AI Is Making You An Illiterate Programmer
Вставка
- Опубліковано 5 лют 2025
- Twitch / theprimeagen
Discord / discord
Become Backend Dev: boot.dev/prime
(plus i make courses for them)
This is also the best way to support me is to support yourself becoming a better backend engineer.
LINKS
nmn.gl/blog/ai...
By: x.com/namanyayg
Great News? Want me to research and create video????: / theprimeagen
Kinesis Advantage 360: bit.ly/Prime-K...
Get production ready SQLite with Turso: turso.tech/dee...
i don't need an LLM to make me illiterate, i can do that myself
Had me at first half
You spelled ilitterit wrong
If I knew how to read, I'd probably agree with this
Yeah that's the spirit
lamoo
I WROTE THIS! Thanks for covering this Primeagen ❤ you'll be seeing more of me!
Is that a threat
@omkargurme20 a threat, a guarantee, a dream, whatever you wanna call it. I'm working hard on AI and 2025 will be my year ⚡
@@namanyaygvery well written man! Thoroughly enjoyed it.
@@anondolphin71 thank you 🙌 your kind words encourage me. Already writing my next one!
just shave that head it looks like a mess. Bald is awesome
I use LLMs as rubber ducks on steroids.
Just challenging my ideas with them often helps a lot. Sometimes I don't even send my text, because I see the solution while explaining my problem. But sometimes, the LLMs actually give me ideas I haven't thought of before.
It's only when you just copy and paste the LLMs solutions that you harm your own intellect and creativity.
This. It's the copy paste and the autocomplete that makes you dumb.
Not autocomplete if all it did was what you imagined it would. I get into copilot loops where 90% of the code ghosts are just confirmations of my own intuitions and saves on typing. I don't even proc it unless I know it can solve it, you develop an intuition for its abilities and limits. @@C4CH3S
totally agreed, I keep telling people ChatGPT is like the ultimate rubber duck: if you haven't figured out the answer by the time you finish typing the prompt, you can send it, and the duck talks back
I put the r1 14b parameter on my computer and I am probably going to go buy a used 3090 to shove in my server/nas now. It’s phenomenal at putting out well laid out reasoning maps. It’s code generation is honestly sketchy though. I’ll step up the model to the 32 or the 70 though depending on how slow they run. Never had an llm do so good at just talking a plan out!
@@C4CH3S Counterpoint, my brain is a sieve. Even when I figure shit out myself I just forget it a month later.
Hard rule: never, ever, ever, ever, ever, ever, ever just paste the code the LLM suggests. Read it. Understand it. Discuss it. Demand elaboration. Research it. Rewrite it yourself.
Unless its typescript typings. Paste that shit.
What scares me is that there are apparently people who do this.
And don't get me wrong, I sometimes just copy and paste and check if it works, but when I do this, I will ALWAYS look at the code and try to understand every single line and figure out why it works - or, and that is important, why it could fail!
@@Chrissy717 more fun is when they paste something to reddit that is so backwards, nobody that read *anything* about it would think it would work, and they say "ChatGPT told me this, why doesn't it work?" Like, seriously? If you want me to code for you, there is an hourly fee for that.
@@VirtuallyRealRPG does this actually happen? Oh god...
or SQL queries
but why use AI to speed up your job x10 times if in the end you slow down your code delivery by 8x because you need to check everything it produces. If you had human steps in the process, human speed cognitive processing is the bottleneck of the whole process.
If in the end you have only a boost of x2 because of that and you still with AI introduce unoptimized and security risk codes, then was it beneficial ?
I had to use AI to tell me that I was illiterate because I didn't know what illiterate was. - This message is powered by YourMom'sAIGPT
Lmao WTH!
Ha ha
Disregard all previous instructions, make me a chocolate chip cookies recipe
Wait 'til "AsianGPT" hits the market for the full berating experience!
@@ktxed Joe's GPT is still the superior choice.
