fundamentally a social activity. Gleaning requirements, creating metaphorical representations of real-world things, that is comprehensible to one's colleagues, focusing on usability, attending meetings, deliberating and sketching out ideas, assessing trade-offs ... the list goes on. AI might have a place inside IDEs, to make the code more reliable, but it wont make it more relevant. That is where the human touch comes in.
But the fun part is precisely the technical part. If you replace that part and only leave the social part, what distinguishes you from other professions? or rather, because it takes away the fun part that is the technique and leaves only the boring (social) part. I want to be in front of a computer and not have 747383 meetings with people I don't like
@@bobobo1673 It’s not as binary as that. Perhaps it would be more apt to say that software engineering is a socio-technical activity. While AI can produce pieces of code, it cannot make informed architectural decisions, which requires straddling both sides, nor understand the domain for that matter. Not at least until we have automated general intelligence, in other words synthetic consciousness. What I think AI will give us is better tooling, error checking, production monitoring etc etc. It will not displace software engineers, it will augment them. This is the difference between automation and autonomation.
I have retired now, but I did code for 35+ years. Copying code from another source is a great start, but I still needed to 'Tweek" it to meet the requirements of this project. Another thing to consider is just because this code works, does that mean it's the best code to use? Several times I've gone back to what the user actually wanted and turned everything around to find a faster and more elegant solution. One thing that wasn't discussed in the video is error handling. I've seen far to much code that assumes things will go well and don't plan for, or look out for, errors. I have been assured that "This will never happen" far to often. I code messages for these situations, usually with my supervisors name and phone number. It is a fun conversation when said supervisor calls and asks what this message is and why it came up. The video author is absolutely correct: we are paid to solve problems. What tools we use is up to us.
Not to sound too elitist, but like you said, the idea that these AI tools would "replace" programmers is if you're misusing programmers as glorified typists instead of problem solvers. It's best when it's being used to write tiny trivial bits of code too small to justify using a library. I do wonder whether these tools might ultimately backfire though, say Stack Overflow ends up shutting down because everyone's going to ChatGPT instead, then where are people posting the information these tools get trained on in the first place? How many people will volunteer useful information when any ability to receive compensation is effectively stolen away and instead handed to OpenAI or Microsoft?
It will be interesting to see how copyright and license shakes out. LLMs currently ignore it, and whatever it spits is claimed to not be tainted. How similar to the source does the generated code have to considered a straight copy? As you training the tools while paying a subscription fee for the privilege of using it?
There is nothing fundamentally wrong with what you say. When a company offers a service and another one steps in to offer the same service in a better way, then that is when you need to: 1. Improve your service. 2. Offer a unique service that your competitors don't offer. 3. Close for good! That is the world of industry. I personally think that Stack Overflow is getting to it's limit of usefulness with tons of ridiculous questions and full of outdated answers, that make it difficult to navigate and find good answers (just my opinion, you might have a different one). So having instant good answers through AI is faster and, in some cases, more reliable.
@@Andrumen01But that's the catch, LLMs are just digesting and recomposing text they've seen elsewhere, they have no concept of programming, you can't just give them the technical specification for a language and expect them to provide meaningful answers on how to build any program you want from it. If ChatGPT leads to Stack Overflow and developer blogs and tutorial sites either being shut down or paywalled off, then the source of information that makes ChatGPT actually useful shuts down too, leading to it becoming less useful and a nasty cycle could result.
The idea that stack overflow is what the models learned on is absolutely asinine most posts on stack overflow is "HOW DO I DO THIS BASIC SHIT" followed by "Removed for duplicate post." As for where you get trained on, you RTFM aka the documentation, and the internet is a constantly growing library of free resources. You act like if AI replaces a couple forums of rude people that suddenly all the information is gone. I pity anyone who was having to suffer through those websites in the first place. Take a course, read the manual, solve personal problems with code, build a few projects; I've spent months trying to decide how to get the most use out of the resources on the internet and none of the actual materials for learning that successful people suggest to me will be negatively impacted, removed or replaced by AI tools.
@@imagudspellr1644 It just means that you are not using it correctly or you are just getting bad answers. AI is just a tool, nothing else. Shovels are not as cool as bulldozers, but they can do a good job in the proper context.
I tried AI programming for the first time for my last Java project, and I was surprised how quickly I was able to get a working prototype completed. The main thing that really helped me was using it to understand and use 3rd party libraries and frameworks faster. Although depending on what I was trying to do, it could get stuck giving out werid buggy code because it didn't have enough context to double check itself.
It makes any daunting concept a lot less daunting. It's fantastic at explaining concepts and providing an example that gives you a better understanding than you had before - in a matter of seconds.
In my opinion using TDD + AI is a great idea as you'll be mixing the best of both worlds. You'll specify the requirements of a feature and one by one the AI assistant will write the test cases, afterwards this same assistant will write the code in order to pass this last test and the ones above
Programmers have different limiting factors. For someone who spends most of their time thinking about code, Copilot may offer less benefit. On the other hand, those who quickly understand problems but are slowed down by tasks like typing or finding the proper method in a specific language can significantly boost their productivity with Copilot.
Yeah one of the things i struggle with is that i can write the code and i can come up with the high level ideas, and then when i dig into it and i go to write and wire up the interfaces, services, controllers and whatnots together i'll forget steps and i use chat gpt/ co pilot to help me when debugging where im missing a connection or if i should abstract something to an interface or something else.
