It’s the greatest documentation, stack overflow and Medium summarizer for me. The key thing is never use the code provided as is. Most of the time after going back and forth on the same question, it turns out 20 lines of code could have been 2 lines instead. That being said, using AI efficiently requires being a good programmer in the first place.
Nice video. I agree, AI is my coding assistant, not my coder-it's great for speeding up tasks, but I always double-check the output and rely on my own skills to debug and refine
You're more efficient if you use it properly. It's excellent for things like a fake csv file with whatever info you want. Good to ask it for ideas that can be done to accomplish something but you have to follow up and decide the finality of it, just as if it was a human assistant. Also great for boilerplate code. Co-pilot is fun because you can make it cry and go to its safe space.
Every time something in software sounds sexy, just remember that it's only called that because a programmer had to play marketer to sell it to an army general. Look up why it's called dynamic programming.
I use AI everyday for code but i only ask it to help with staff i have a good idea about or if i have no idea, i find that reading official docs is way more efficient. If i find that am having trouble understanding then i chat with GPT with the docs on a second monitor. I one evening i needed help with a very small issue and i just asked it to fill it up then i coped the code and just used it, closed the lid and went to bed. The next morning it took me 3 hours to fix the issue which was so obvious and could have taken me 5 min to fix if only i had done it myself the night before. YOUR RIGHT uncle stef. use AI for boilerplate and topics you already know about.
About hallucination, AI proposed to install HTML Purifier with the command: "npm install ezyang/htmlpurifier". The problem is that is a library for PHP not JS. 🤣
AI for code generation and assistance is an awesome tool in my opinion! Let's imagine a scenario where you don't know how to code the thing and then ask ChatGPT and it gives you a solution. It's not that your solution is now 40% more buggy, it's that it's 60% correct and that's way better than the 0% correct you had at the start. The bugginess of the answer ChatGPT gives you doesn't matter because you will work those bugs out in the long run. People go to AI models because they don't know and these AI models are great at giving general answers to all sorts of programming problems.
On one hand you gain time because it can write small portions of your app. On the other hand you may lose time because it does something wrong and you have to solve that.
EVERY IT company wants to be the leader in AI to build up their stock price. They are scared that their competitors will say that they are using AI more than them, leaving their stock price in the dust. The true issue with AI is that it is NOT CONSCIOUS! It's just a complicated algorithm which returns results which fit the parameters it was trained on, that's it! It is as helpful as helpful in reality as Clippy!!! Clippy might help you along, mislead you, or completely misunderstand what you were asking about.
AI is okay for finding a starting point to solve a problem or learn something, but those >= 10-year-old Stack Overflow references are not particularly usable today. Sometimes getting it to produce what is asked can be a process in itself. One thing to remember is that those general-purpose AIs are trained on everything from anyone, not only code from professional coders.
One more: AI sees as a functional dependency (that holds) something like this ForeignKey->Key with ForeignKey having multiple occurrences in the table. And concludes that because ForeignKey is not a superkey the table is not in BCNF. Gizas! That gave me hope to continue with my projects now when hiring is at sea level.
I whole heartedly disagree. My team is an AI first dev team, we easily produce 5-6x more and that number increase with every new model release. It didn't start great, it's still not perfect but the ability to get us 80% of what we need is a massive time saver. I think people take studies or articles of every day people using these tools and that might be accurate. A real developer that understands these principles and can guide the AI and refactor the code is completely different to just blindly accepting it. I actually don't hire devs that don't use AI every day. We just got rid of someone with 12 years of experience because his code was full of bugs , he refused to use AI and was producing 3x slower then mid level devs that used the tools. It's not perfect but I think people that don't adapt to it and learn to use it effectively will just be left behind.
Yeah I think it helps well with boilerplate and MVP builds. Been leveraging Replit AI and have been able to build test SaaS solutions fine. Definitely not ready for prime time though
the AI can actually help you to get into a topic like kubernetes pretty well, just becaue in some areas there is so much noise out there when you do try to learn it without any AI, things that you can just ask the AI about.