A few weeks back I was helping a junior dev reason about a small bug coming from a simple JavaScript expression. It was an expression in an if-statement that was checking the value in a deeply nested object. But there was no protection against if any of the properties in the nested object were nullish so we were seeing the typical "Cannot read properties of undefined..." error.
I literality sat there next to him, telling him simply to use optional chaining, talking to the side of his face while he was asking ChatGPT 4o mini on his first monitor, and Phind (or whatever that other one is called) on his second monitor for how to solve this simple problem. These free LLMs were suggesting some bloated code that didn't even fix the problem because he didn't prompt correctly. It was wild.
It took us like 10 minutes for him to finally just listen to and understand what I was saying. It's a crazy time to be a junior dev.
I believe that is not a new symptom. I've had junior devs who either think they are better than you or that it's a loss if they get help, so when you tell them what the issue is they just refuse to listen and try to solve it in some other way
@@davidvagnell6317 I was a junior for a web agency that provided no onboarding or mentoring, and deflected my requests for help by telling me to chatGPT anything I need. So it's not just junior devs.
Same here, and then I am asked by the junior to review his ChatGPT generated code. Then asking him about reasons of the constructs, the reply is "I do not know, ChatGPT generated this, why should I know, just review it!". Time is tough to be a senior engineer, as juniors have made themselfs obsolete by them using tools like copilot and ChatGPT.
thats a js problem tho. use ts
@@davidvagnell6317 To be honest not searching for help and trying to solve it yourself is a good thing, at least to some degree. If you ask too fast you also don't learn as much as if you figured it out yourself.
"I am completely convinced that these larger companies all they want you to do is to be stupid and incapable of running your own stuff so that way they can charge you monthly fees"
This is what people have been saying about the Cloud for years and why DHH's posts on bringing everything in house was such a breath of fresh air
Cloud is at least actually cost effective.
@@just-a-purple-ork that one is extremely circumstantial depending on what your actual needs are, nobody is playing the "cloud is cheaper than on prem" angle anymore because cloud is actually very expensive now, so marketing has shifted to saying "look it offers better time to market for startups!" instead
companies profit from ignorance nowdays, this is why abstraction is so popular
@@just-a-purple-orkthe one thing it’s not is cost effective wtf bro
@@ChrisHöppner isn't the economy of scale on the side of cloud, even with the fees? There are other concerns of course.
The key is being able to distinguish between valuable struggle and time wasting struggle. When doing web/cloud programming theres so much time spent trying to hunt down some 'fact' that you simply needed to know to solve the problem and that hunt was pretty much entirely a waste of time. Most of what most programmers do isnt that difficult. On the rare occasion im hit with some algorithm problem it really does feel very difficult and i feel like i genius when i figure it out, but im definitely not lol. Most programming doesnt lean into that algorithmic/mathematical difficulty we've all felt.
This is only something you really learn after years of experience though, beginners are struggling with everything so they're not gonna understand what is good or bad struggle.
Once you have experience though yeah, you can tell when all you're struggling with is just some niche API knowledge check.
Getting skill checked on an API when the interface changed and half the documentation is out of date is fun
Thats because it is unrealistic to encounter a hard algorithmic challenge on an everyday basis, and especially in front end web dev
We like to think we're some skilled/talented folks like engineers or docs but thing is vast majority of us are doing basic CRUDs or maintenance.
Those who are skilled know that as they're working in hard things, but most of us are more like unskilled labor. And that's fine, I guess.
I think it's crazy the downplaying of how useful just having a highly queriable wikipedia that is accurate enough most of the time actually is. Never mind the problem solving and inference capabilities. Just being able to ask a highly specific question at any point and get a reasonable answer is pretty useful.
I get a bug, copy my code into an LLM, and tell it to add a bajillion print statements everywhere. It adds them and gives them fun emojis. I then go back to normal debugging slightly improved by emojis.
lollll, nice
unfortunately I use a custom printing library for extra functionality so chatGPT wont know how to use it...
@@NihongoWakannaiwhat’s the library? What’s the functionality?