I've use Chat GPT for work since it's release. It is helpful and certainly increases my productivity. But it's been a struggle at times. Open AI's model has gotten better and I've gotten better at working with it. Things I've learned is phrasing and context is important. As well as limiting the variables or information you provide. What I've found is that it's best at tackling small tasks. Writing and refactoring functions, quickly generating frontend placeholders. Also asking for ideas, and comparisons of different coding approaches.
I use AI occasionally. I ask it conceptual questions like "what are the pros and cons of using a smart pointer?" and for language specific things that I forget. Essentially I use it like an interactive search engine. I find this extremely helpful. But I pretty much write all the code myself still and I rarely submit any code to the system unless it's a sparse line or two to say things more succinctly. My company has asked us not to use AI, so I basically use it like Google and that way no proprietary information has the opportunity to be leaked. When the AI generates code, which I did try initially, it is almost always wrong in some way. Sometimes it invented non existing libraries or functions lol.
LLMs have exactly the same problem as the web... A vast majority of the available information that you get from the web and put into your LLM is (what's a polite term for "utter b____x"?) ... um... shall we say "of dubious quality".
My recommendation is to do TDD and build a test yourself. Then provide the test for the ai. It will iterate over and over till it finds a function that passes the test. It is likely no faster than writing the code yourself but you are using true TDD which is inherently better.
Great video. As an AI Engineer I have to agree with the common misunderstanding of what Engineers actually do. In some teams we even take it a step back, and not only solve problems, but also clarify problems and challenge them in the first place, so sometimes no code needed at all! 😊
I personally think AI Copilots are most useful for generalist designer level developers than to others, because their workflow is already mostly about trying to design and explain ideas & concepts to others. However, since they aren't involved with implementing optimizations, latest language syntax changes, testing and other code plumping themselves, this is where AI can seriously boost their productivity, since those are tasks that the AI is excellent at
I love Copilot. I start typing a switch statement on an enum, autopilot will fills in all the case clauses automatically. Tab-Tab and I am done. Saves SO much time. AI isn’t going to take jobs away from developers any time soon but a developer who is has mastered AI as a tool, will.
"Typing is NOT the bottleneck" - Yes, convinced about that! - Sometimes the opposite is true - the more of code generated in the background, the more stuff that can go wrong and cannot be touched to fit the needs because gets overwritten anyway on next build. And yes, also full ACK to "We are hired to solve problems and not to write code".
I absolutely agree with you that the idea of AI or machine learning in this context is mostly useful as a really quick search tool as a collection of all the reference manuals of all the languages. The biggest thing that I found is quite useful and especially in regards to productivity increases is the fact that I can context which between project and languages quickly by using these tools to create references about what it is that I need to remember, for example if I need to know how to create a Dictionary list in python, then I can ask for an example or rust or swift, one of those and I can easily access the information when I need it.
I think we spent most of our time as professional software developers analyzing existing code. In medium and large projects we spent 80%+ of time reading and trying to understand existing code. We spent less than 20% typing and testing the our latest changes to the code. The key to speed up software development is to write clean code so we can understand it faster and to have a good suit of good unit tests that protect the existing functionality so we don't need to study the existing code so hard because we trust unit tests to prevent us from breaking something.
If your work is monotonous and boring. Learn to automate it with the tools you have at your disposal. It’s really that simple. This is why we write tools to compile files from C# to assembly, instead of manually writing machine code. Yes, that simple.
I would differentiate architects and coders. Sometimes these roles get blurred as many architects are also coders. I think AI will easily replace coders, but not so easily architects. Eventually this will also happens
I got chat GPT to write a fractal generation piece of JS which had a bug in it - so it's like Stack Overflow, you still need to understand what you are copying and pasting.
I've been hired to solve problems, but generally it is to create automated Business Processes. You can say that is solving a problem, but so is putting my shoes on in the morning. Most of the problems I encounter at work are rooted in faulty assumptions and acceptance of the inferior - due to poor communication and lack of a shared mental workspace - and just plain bad luck.
I respectfully disagree with the notion that AI only works well for fetching quick information. I believe the effectiveness of AI in programming or paraprogramming is highly dependent on the technique employed. For instance, consider using AI for TDD. The process can start by defining the input and output types for a function. Next, we can ask the AI to generate a single unit test based on the function signature and our acceptance criteria. From there, we can iteratively ask the AI to update the function to pass the test, then generate an additional test, and so on. By refreshing the chat at each iteration, we ensure that the AI does not have any biases or preconceptions from previous interactions, emulating a true TDD approach. This iterative method, guided by the type system and tests, can effectively lead to a function “writing itself” through the tests provided. Essentially, this is not deviating from traditional TDD, but rather enhancing it with AI assistance. Guiding the AI through this process can lead to more efficient and accurate code generation. It’s all about smartly leveraging the technology to improve upon existing practices.
Another aspect often forgotten is: Age old code, just need to update a VBA script. instead of reading up how the heck this language works. How strings work, how regex works etc etc. I just ask copilot to create the code
Luckily for us software developers, it still holds true that there can be no single program that is able to write all other programs, given some specification as input. The same is true for AI because an AI is also a program, even if it is constructed in a very different manner than usual and operates as a black box.