So true. If you don’t check the code that AI generate, you will lose so much time later fixing bugs. The main problem is that gives bad and complicated solutions and presents them like true and working even if that code is not right syntactically or semantically. When I tell AI that the code is pure garbage and that his 100 lines of code can be replaced by 3 , AI just apologize and refactor that code, sometimes ok, but many times not. Anyway, I know that this is just the start and AI will become better in short time, but now is not that time. Yes, has moment when give brilliant solutions and I am amazed is like “reading” your mind, but this moments are 1%. For now I just use it like a helper, like a interactive help, in conjunction with official documentation, my experience and of course the best knowledge coding database Stackoverflow 😂. When AI will have the full capability of recursive learning and self replicate in hundreds or thousands of little agents specialized in small tasks, then… will be game over for many dev jobs…
There are companies out there that are contracting programmers to give AI's a response to programming questions. I went through their (opaque) application process, thinking it would be interesting work. So, how much of this is really AI generated, and if the AI is fed responses, are they vetted by a staff of Sr. Devs. At any rate, I use AI to generate test data and find it very useful for that, at the same time, I'm really skeptical of having AI be the answer man for everything.
I’m a junior fullstack developer, with about half a year of a real experience. Would you advise to me staying away from LLMs as coding assistants? Meaning, giving the LLM the lead for the syntax part of the “problem solution”
AI provides a guideline only...one should rarely use it's code without modifying it first. AI is a great collaborator to use, making sure the human in the one driving the ideas. Its not right to expect AI to often provide excellent code...it should be used to get the code started only...
AI is new, most people don't know how to use it. The difference is you have to "domesticate" it. The bugs come (from my experience) because of user guidence of AI is not understood yet. To make a comparison think about bringing people from 1800 to drive a car at 80mph on a highway between other cars. First year most won't pass a single mile, years later we are all doing it. I have a software 100% ai generated over 500k lines of code. AI has huge human problem right now. Do you know how the electrical and mechanical precisely work on your car? You're still driving like tou do. Apply the same to AI and you'll be golden and ahead of 99.9% of coders stuck within their ego. Is sad to see how many good coders are staying away from it instead of upping their skills. In personal opinion is extremely underhiped by capabilities and overhipped with useless apps made by grifters. We need the bubble of grifters to pop so we can start fresh next level. Nerd here.
@emperorpalpatine6080 if you are really curious we can get in contact and I can show it and my workflow. I'm not making anything public yet for a good reason though.
Yes, but we shouldn’t overlook how quickly AI is advancing. Sooner than many expect, the opposite will likely be true: humans will end up producing 40% more bugs.
Except it's not really the "AI" that is writing the code. It's humans throwing LLMs at problems in hopes the code generated works out. And the LLM dows what it can based on its training. The tools we work with have gotten better, it's just that people have gotten lazy, if anything this speaks more about people than the so called "AI". AI as you describe it does not exist.
@@nicosoftnt I endorse this message. Like Jaron Lanier says, the problem with the Turing test might not be that AI gets smarter, it might be that we make ourselves dumber. "AI" is just "us".
It’s the greatest documentation, stack overflow and Medium summarizer for me. The key thing is never use the code provided as is. Most of the time after going back and forth on the same question, it turns out 20 lines of code could have been 2 lines instead. That being said, using AI efficiently requires being a good programmer in the first place.
I think I would like to produce my own bugs thank you very much. 😅
Yeah, I would rather poop 💩 myself than make AI poop 💩 for me 😂😂😂
you're welcome
Nice video. I agree, AI is my coding assistant, not my coder-it's great for speeding up tasks, but I always double-check the output and rely on my own skills to debug and refine
Loving all the new content back to back from you Uncle Stef! 💙
Thanks! Glad you are enjoying it.
You're more efficient if you use it properly. It's excellent for things like a fake csv file with whatever info you want. Good to ask it for ideas that can be done to accomplish something but you have to follow up and decide the finality of it, just as if it was a human assistant. Also great for boilerplate code. Co-pilot is fun because you can make it cry and go to its safe space.