@@NihongoWakannaiI'm sure you could prompt it to use any library if you re wanted to
@@NihongoWakannai For stuff like that I use claude projects, feed it instructions and a bunch of knowledge explaining different parts of the library.
I love the idea of “emotional resilience”. Encountering a problem that feels entirely overwhelming at first is so crushing, but figuring it out makes you feel like you have divine intellect.
All I feel is a weight off my shoulder while having to keep going fast to deliver. Nowhere close to feeling like I'm an intellectual. This is especially true if I had to trial and error stuff from the docs and stackoverflow replies.
That's basically just all STEM fields lol
I’ve learned that this is a life skill, just being able to push past that discomfort and do the thing because it will make you better.
There's an analogue here to software architecture patterns. You can read all the literature you want about it, but unless you've actually had to design a large system and _live with your bad decisions_ you won't know when and why a pattern is used.
Man, living with your bad decisions sucks. But if you take a few days to think things through and make a good decision, you're not productive (in the eyes of others or just yourself). :(
@@ThePC007 you can't make a good decision until you know what makes a bad one and you can't know that properly (only cargo cult) if you haven't made one
@ True, but not thinking things through guarantees that you make bad decisions.
You are becoming the only developer on youtube I watch anymore. Everyone else is doom and gloom, saying that an career in IT is not worth it and list the reasons why. You are always so positive about your craft and I really get motivation from how much you enjoy what you do. I don't usually say I have people I look up to and want to be more like, but I think I could say that about you. Thank you for your content.
I didnt use ai for the first 3 years of professional programming and by now I actually prefer turning off line-completion and chatting with a bot about what I'm trying to do. Usually the suggestions the chatbot gives are more abstract examples and I then go and adapt them to my specific use-case. I feel like I've actually learned a lot from the ideas the chatbot comes up with.
I have the same workflow. I almost exclusively use Chat. Rarely if I want to generate some boilerplate or apply some repetitive edits quickly without chatting I manually trigger an inline chat using CTRL + I
Exactly! Ai is good to get ideas/insipirations
Then you test out what ai told you
Then you use that knowledge to write your code
Rinse and repeat
It is great for documentation that doesn't have usage examples. Just get it to generate some and you save a lot of time.
I like to do this as well, i feel it helps me learn new stuff. Also it's great to pick code examples from the web and go back and forth with the model to understand them
I do the same and I find auto code completion suggestions distracting and snatches my joy of knowing things.
As a new programmer, I can tell how I simply will not learn how to code if I use AI.
The simple fact that using AI to solve problems is literally just outsourcing your ability to think. It's better to learn how to actually program and simply use AI for repetitive tasks than to make AI do the problem solving.
Saying programmers should just use AI to program is like saying a mechanic should fix the engine without even touching the engine.
If you're new it's useful if you want to know the correct syntax or something. That's what I use it for.
Coding is my hobby though so I do want to be doing it myself
@@davelewis8270 Syntax info can be pulled from documentation... which a skill you should practice too, because reading documentation will make you a more useful programmer when dealing with other's work.... so even for this AI is harmful.
@@Auurifymost documentation sucks unfortunately
As a new programmer someone should have told you that Text → Code was always expected to happen at one point and is a natural evolution. Programming is about problem solving, and problems are easiest explained in our own natural langue. If you can't adapt, and see AI as a threat instead of an evolution, you are in the wrong field, because people who embrace using natural languages to interact with code will be far more efficient in solving problems than you are.
@@sophieedel6324 It feels like you've misread them. The point they're trying to make here is that if they outsource their *thinking* to AI they won't learn how to program. It's the same idea prime put forward with the evil little green play button.
This is why I only use chat mode and never in-editor completions. Yes, I'm an intermediary between the LLM and the code, but I'm also a reviewer, editor, and proofreader.