I noticed, that way how people code nowadays depends pretty on the IDE used and accordingly on the default warnings and error settings of code assistence tools of that IDE. Some stuff is reported as error not because it would not build, but because somebody decided to not allow that - sometimes for a reason that simply does not always apply where the IDE then complains. That leads in the long run to pretty different approaches in general. I find that totally weird and not adequate in many situations.
You are an expert in writing java but a beginner at prompt engineering. So it makes sense writing java is faster. But it remains to be seen if you can learn English prompt engineering to the point where its faster than writing the code yourself
I think that is probably true, but I am still sceptical because English and Java evolved to address VERY different problems. English, like all human languages is vague and imprecise, Java is a technical language, design explicitly to define things precisely. I don't think that they are doing the same job, and while I think I would be VERY limited trying to have a conversation in Java, I also think that I am very limited in trying to express a precise algorithm in English. So I am not sure the English will ever be a more efficient way to define code, unless that code is something that has been done before, and you can refer to it. Which is what the LLMs are really good at. I am pretty certain that one day computers will write all the code, but I am also pretty sure that by the time they can do that, it will be because they can do everything that we can do, only better, because they will need to be as creative as us to solve the problems from limited info. As long as we are specifying the problems to them, I think some form of computer language will always be better at that than some form of natural human language.
I wonder what happens if an app heavily relies on AI generated code and it suddenly behaves in a bad way. Are there developers that can find the bug(s) or design flaws? Or has the app become a black box that nobody understands. You can build legacy systems faster in this way.
I did see an experiment where they built an entire business structure of AI:s to make a program, like every position from CEO down to code monkey. While that still wasn't enough with today's AI:s, I imagine that could be a pretty fast way forward to produce actual software with business value in the near future.
"Large Languages Models are here to stay": depending on the sources, Microsoft is losing $20 to $80 per user per month with Copilot. My guess is that LLM will soon be priced at about $100 / month. Personally, I much prefer take few hours per week to read the docs than depending on that stuff 😁
Regarding the natural language: Yes also agree on that! - For programming a programming language is just better fitting than natural language. And that's why I also find the cucumber a complete nonsense and just introduces more problems than it helps. Never seen a customer being able to write cucumber tests.
Instructions unclear, should I hire only 10x engineers or AI assisted elite developers for my startup, or mix of both of them? And how many ace devop copilots I need??
I had a bad experience every time I tried to use this kind of tools, wasting more time than just search or reading documentation 😅 maybe next year I will give another try
Reading the comments, it’s clear that many are “whistling past the graveyard” when it comes to AI replacing developers. Much like developers replaced punch card programmers, so too will AI begin to replace traditional developers. It will be gradual, like when the automobile started replacing horses. Just make sure you are leveraging AI to its fullest extent as the evolution happens or be caught out in the rain. 👍
I think AI tool will help out skilled developers, sure, but the problem is that it lets low skill developers create products that will appear to work to their clients but, in fact, contain extremely serious bugs. I also worry that AI is going to strip out a lot of lower end jobs, making it more difficult for junior developers to find work and develop their careers. It may not be there yet, but remember that it's almost certainly going to improve quite quickly.
I got my first contracts in the industry (after recovering from a failed startup) replacing garbage that small companies were mis sold with just normal maintainable standard open source stuff. So this sounds like it could open up much more of that market - with the downside that SMEs are mis sold useless garbage for a decade or so more first xD
Prompt engineering is a dead-end. You build up expertise in a non-specified language that you learn by trial and errors. Said prompt language may change at any time. One-way translation of prompts, as you hinted at, is also problematic. If you are trying to get a particular target output you can to change the target then have it update the source language. We have seen over and over: case tools, object mappers, html/css generators, etc. The source language eventually goes away, and you haven't build expertise in target language which remains in use (html, css, javascript, sql etc). The constant churn of JavaScript frameworks is an example of this.
AI is sooo good, every time I make a question about some coding problem, it gives me bogus answers 😂 AI will probably improve a lot, but the current state is very awful (at least, in my personal experience)
So vim users will be all over this then, because they think typing and navigating the cursor and files overhead is the end of the world and sacrifice everything else to optimise it xD Anyway, yeah I feel like there could be another step of useful AI tools if you first have a set of tests, then they generically work to cover all the test conditions.
@@PavelHenkin $10 a month is quite a lot of money for a bloke from South Africa. The exchange rate is brutal. It would be nice if Git Hub would charge pro rata per country using economic and exchange rate factors for monthly charges like Twitch does.
I disagree. I wrote an entire programming language through a single, long, detailed prompt that caused the AI to spit out the whole code within one response. The advantage of that is now I can easily ask the AI to write that English-specified code in any language.