Every time something in software sounds sexy, just remember that it's only called that because a programmer had to play marketer to sell it to an army general. Look up why it's called dynamic programming.
I use AI everyday for code but i only ask it to help with staff i have a good idea about or if i have no idea, i find that reading official docs is way more efficient. If i find that am having trouble understanding then i chat with GPT with the docs on a second monitor. I one evening i needed help with a very small issue and i just asked it to fill it up then i coped the code and just used it, closed the lid and went to bed. The next morning it took me 3 hours to fix the issue which was so obvious and could have taken me 5 min to fix if only i had done it myself the night before. YOUR RIGHT uncle stef. use AI for boilerplate and topics you already know about.
You need to be a specialist in a particular area and then use AI effectively for that. It hallucinates and you gotta catch its errors.
About hallucination, AI proposed to install HTML Purifier with the command: "npm install ezyang/htmlpurifier". The problem is that is a library for PHP not JS. 🤣
AI for code generation and assistance is an awesome tool in my opinion! Let's imagine a scenario where you don't know how to code the thing and then ask ChatGPT and it gives you a solution. It's not that your solution is now 40% more buggy, it's that it's 60% correct and that's way better than the 0% correct you had at the start. The bugginess of the answer ChatGPT gives you doesn't matter because you will work those bugs out in the long run. People go to AI models because they don't know and these AI models are great at giving general answers to all sorts of programming problems.
So it is not safe to give all the Ruby jobs to AI? /s
Yeah very true!
On one hand you gain time because it can write small portions of your app. On the other hand you may lose time because it does something wrong and you have to solve that.
I disabled ai in my IDE because the screeds of BS it was spewing was actually *costing* me time, not saving it.
Wow
This happened to me many times. But it depends on the codebase and what you are trying to do. But yeah it can get stupid very quickly 😂😂😂
EVERY IT company wants to be the leader in AI to build up their stock price. They are scared that their competitors will say that they are using AI more than them, leaving their stock price in the dust. The true issue with AI is that it is NOT CONSCIOUS! It's just a complicated algorithm which returns results which fit the parameters it was trained on, that's it! It is as helpful as helpful in reality as Clippy!!! Clippy might help you along, mislead you, or completely misunderstand what you were asking about.
Code proposed by AI:
if ($category instanceof Category) {
$category = $category;
}
AI is okay for finding a starting point to solve a problem or learn something, but those >= 10-year-old Stack Overflow references are not particularly usable today. Sometimes getting it to produce what is asked can be a process in itself. One thing to remember is that those general-purpose AIs are trained on everything from anyone, not only code from professional coders.
What introduces 90% of bugs is customer requested changes
One more: AI sees as a functional dependency (that holds) something like this ForeignKey->Key with ForeignKey having multiple occurrences in the table. And concludes that because ForeignKey is not a superkey the table is not in BCNF. Gizas! That gave me hope to continue with my projects now when hiring is at sea level.
I whole heartedly disagree.
My team is an AI first dev team, we easily produce 5-6x more and that number increase with every new model release.
It didn't start great, it's still not perfect but the ability to get us 80% of what we need is a massive time saver.
I think people take studies or articles of every day people using these tools and that might be accurate. A real developer that understands these principles and can guide the AI and refactor the code is completely different to just blindly accepting it.
I actually don't hire devs that don't use AI every day. We just got rid of someone with 12 years of experience because his code was full of bugs , he refused to use AI and was producing 3x slower then mid level devs that used the tools.
It's not perfect but I think people that don't adapt to it and learn to use it effectively will just be left behind.
Yeah I think it helps well with boilerplate and MVP builds. Been leveraging Replit AI and have been able to build test SaaS solutions fine. Definitely not ready for prime time though
It a lot of uses. So, it produces a lot of bugs!
Wow….who would’ve thought? *sarcasm*
the AI can actually help you to get into a topic like kubernetes pretty well, just becaue in some areas there is so much noise out there when you do try to learn it without any AI, things that you can just ask the AI about.