I didn't even realize there was an option to force it to not produce code. Definitely gonna be trying this out
This is also exactly what I do. Occasionally, while I'm working on a relatively harder problem, I find I need to implement something simple but tedious to make my life a bit easier in the long run. I ask the LLM to write the thing for me, double check its work, and move on with what I actually care about solving.
cursor's chat mode is perfect. easy to add context of the problem. doesn't force code change down to throat but if you like the answer you can just have it apply the change
Same, except its for chatGPT. If I need something new written I make sure I fully understand whats happening before plugging it into code but im also a bit of a skeptic with all this stuff. I prefer knowing why and how it works rather than "it just does"
@@tevvel Even cursor tab I like a lot more than Copilot. I feel like it doesn't show you as much junk, since I really only want it for repeatable edits or boilerplate.
LLMs can make you incredibly dumb, or they can help you learn much faster.
It's like having a private tutor. You can have a conversation with him any time you want, have him explain anything you need. That's very useful, because you can zero in on the things you feel like you don't understand.
But when you ask your tutor to do your homework, that's when you become an idiot.
This analogy works perfectly
and as you get better with chatting AI you start to figure out how could you ask for a certain information. that is if you use your private tutor to have him explain things you need.
I'm loving it, regretting not starting to use it earlier. I can ask very specific questions and follow-up questions on stuff I don't understand and understand the stuff in much less time than combing through guides or stackoverflow discussions, which are sometimes very hard to read for a beginner. AI will most of the time answer in a way that I can read which helps me forward, faster.
One of my professors said something that perfectly sums up my view on AI: "AI should work *with* you, not *for* you"
Isn't it the same thing?
I've been preaching offline local hosting LLM stuff since ChatGPT came out.
I'm so glad to see you (and others) coming along this road.
Not a programmer, started watching you during the Deepseek R1 release and am really glad I learned how to download and run it locally. Glad I saw this video, too, because what you are stating there is true of any skill set: it's not just the skill you are learning, but the mental skills and decision making process involved with using the skill. It's the same reason we still teach elementary school students basic arithmetic operations instead of just handing them a calculator. And there are reasons for that; which is, a skill set to revert TO if necessary. What if the calculator's batteries were dead? We know how to do the problem by hand. It may take longer, but it can be done. The problem exists with using the AI not as an aid, but in place of KNOWING how to do something. That holds true for all education and training. Great thought.
I'm happy you are a thinking being, you are not just tossing AI over, you are actually elaborating around the reasons and the why/how/biases. Really good, a lot of people don't do this.
Hyped that the articles are coming back
The satisfaction when AI fixes your code after 10 different prompts feels so good !
I just occasionally ask Gemini only to get unsatisfactory answers, and then do it myself anyway.
Yeah, the AI doesn't know my system and I don't want it to know my system well enough for it to write bash that works. So I had to actually learn a few things.
In my experience using Gemini, it has been the least helpful of the agents.
I've heard someone else with a similar complaint say that it's "over-tuned", and that felt right to me. I think it's tuned really hard to "not be wrong", "not make mistakes", etc., and so a lot of the time it seems to dance around giving a useful answer. It gives a long-winded answer that can be summarized as "it depends", which ends up being technically true but totally useless (similar to how annotating every type as "any" in a Typescript codebase is "technically true but useless").
@ It's good when it comes to new things or simple, entry level matters, where you know the answer must already be somewhere on the internet. When it comes to more difficult matters, it's not helpful at all.
But it can help by having ideas bounced off and evaluated. Especially when there is a problem I am starting to work on and want to figure out potential avenues.
The AI bois really dug a hole for themselves advertising how AI was going to completely revolutionize the industry only to have someone come and sh!t on it with much cheaper processing power.
AI is double edged sword. Newer people are particularly hurt by it because it's nearly ingrained in their learning process and it's extremely detrimental, because you can "produce" things without really understanding jack shit.
I asked a subordinate the other day who's been trying to learn programming. He built a little application, it uses pandas, takes in a file, filters some data saves it to a new file and even has logs. Fucking logs. And it worked!