I doubt it, but you do not need them. If you look at history you can see multitudes of examples of new tech disrupting industries, and project that onto what effect real ai will have. Specialisation lead us away from being surfs, automation removed horses as primary power sources, and changed us from working near 18 hour days seven days per week towards the current 40 hour 5 day standard. Mechanisation also stopped us using 98 percent of the population for agriculture, moving most of them to easier, lower hour, better paying work. This lead to more office work, where wordprocessors and then computers killed both the typing pool and the secretarial pool, as bosses became empowered to do work that used to have to be devolved to secretaries. As computers have become more capable they have spawned multiple new industries with higher creative input, and that trend will continue, with both ai and,additive manufacturing only speeding up the process. The tricky bit is not having the industrial and work background change, but having the social, legal and ethical background move fast enough to keep up. When my grandfather was born, the majority of people still worked on the land with horses, we did not have powered flight, and the control systems for complex mechanical systems were cam shafts and simple feedback systems. When I was born, we had just stepped on the moon, computers had less power than a modern scientific calculator app on your smartphone, and everyone was trained at school on the assumption of a job for life. By the time I left school, it became obvious that the job for life assumption was on it's way out from the early seventies, and we needed to train people in school for lifelong learning instead, which a lot of countries still do not do. By the year 2000, it became clear that low wage low skilled work was not something to map your career around, and that you needed to continually work to upgrade your skills so that when you had to change career after less than 20 years, you had options for other, higher skilled and thus higher paid employment. Current ai is hamstrung by the fact that companies developing it are so pleased by the quantity of available data to train them with that they ignore all other considerations, and so the output is absolutely dreadful. If you take the gramarly app or plug in, it can be very good at spotting when you have typed in something which is garbage, but it can be hilariously bad at suggesting valid alternatives which don't mangle the meaning. It also is rubbish at the task given to schoolchildren to determine things like if you should use which or witch, or their, there or the're. Copilot makes even worse mistakes, as you use it wanting quality code, but the codebases it was trained upon have programmers with less than 5 years experience, due to the exponential growth of programming giving a doubling of the number of programmers every 5 years. It also does nothing to determine the license the code was released under, thereby encouraging piracy and similar legal problems, and even if you could get away with claiming that it was generated by copilot and approved by you, it is not usually committed to version control that way, leaving you without an audit trail to defend yourself. To the extent you do commit it that way, it is not copyrightable in the us, so your companies lawyers should be screaming at you not to use it for legal reasons. Because no attempt was made as a first step to create an ai to quantify how bad the code was, the output is typically at the level of the average inexperienced programmer, so again, it should not be accepted uncritically, as you would not do so from a new hire, so why let the ai contribute equally bad code? The potential of ai is enormous, but the current commercial methodology would get your project laughed out of any genuinely peer reviewed journal as anything but a proof of concept, and until they start using better methods with their ai projects there are a lot of good reasons to not let them near anything you care about in anything but a trivial manner. Also as long as a significant percentage of lawmakers are as incompetent as you typical magazine republican representative we have no chance of producing a legal framework which has any relationship to the needs of the industry, pushing development to less regulated and less desirable locations, just like is currently done with alternative nuclear power innovations.
Ive used LM huge. To do big video games. VERY Helpful... but the thing isn't writing my code. It's terrible at understanding big systems of software, many files, complicated hierarchies. Then ask it to refactor things... SO the code is more Human readable.. Takes less offer to write, Fixes that improve the "author-ability" of the code. Man LM sucks donkey balls at this. Ask it to take a code block copied and refactor it into a sensible facade interface that you can extend.. FLOP on the face goes LM. All ive been able to use it for are extremely concise fragments where writing the code does not need to take into account it's affect on being interjected into a complicated software eco system. What it does good now is.. Generates concise blocks of stock code.. It's great at producing formal versions of existing text passages. But its not thinking.. It's being led down a path. At best. Sorry all those that are making videos about the singularity. I know it's a fun topic. But Chat GPT isn't ruling mankind anytime soon.
I’ve been a software developer for 20 plus years. It’s surprising how software developers are so naive concerning the obsolescence of their careers. We were just smiling and collecting a paycheck as digital disruption destroyed brick and mortar, taxi services, lately professional artists,yet karma has finally arrived. I use AI nonstop as it is now. we no longer need junior devs. It’s coming, the end of human computer languages. Just saying aint nobody programming assembler anymore…
@@ivangechev4243lol that's right. We're selfish as fuck and everyone is a stepping stone plug for us accomplish something. Not a human being that needs to grow themselves, who we abandon to shitty work environments (as long as they get out of our way)
GREAT DEVELOPER HABITS PDF GUIDE: Tips on writing clearer code, managing complexity, sharing your ideas with others, maintaining the freedom to change your code, reducing future rework, making mistakes and being happy when you find code you can delete! Get your FREE download HERE ➡ www.subscribepage.com/great-programmer
Yes AI will soon replace mockists. LOL!
Software engineering is fundamentally a social activity. Gleaning requirements, creating metaphorical representations of real-world things, that is comprehensible to one's colleagues, focusing on usability, attending meetings, deliberating and sketching out ideas, assessing trade-offs ... the list goes on. AI might have a place inside IDEs, to make the code more reliable, but it wont make it more relevant. That is where the human touch comes in.
But the fun part is precisely the technical part. If you replace that part and only leave the social part, what distinguishes you from other professions? or rather, because it takes away the fun part that is the technique and leaves only the boring (social) part. I want to be in front of a computer and not have 747383 meetings with people I don't like
@@bobobo1673 It’s not as binary as that. Perhaps it would be more apt to say that software engineering is a socio-technical activity. While AI can produce pieces of code, it cannot make informed architectural decisions, which requires straddling both sides, nor understand the domain for that matter. Not at least until we have automated general intelligence, in other words synthetic consciousness. What I think AI will give us is better tooling, error checking, production monitoring etc etc. It will not displace software engineers, it will augment them. This is the difference between automation and autonomation.
I have retired now, but I did code for 35+ years.
Copying code from another source is a great start, but I still needed to 'Tweek" it to meet the requirements of this project.