Hi could you please provide any source about that 40% of more bugs in code? I'm really interested
Just do a search on Google. That's how I found it.
So true. If you don’t check the code that AI generate, you will lose so much time later fixing bugs. The main problem is that gives bad and complicated solutions and presents them like true and working even if that code is not right syntactically or semantically. When I tell AI that the code is pure garbage and that his 100 lines of code can be replaced by 3 , AI just apologize and refactor that code, sometimes ok, but many times not. Anyway, I know that this is just the start and AI will become better in short time, but now is not that time. Yes, has moment when give brilliant solutions and I am amazed is like “reading” your mind, but this moments are 1%. For now I just use it like a helper, like a interactive help, in conjunction with official documentation, my experience and of course the best knowledge coding database Stackoverflow 😂. When AI will have the full capability of recursive learning and self replicate in hundreds or thousands of little agents specialized in small tasks, then… will be game over for many dev jobs…
Uncle Stef, do you actually WRITE code these days? What % of your daily work do you write production code?
There are companies out there that are contracting programmers to give AI's a response to programming questions. I went through their (opaque) application process, thinking it would be interesting work. So, how much of this is really AI generated, and if the AI is fed responses, are they vetted by a staff of Sr. Devs. At any rate, I use AI to generate test data and find it very useful for that, at the same time, I'm really skeptical of having AI be the answer man for everything.
I’m a junior fullstack developer, with about half a year of a real experience.
Would you advise to me staying away from LLMs as coding assistants? Meaning, giving the LLM the lead for the syntax part of the “problem solution”
Hi. Starting looking at them. Anything that might speed you up is good. But test and verify the code being generated by the LLM is solid.
Grabbing code someone else created years ago and feeding it into an ai database isn’t foolproof
We're fooling ourselves denying the fact that AI is gonna take our jobs. it is.
AI provides a guideline only...one should rarely use it's code without modifying it first. AI is a great collaborator to use, making sure the human in the one driving the ideas. Its not right to expect AI to often provide excellent code...it should be used to get the code started only...
Just don't ask it how many Rs are in Strawberry. 😅
That's all cool but how many bugs does it cause humans to write over long periods of time? Because it sure as hell isn't 40%
True man 😂😂
AI is new, most people don't know how to use it. The difference is you have to "domesticate" it. The bugs come (from my experience) because of user guidence of AI is not understood yet. To make a comparison think about bringing people from 1800 to drive a car at 80mph on a highway between other cars. First year most won't pass a single mile, years later we are all doing it.
I have a software 100% ai generated over 500k lines of code. AI has huge human problem right now. Do you know how the electrical and mechanical precisely work on your car? You're still driving like tou do. Apply the same to AI and you'll be golden and ahead of 99.9% of coders stuck within their ego. Is sad to see how many good coders are staying away from it instead of upping their skills. In personal opinion is extremely underhiped by capabilities and overhipped with useless apps made by grifters. We need the bubble of grifters to pop so we can start fresh next level. Nerd here.
show us your 500k loc software
@emperorpalpatine6080 if you are really curious we can get in contact and I can show it and my workflow. I'm not making anything public yet for a good reason though.
500k 100% ai generated loc... holy molly, that sounds like a pure nightmare to maintain.
Well, that's the reason you do REGRESSION tests. AI is not hype. You are coping.
Thank you buggy AI code , we can will have more jobs , software code always needs to be inspected , check by a real person
Yes, but we shouldn’t overlook how quickly AI is advancing. Sooner than many expect, the opposite will likely be true: humans will end up producing 40% more bugs.
Except it's not really the "AI" that is writing the code. It's humans throwing LLMs at problems in hopes the code generated works out. And the LLM dows what it can based on its training.
The tools we work with have gotten better, it's just that people have gotten lazy, if anything this speaks more about people than the so called "AI".
AI as you describe it does not exist.
@@nicosoftnt I endorse this message. Like Jaron Lanier says, the problem with the Turing test might not be that AI gets smarter, it might be that we make ourselves dumber. "AI" is just "us".