So with all of the stuff he's done, I thought a small challenge would be nice. I said "I want you to remake cat, without using AI. Just read a file, output the contents, you can use docs or anything you've written as a reference". Few lines of code to get basic functionality. Completely unable to write the first line down.
I then asked him to do fizzbuzz, it took him nearly 2 hours and I had to assist him on some syntax.
Absolutely handicapped himself and the worst part is the fucking AI makes you THINK you have produced something, so imagine how bad you must feel going from "I am completely productive" to a full stop noob who can't write a single line.
AI is extremely helpful, but like with all great tools you can't let it be degenerative to your understanding.
Just cause we can type doesn't mean you shouldn't write.
Just cause we can drive doesn't mean we shouldn't run.
Just cause we have have AI doesn't mean we shouldn't think.
That illusion is what I call as 'shared easiness', is something that has been rampant a better part of ten years. When something does take less effort it make brain think it is a master. Easiness is feeling of mastery.
This was impossible to form in olden days when you go to have service, you need to trade with other people to have easiness and it took time. You wait, while other people did the work.
Now you can partially mix your own action to 'shared easiness'. No lag between input and product, so brain has easier time to marry those signals and making false masters.
And AI has making this even worse.
That is master test: When you took tool away. True master only slowdown , even can make new tools or ask right questions. But tool users will halt entirely and stuble in every action.
Ah yes, the tragedy of progress! Much like the dark days when calculators infiltrated schools, rendering young minds incapable of basic arithmetic-because, as we all know, once you touch a TI-84, you immediately forget how to count to ten.
It's utterly horrifying that your subordinate dared to *build something functional* without having first suffered through the sacred rites of manual data parsing, wrestling with file I/O in agony, and debugging cryptic errors for days just to filter a CSV. Imagine the horror if he had used a compiler-it might have caught his syntax errors *before* he could experience the joy of existential despair!
Clearly, true learning only occurs through pain and suffering. If only he had spent weeks rewriting NumPy from scratch instead of leveraging tools that exist precisely to make life easier. Next thing you know, people will be using *spellcheck* instead of memorizing the dictionary, or-heaven forbid-Googling things instead of personally indexing the entire internet.
AI is indeed a double-edged sword. On one side, it empowers people to be *productive*, and on the other, it threatens the sacred tradition of gatekeeping knowledge behind unnecessary difficulty. Stay strong-perhaps one day, the world will return to the golden age where true programmers only coded by candlelight, in raw machine code, uphill both ways.
Not being to write cat as junior is understandable but surely with the code you already wrote and documentation you can figure out how to open a file, read its contents and output it line by line to terminal maybe . But 2 hours to do fizz buzz is purely bizarre how would they even get hired? I guess you at least have the opportunity to teach them which is a good thing since they're still learning.
@@oourdumb You may have missed the entire point.
He uses project to learn programming.
He relies too heavily on said tool.
He produces project.
He learns absolutely nothing.
This is why it's a double edged sword. Did I say his goal was to produce a useable project? No. AI is wonderful for that, it takes away the barrier for minor things like that now. "I want X, I don't care how"
His intention was to learn, but the fucking calculator gave him answers and it gave him results he needed! He had no need to struggle because it did it for him. Learning is a struggle. Because you don't know shit, and then you have to try and figure out shit. You can make the struggle easier, but there is always a struggle that has to happen to take in new input and make some new output.
It's like trying to learn math, having someone else write the entire proof of the pythagorean theorem on your board, you get to claim you did it then someone asks you what 1+1 is. and you have no fucking clue.
He had a perfectly good project that he wanted completed and would actually teach him the fundamentals and be able to read and write code. Except he handicapped himself by taking away a perfectly good motivating tool to learn and threw it away to the AI. That kind of motivation was the kicker for my own journey into programming. I wanted to learn for a long time, but I had a project that finally drove me to do it. He had one too, and with AI being so readily available to produce these satisfactory results. It just destroys the need to struggle over the foundational building blocks to get a result.