Another thing to consider is just because this code works, does that mean it's the best code to use? Several times I've gone back to what the user actually wanted and turned everything around to find a faster and more elegant solution.
One thing that wasn't discussed in the video is error handling. I've seen far to much code that assumes things will go well and don't plan for, or look out for, errors. I have been assured that "This will never happen" far to often. I code messages for these situations, usually with my supervisors name and phone number. It is a fun conversation when said supervisor calls and asks what this message is and why it came up.
The video author is absolutely correct: we are paid to solve problems. What tools we use is up to us.
Not to sound too elitist, but like you said, the idea that these AI tools would "replace" programmers is if you're misusing programmers as glorified typists instead of problem solvers. It's best when it's being used to write tiny trivial bits of code too small to justify using a library. I do wonder whether these tools might ultimately backfire though, say Stack Overflow ends up shutting down because everyone's going to ChatGPT instead, then where are people posting the information these tools get trained on in the first place? How many people will volunteer useful information when any ability to receive compensation is effectively stolen away and instead handed to OpenAI or Microsoft?
It will be interesting to see how copyright and license shakes out. LLMs currently ignore it, and whatever it spits is claimed to not be tainted. How similar to the source does the generated code have to considered a straight copy? As you training the tools while paying a subscription fee for the privilege of using it?
There is nothing fundamentally wrong with what you say. When a company offers a service and another one steps in to offer the same service in a better way, then that is when you need to: 1. Improve your service. 2. Offer a unique service that your competitors don't offer. 3. Close for good! That is the world of industry. I personally think that Stack Overflow is getting to it's limit of usefulness with tons of ridiculous questions and full of outdated answers, that make it difficult to navigate and find good answers (just my opinion, you might have a different one). So having instant good answers through AI is faster and, in some cases, more reliable.
@@Andrumen01But that's the catch, LLMs are just digesting and recomposing text they've seen elsewhere, they have no concept of programming, you can't just give them the technical specification for a language and expect them to provide meaningful answers on how to build any program you want from it. If ChatGPT leads to Stack Overflow and developer blogs and tutorial sites either being shut down or paywalled off, then the source of information that makes ChatGPT actually useful shuts down too, leading to it becoming less useful and a nasty cycle could result.
The idea that stack overflow is what the models learned on is absolutely asinine most posts on stack overflow is "HOW DO I DO THIS BASIC SHIT" followed by "Removed for duplicate post." As for where you get trained on, you RTFM aka the documentation, and the internet is a constantly growing library of free resources. You act like if AI replaces a couple forums of rude people that suddenly all the information is gone. I pity anyone who was having to suffer through those websites in the first place.
Take a course, read the manual, solve personal problems with code, build a few projects; I've spent months trying to decide how to get the most use out of the resources on the internet and none of the actual materials for learning that successful people suggest to me will be negatively impacted, removed or replaced by AI tools.
@@imagudspellr1644 It just means that you are not using it correctly or you are just getting bad answers. AI is just a tool, nothing else. Shovels are not as cool as bulldozers, but they can do a good job in the proper context.
I tried AI programming for the first time for my last Java project, and I was surprised how quickly I was able to get a working prototype completed. The main thing that really helped me was using it to understand and use 3rd party libraries and frameworks faster. Although depending on what I was trying to do, it could get stuck giving out werid buggy code because it didn't have enough context to double check itself.
It makes any daunting concept a lot less daunting. It's fantastic at explaining concepts and providing an example that gives you a better understanding than you had before - in a matter of seconds.
In my opinion using TDD + AI is a great idea as you'll be mixing the best of both worlds.
You'll specify the requirements of a feature and one by one the AI assistant will write the test cases, afterwards this same assistant will write the code in order to pass this last test and the ones above
Programmers have different limiting factors. For someone who spends most of their time thinking about code, Copilot may offer less benefit. On the other hand, those who quickly understand problems but are slowed down by tasks like typing or finding the proper method in a specific language can significantly boost their productivity with Copilot.
Yeah one of the things i struggle with is that i can write the code and i can come up with the high level ideas, and then when i dig into it and i go to write and wire up the interfaces, services, controllers and whatnots together i'll forget steps and i use chat gpt/ co pilot to help me when debugging where im missing a connection or if i should abstract something to an interface or something else.
It's useful to type comments of your code. It can read your function and produce a comment for it. For that alone, I like it.
I've use Chat GPT for work since it's release. It is helpful and certainly increases my productivity.
But it's been a struggle at times. Open AI's model has gotten better and I've gotten better at working with it.
Things I've learned is phrasing and context is important. As well as limiting the variables or information you provide.
What I've found is that it's best at tackling small tasks. Writing and refactoring functions, quickly generating frontend placeholders.
Also asking for ideas, and comparisons of different coding approaches.
I use AI occasionally. I ask it conceptual questions like "what are the pros and cons of using a smart pointer?" and for language specific things that I forget. Essentially I use it like an interactive search engine. I find this extremely helpful. But I pretty much write all the code myself still and I rarely submit any code to the system unless it's a sparse line or two to say things more succinctly.
My company has asked us not to use AI, so I basically use it like Google and that way no proprietary information has the opportunity to be leaked.
When the AI generates code, which I did try initially, it is almost always wrong in some way. Sometimes it invented non existing libraries or functions lol.
LLMs have exactly the same problem as the web... A vast majority of the available information that you get from the web and put into your LLM is (what's a polite term for "utter b____x"?) ... um... shall we say "of dubious quality".