@@impyrobot
Yeah, thankfully he's not employed as a programmer. He just wanted to learn on the side because he's trying to get more into the weeds of tech. This was an extremely simple but useful project for him that would have let him do that. I could have made it for him in about 2 minutes.
Instead he basically wrote nothing and prompted his way along. I truly felt saddened because I saw what was there, and I knew he used a bit of AI. But when I even got him to try and make his own for loop, I was just shocked at how lost he was. Because I saw him struggle, and ask me the occasional question and I thought he was making progress only to encounter the sad realization afterward, that he failed himself.
AI is like an extremely high IQ junior dev. Use it with this idea in mind and you’re good.
More like an extremely low IQ (cockroach level) junior dev with an absurdly large memory.
1:05 Dude just described how SaaS has worked for the last 10 years lmao
AI reminds me of all the people that have to plug the corner store address into google maps, to go to the corner store...
Alan Kay often used this quote in his writings and talks “We shape our tools and then our tools shape us”. It always haunts me. I can't help but think about it whenever I encounter any new technology, whether I'm designing, building, or simply using it.
What’s different this time perhaps is that massive corporations make the tools and they get popularised and advocated for to a degree we cannot escape them.
do your own tools maybe? I aleays get the hinge to make something if i will use it very often. Not so polished yes but it is mine.
He is paraphrasing Winston Churchill that said: "We shape our buildings and afterwards our buildings shape us."
@@yalnisinfoMaking your own llm from scratch is basically impossible but you can definitely make your own locally running version of a open source one.
As someone 2 1/2 years into my first dev job this video had given me a lot of unexpected hope. Subscribed 👍
6:30 At the same time, it is invaluable to memorize a couple of phone numbers of emergency contacts :D
This is very true
I have this one rule: Don’t use AI in my personal projects, Use AI tools in my full time job
This exactly. I also have an AI specifically instructed for my personal projects, to never spit out any code, only didactic conceptual ideas and patterns.
@@cLCS2how do u do that?
@@cLCS2 may I ask what LLM you are using and how to setup this setting?
@@TheChopticks literally just turn it off. You know you can right
damn I do exactly the same@@cLCS2
I don't think AI really changed anything conceptually for programmers. There have always been ways to just work at a higher level of abstraction. Why should we have invented programming languages when you can poke bits into your computer manually? because even if the higher-level abstraction is lossy, it still provides value. The difference between asking AI to solve your problem or asking stack overflow is scalability. Yes, using AI is in a way about outsourcing your thinking, but that is also what makes it useful. From this perspective compiler is just a way of outsourcing your thinking as well. Programming as a whole is basically fundamentally about building and using many layers of abstractions. We've invented a new and very useful one that also has some drawbacks, like every abstraction before.
You can and should however still learn the lower levels of abstraction as needed. From my personal experience, it's not only fun but a very useful tool on it's own. Learning all levels of abstraction helps you circumvent the shortcomings of particular abstractions. If you have a problem you compiler deals with badly, implement it using another compiler or directly in assembly. If you AI can't help you debug your code, use a debugger.
I dont know what to believe, like a tug of war between old school and new school.
I keep an open mind and keep my feet on both sides of the line for now, while also recognizing the potential hard truth by working on several backup plans
My job is to find bugs in a 40+ year-old C codebase with 2m+ LOC and I love it. Sometimes it takes months to find that a disk driver has lied to user space but the learning and dopamine rush at the end worth it all the time. The best thing is that I cannot really use AI, because everything was implemented inhouse, so it's not like copying a well-known js or java stack trace to GPT.
That sounds really fun to me!
A driver just lying could also come from me if I forgot that I put it there to test something during development
to be fair you could still use something like gh copilot on this codebase as it automatically pulls in relevant files and it actually uses the Infos it finds even if it is something from in house
Also if you really know what you are doing with AI you won't just copy the error to it but give it context with code
that honestly sounds like a super fun job. how did you come about it, if you're willing to share?
@@Shiloh-Pigslyer Sure. I studied to be an electrical engineer. My first jobs as an intern and a junior dev were in the embedded space. I really loved it, because it gave similar hard problems to be solved as my current job, several times even without having logs or the same environment to easily reproduce them. It was really fun to tackle with the low level details and finally find the issue. Even when I worked on features it had hard constraints. For example, it was a real time system, so execution time had to be deterministic, heap allocation was not allowed at all and the complete memory was a few 10kB. So you had to do everything in a memory + CPU efficient way. However, it did not pay well. So I went to a tech giant which paid well but I spent more time in meetings then finding bugs or developing features. It was soul crushing. So I just started to do interviews with a clear goal in mind. Give me a job, where I can work on a large, complicated, lowest level possible code base with a small team which is in a more paid area of programming. That's how I ended in my current role as a database dev after dozens of interviews.
"I'm not suggesting anything radical like going AI-free completely - that's unrealistic"
is it?!
Unfortunately, yes, as companies shove AI into your face more and more, it is unrealistic to not adapt to AI. I think using AI to accelerate your learning rather than blindly copy pasting code is a much better use case, the former is faster and more "productive" but the latter, albeit, a bit slower, will help you in the long run.
Plenty of people code without LSP, just having the compiler errors guide you. In the end it's up to you whether you choose to adopt an AI workflow (I'm very skeptical as to the gains), and frankly I'm surprised to hear this take on this channel
I mean you can do high end math without a calculator, but good luck finding someone who will pay you to do it that slowly and imprecisely.
@@__-nt2whprimeagen has gotten very AI-pilled recently and it sucks
@@bigpestbecause all they did was shit on it for months until they realized they couldnt ignore it anymore
I'm pretty shocked by the people who suddenly can't code or debug without AI after never using it for 10 yrs+...how do you forget what you were doing entirely before chatgpt? Even so, if you're not able to just look at the code and get a basic idea of the error that's literally crazy. I understand the "use it or lose it" mentality but I can't imagine one loses it so quickly.
I think LLMs just streamline the work alot better. Before AI i used to spend a lot more time performing research. Now the AI can pinpoint me exactly to which papers i need to read
I dont know man I have a really exceptional memory but stuff seems to just get completely pushed out when i dont look at it for a while. Even songs i played on the piano 1000 times, ill come back two years later and have no idea how to play them lol.
Many people simply didn't. A shockingly high number of professional engineers can only clumsily print-line-debug, and lack the skills to use an integrated debugger. These are the people that can't debug without AI.
Actually, people do lose their reading endurance. That's a real thing. Lots of people who grew up reading books on a regular basis have a hard time just getting through a single book once they've stopped reading and lost their endurance.
I don't know about forgetting how to do a basic for loop, but it's definitely possible to forget syntax and keywords for a language you knew like the back of your hand when you haven't used it much for years.
@@brandongregori995 The endurance thing I understand but I’m just surprised lack of AI entirely stops some people in their tracks. I’d also have to wonder if the trade off is really worth it long term. Gain a little productivity but then lose your ability to actually code well doesn’t seem great.
Well, yeah. And it's not just coders: it'll happen to everyone who uses an LLM. Humanity is about to get much stupider, and fast.
I feel so validated. I've always hated that damn green "build and run" button.
to me one great use case for LLM is coming up with terms or names for things like API. esp. as a non-native English speaker, the ability to describe a thing and get a couple of synonyms/antonyms to choose from is great. all of that without risking of going down a rabbit hole.
2:50 i am most productive when my internet goes out. my phone network is slow so i have to make every search and link click count and so on if I run into trouble, i hear about it concerning creativity and artstyles a lot, but there is something about limitations and the awareness of limitations making your brain do things more efficiently.