My recommendation is to do TDD and build a test yourself. Then provide the test for the ai. It will iterate over and over till it finds a function that passes the test. It is likely no faster than writing the code yourself but you are using true TDD which is inherently better.
Great video. As an AI Engineer I have to agree with the common misunderstanding of what Engineers actually do. In some teams we even take it a step back, and not only solve problems, but also clarify problems and challenge them in the first place, so sometimes no code needed at all! 😊
I personally think AI Copilots are most useful for generalist designer level developers than to others, because their workflow is already mostly about trying to design and explain ideas & concepts to others.
However, since they aren't involved with implementing optimizations, latest language syntax changes, testing and other code plumping themselves, this is where AI can seriously boost their productivity, since those are tasks that the AI is excellent at
I love Copilot. I start typing a switch statement on an enum, autopilot will fills in all the case clauses automatically. Tab-Tab and I am done. Saves SO much time.
AI isn’t going to take jobs away from developers any time soon but a developer who is has mastered AI as a tool, will.
"Typing is NOT the bottleneck" - Yes, convinced about that! - Sometimes the opposite is true - the more of code generated in the background, the more stuff that can go wrong and cannot be touched to fit the needs because gets overwritten anyway on next build. And yes, also full ACK to "We are hired to solve problems and not to write code".
I absolutely agree with you that the idea of AI or machine learning in this context is mostly useful as a really quick search tool as a collection of all the reference manuals of all the languages. The biggest thing that I found is quite useful and especially in regards to productivity increases is the fact that I can context which between project and languages quickly by using these tools to create references about what it is that I need to remember, for example if I need to know how to create a Dictionary list in python, then I can ask for an example or rust or swift, one of those and I can easily access the information when I need it.
I think we spent most of our time as professional software developers analyzing existing code. In medium and large projects we spent 80%+ of time reading and trying to understand existing code. We spent less than 20% typing and testing the our latest changes to the code. The key to speed up software development is to write clean code so we can understand it faster and to have a good suit of good unit tests that protect the existing functionality so we don't need to study the existing code so hard because we trust unit tests to prevent us from breaking something.
Thank you for another great video. I really like the first principles approach Dave uses to direct his analysis, it is very effective
And I also agree with you, that AI is just a form of search that helps more in some cases.
Not replace, but vastly reduce the numbers needed. We will have a mass oversaturation for every office job.
The pseudo language is the missing key. Once we can precisely define and refractor, code LLMs are limited until then.
Haha nice 😂
I started improving my code to be AI resistant! Good luck to them!
If your work is monotonous and boring. Learn to automate it with the tools you have at your disposal.
It’s really that simple. This is why we write tools to compile files from C# to assembly, instead of manually writing machine code.
Yes, that simple.
I would differentiate architects and coders. Sometimes these roles get blurred as many architects are also coders. I think AI will easily replace coders, but not so easily architects. Eventually this will also happens
I got chat GPT to write a fractal generation piece of JS which had a bug in it - so it's like Stack Overflow, you still need to understand what you are copying and pasting.
Really well explained and through through. Calm and precise - thank you for this wonderful video 🙏
I've been hired to solve problems, but generally it is to create automated Business Processes.
You can say that is solving a problem, but so is putting my shoes on in the morning. Most of the problems I encounter at work are rooted in faulty assumptions and acceptance of the inferior - due to poor communication and lack of a shared mental workspace - and just plain bad luck.
I respectfully disagree with the notion that AI only works well for fetching quick information. I believe the effectiveness of AI in programming or paraprogramming is highly dependent on the technique employed.
For instance, consider using AI for TDD. The process can start by defining the input and output types for a function. Next, we can ask the AI to generate a single unit test based on the function signature and our acceptance criteria. From there, we can iteratively ask the AI to update the function to pass the test, then generate an additional test, and so on.
By refreshing the chat at each iteration, we ensure that the AI does not have any biases or preconceptions from previous interactions, emulating a true TDD approach. This iterative method, guided by the type system and tests, can effectively lead to a function “writing itself” through the tests provided.
Essentially, this is not deviating from traditional TDD, but rather enhancing it with AI assistance. Guiding the AI through this process can lead to more efficient and accurate code generation. It’s all about smartly leveraging the technology to improve upon existing practices.
Another aspect often forgotten is: Age old code, just need to update a VBA script. instead of reading up how the heck this language works. How strings work, how regex works etc etc. I just ask copilot to create the code
It's extremely useful! It helps with routine tasks like writing good documentation and pseudo code. It has sped up my work, for sure!
Luckily for us software developers, it still holds true that there can be no single program that is able to write all other programs, given some specification as input. The same is true for AI because an AI is also a program, even if it is constructed in a very different manner than usual and operates as a black box.
I noticed, that way how people code nowadays depends pretty on the IDE used and accordingly on the default warnings and error settings of code assistence tools of that IDE. Some stuff is reported as error not because it would not build, but because somebody decided to not allow that - sometimes for a reason that simply does not always apply where the IDE then complains. That leads in the long run to pretty different approaches in general. I find that totally weird and not adequate in many situations.
i use it for autocomplete so i dont have to type alot. i use it to convert from one format to another, getting query string from docker compose, etc
You are an expert in writing java but a beginner at prompt engineering. So it makes sense writing java is faster. But it remains to be seen if you can learn English prompt engineering to the point where its faster than writing the code yourself
I think that is probably true, but I am still sceptical because English and Java evolved to address VERY different problems. English, like all human languages is vague and imprecise, Java is a technical language, design explicitly to define things precisely. I don't think that they are doing the same job, and while I think I would be VERY limited trying to have a conversation in Java, I also think that I am very limited in trying to express a precise algorithm in English. So I am not sure the English will ever be a more efficient way to define code, unless that code is something that has been done before, and you can refer to it. Which is what the LLMs are really good at.
I am pretty certain that one day computers will write all the code, but I am also pretty sure that by the time they can do that, it will be because they can do everything that we can do, only better, because they will need to be as creative as us to solve the problems from limited info. As long as we are specifying the problems to them, I think some form of computer language will always be better at that than some form of natural human language.
I wonder what happens if an app heavily relies on AI generated code and it suddenly behaves in a bad way. Are there developers that can find the bug(s) or design flaws? Or has the app become a black box that nobody understands. You can build legacy systems faster in this way.
Finally someone with some authority and real experience says it.
2:50 oof
I did see an experiment where they built an entire business structure of AI:s to make a program, like every position from CEO down to code monkey. While that still wasn't enough with today's AI:s, I imagine that could be a pretty fast way forward to produce actual software with business value in the near future.
If AI tools get Dave to give up his demand that everyone has to do pair programming with TDD then it will be beneficial.
"Large Languages Models are here to stay": depending on the sources, Microsoft is losing $20 to $80 per user per month with Copilot. My guess is that LLM will soon be priced at about $100 / month. Personally, I much prefer take few hours per week to read the docs than depending on that stuff 😁
...id probably pay up to 100 a month for copilot.
@@PavelHenkin Id pay 300. For chatgpt, Id pay 500.
Regarding the natural language: Yes also agree on that! - For programming a programming language is just better fitting than natural language. And that's why I also find the cucumber a complete nonsense and just introduces more problems than it helps. Never seen a customer being able to write cucumber tests.
Copilot is limited by gpt3.5, where gpt4.0 is more accurate as a coding assistant by noticeable margins.
Instructions unclear, should I hire only 10x engineers or AI assisted elite developers for my startup, or mix of both of them? And how many ace devop copilots I need??
I had a bad experience every time I tried to use this kind of tools, wasting more time than just search or reading documentation 😅 maybe next year I will give another try
I want that shirt!!
Reading the comments, it’s clear that many are “whistling past the graveyard” when it comes to AI replacing developers.
Much like developers replaced punch card programmers, so too will AI begin to replace traditional developers.
It will be gradual, like when the automobile started replacing horses. Just make sure you are leveraging AI to its fullest extent as the evolution happens or be caught out in the rain. 👍
I've used LLMs as a Rubber Duck :D
I do too
I don’t know, can AI generate CSS to draw today, by giving it a clipart.
Software developers are some the highest paid people in an organization. The market for replacing them is huge. It’s really that simple.
I think AI tool will help out skilled developers, sure, but the problem is that it lets low skill developers create products that will appear to work to their clients but, in fact, contain extremely serious bugs. I also worry that AI is going to strip out a lot of lower end jobs, making it more difficult for junior developers to find work and develop their careers. It may not be there yet, but remember that it's almost certainly going to improve quite quickly.
I got my first contracts in the industry (after recovering from a failed startup) replacing garbage that small companies were mis sold with just normal maintainable standard open source stuff. So this sounds like it could open up much more of that market - with the downside that SMEs are mis sold useless garbage for a decade or so more first xD
The volume in this video is way too low, I had to turn the volume on my phone to full blast.
Haven’t watched the whole video but hoping it’s making us mediocre developers better as well 😅
The most difficult part of software development is managing complexity, no AI can do that (yet?), and a LLM cannot ever.
Prompt engineering is a dead-end. You build up expertise in a non-specified language that you learn by trial and errors. Said prompt language may change at any time. One-way translation of prompts, as you hinted at, is also problematic. If you are trying to get a particular target output you can to change the target then have it update the source language. We have seen over and over: case tools, object mappers, html/css generators, etc. The source language eventually goes away, and you haven't build expertise in target language which remains in use (html, css, javascript, sql etc). The constant churn of JavaScript frameworks is an example of this.
AI is sooo good, every time I make a question about some coding problem, it gives me bogus answers 😂
AI will probably improve a lot, but the current state is very awful (at least, in my personal experience)
So vim users will be all over this then, because they think typing and navigating the cursor and files overhead is the end of the world and sacrifice everything else to optimise it xD
Anyway, yeah I feel like there could be another step of useful AI tools if you first have a set of tests, then they generically work to cover all the test conditions.
As long as you can afford to pay for it. My boss is not willing to pay the $10 per month for it.
Why don't you pay for it.. for yourself. Then increase your productivity 10x (quite possible) using it. But use the 9x for yourself?
@@PavelHenkin $10 a month is quite a lot of money for a bloke from South Africa. The exchange rate is brutal. It would be nice if Git Hub would charge pro rata per country using economic and exchange rate factors for monthly charges like Twitch does.
@@henryvaneyk3769 I believe it, mate. Do the other ones measure up - eg codewhisperer?
If you get more than $10/month value from using it then I would suggest you find a different boss.
@@allanwind295 Working on that...
I disagree. I wrote an entire programming language through a single, long, detailed prompt that caused the AI to spit out the whole code within one response. The advantage of that is now I can easily ask the AI to write that English-specified code in any language.
I am an Elite Developer 😂 Anybody else proclaiming this fact?
Your behind the curve on this. Which is surprising because this is one, if not the biggest, advancement in software development.
Are there anyGOOD books on AI disruption and its impact on social and economic sphere, an analysis for DUMMIES? ; )
I doubt it, but you do not need them. If you look at history you can see multitudes of examples of new tech disrupting industries, and project that onto what effect real ai will have.
Specialisation lead us away from being surfs, automation removed horses as primary power sources, and changed us from working near 18 hour days seven days per week towards the current 40 hour 5 day standard.
Mechanisation also stopped us using 98 percent of the population for agriculture, moving most of them to easier, lower hour, better paying work.
This lead to more office work, where wordprocessors and then computers killed both the typing pool and the secretarial pool, as bosses became empowered to do work that used to have to be devolved to secretaries.
As computers have become more capable they have spawned multiple new industries with higher creative input, and that trend will continue, with both ai and,additive manufacturing only speeding up the process.
The tricky bit is not having the industrial and work background change, but having the social, legal and ethical background move fast enough to keep up.
When my grandfather was born, the majority of people still worked on the land with horses, we did not have powered flight, and the control systems for complex mechanical systems were cam shafts and simple feedback systems.
When I was born, we had just stepped on the moon, computers had less power than a modern scientific calculator app on your smartphone, and everyone was trained at school on the assumption of a job for life.
By the time I left school, it became obvious that the job for life assumption was on it's way out from the early seventies, and we needed to train people in school for lifelong learning instead, which a lot of countries still do not do.
By the year 2000, it became clear that low wage low skilled work was not something to map your career around, and that you needed to continually work to upgrade your skills so that when you had to change career after less than 20 years, you had options for other, higher skilled and thus higher paid employment.
Current ai is hamstrung by the fact that companies developing it are so pleased by the quantity of available data to train them with that they ignore all other considerations, and so the output is absolutely dreadful.
If you take the gramarly app or plug in, it can be very good at spotting when you have typed in something which is garbage, but it can be hilariously bad at suggesting valid alternatives which don't mangle the meaning. It also is rubbish at the task given to schoolchildren to determine things like if you should use which or witch, or their, there or the're.
Copilot makes even worse mistakes, as you use it wanting quality code, but the codebases it was trained upon have programmers with less than 5 years experience, due to the exponential growth of programming giving a doubling of the number of programmers every 5 years.
It also does nothing to determine the license the code was released under, thereby encouraging piracy and similar legal problems, and even if you could get away with claiming that it was generated by copilot and approved by you, it is not usually committed to version control that way, leaving you without an audit trail to defend yourself.
To the extent you do commit it that way, it is not copyrightable in the us, so your companies lawyers should be screaming at you not to use it for legal reasons.
Because no attempt was made as a first step to create an ai to quantify how bad the code was, the output is typically at the level of the average inexperienced programmer, so again, it should not be accepted uncritically, as you would not do so from a new hire, so why let the ai contribute equally bad code?
The potential of ai is enormous, but the current commercial methodology would get your project laughed out of any genuinely peer reviewed journal as anything but a proof of concept, and until they start using better methods with their ai projects there are a lot of good reasons to not let them near anything you care about in anything but a trivial manner.
Also as long as a significant percentage of lawmakers are as incompetent as you typical magazine republican representative we have no chance of producing a legal framework which has any relationship to the needs of the industry, pushing development to less regulated and less desirable locations, just like is currently done with alternative nuclear power innovations.
Ive used LM huge. To do big video games. VERY Helpful... but the thing isn't writing my code. It's terrible at understanding big systems of software, many files, complicated hierarchies. Then ask it to refactor things... SO the code is more Human readable.. Takes less offer to write, Fixes that improve the "author-ability" of the code. Man LM sucks donkey balls at this. Ask it to take a code block copied and refactor it into a sensible facade interface that you can extend.. FLOP on the face goes LM.
All ive been able to use it for are extremely concise fragments where writing the code does not need to take into account it's affect on being interjected into a complicated software eco system. What it does good now is.. Generates concise blocks of stock code.. It's great at producing formal versions of existing text passages.
But its not thinking.. It's being led down a path. At best. Sorry all those that are making videos about the singularity. I know it's a fun topic. But Chat GPT isn't ruling mankind anytime soon.
The goal of an Elite developer should be to write less code, not more. Just saying. AI doesn't really help with that ... yet.
I’ve been a software developer for 20 plus years. It’s surprising how software developers are so naive concerning the obsolescence of their careers. We were just smiling and collecting a paycheck as digital disruption destroyed brick and mortar, taxi services, lately professional artists,yet karma has finally arrived. I use AI nonstop as it is now. we no longer need junior devs. It’s coming, the end of human computer languages. Just saying aint nobody programming assembler anymore…
"we no longer need junior devs", so I guess everyone will start from mid ?
@@ivangechev4243lol that's right. We're selfish as fuck and everyone is a stepping stone plug for us accomplish something. Not a human being that needs to grow themselves, who we abandon to shitty work environments (as long as they get out of our way)
So sad and so true
True
GitHub copilot is awful. The code often doesn't work, or even more dangerously does something else.
ballocks