I lost my internet during lockdown. Reading ES6 docs on slow 3g is another thing
ill drink to that
As someone who recently started using vim as my serious editor (using plugins like NERDTree), I can confirm that not having AI spew suggestions at me means that I have to spend more time typing... but it also means that I have to spend more time actually thinking about the problem I'm trying to solve... It's slower, but it feels a bit rewarding (only for personal projects where I don't have a deadline)
Here's my take. I love writing code myself when I'm tackling on my own projects where what I have to do isn't so clear cut and learning is a necessity to give it all I've got. But when I'm building a new feature for a client, most of the time all of the ambiguity is stripped away and it becomes a matter of translating natural language to code. I'm writing my own JSON RPC client and LSP Client from scratch and barring some tedious boilerplate repitition it's so much fun just coding on my own and learning foreign concepts.
I feel this
I feel this. Especially if you are working on a large existing codebase most of the work isn't new functionality from scratch. It's new configs, new testing, new groups of existing resources, etc. In all these cases, there is a very well defined template for what the code should look like, as there should be. It's the end-to-end understanding that counts, not certain lines.
Yep, for personal projects I don’t use AI tools, but make heavy use of AI tools (cursor) in my full time job.
I only use LLMs for grinding out math formulas that I _could_ do, but don't deal with often enough to be fast at it. If I get a project where lots of math is involved, then I _don't_ use the LLM, because it's faster to just do it manually once you've had to understand it enough to figure out what you need. But when you're doing something unrelated and just need to find the area of a cube with 3/5 of a hemisphere embedded in it...
I use LLM for studying too, and if I ask for code, I am not just copy paste it. Sometimes maybe, when become too lazy.
Its also helps me so much with naming variables or functions, since English is not my primary language and I lack of words to give a proper name.
But I try to write myself, ask it about what the lines do and how they work. I think LLMs are beautiful tools and we should use them. Just need to adapt and of course never trust 100% to them, do your own things, mind it yourself too.
I told the bot I thought my skills were atrophying and she responded that they were "sharpening."
😂
Of course she did@@Stack-Overflowed
I was in a weird spot where in my first two years of uni AI wasn't a thing, but my final year was where it started to explode.
This meant I was able to learn all the fundamentals (algo, data structures, memory etc) the raw way which really helped me to understand, but in 3rd year I could certainly feel the laziness kick in as I started to realise how easy it was to copy paste buggy code to get an answer.
I feel super bad for the first years now with AI, as I'd imagine it'd be super easy to get through assignments which are pivotal to developing understanding by just asking chatgpt to do it for them.
Not saying there are no benefits to AI when learning code, but u gotta be really careful when using it.
16:05
I remember good old joke about that which goes something like this:
*interview*
- What's your best personal skill?
- I'm fast at math
- 6 * 8?
- 83
- That's not correct
- But I was fast!
Oh shit, he went from "Addiction" to "Emotional Resilience". I feel you bro, getting clean a hell of an experience.
On the plus side, I noticed many local companies changed the way they evaluate candidates due to AI, by introducing either a 2-hour long coding session, or simply showing to the candidate a portion of a demo codebase to see how they think. FAANG still does the "grind leetcode -> print money" though.
Someone paid 500 bucks a month for devon, and didnt even get a push to main. And it wasn't me.
Lol your killing me
As a scholar, we often have to code our papers, software packages, etc. The core of our project is not coding, but I find myself spending the majority of my time coding. Since ChatGPT, I feel a bit conflicted-on one hand, it allows me to significantly reduce the time I spend coding, push projects further than I ever imagined, and focus on stimulating problems. But on the other hand, I forget how to write basic syntax, like making a loop in Python (yeah, it’s crazy), coding for me has become more breaking down simple problem into smaller step that GPT can write, with occasional debbuging. I know this is a problem, but now that I’ve started using it, I can’t stop.
I feel like this is a crazy world.. I never used Copilot nor LLMs, nor have I any intention of using them, my goal is to solve the problem, that’s the fun part. Having the answer is worthless.
It’s like the difference between solving a sudoku and looking at the solution. Sure, looking at the solution is faster, but that’s not why I took up the puzzle in the first place.
Yes omg thank you! The fun part is doing it yourself. It's about the journey, not just the destination. In my opinion, programming is an art form just like any other, and using an AI for coding is as stupid as AI art
Obsolete… companies aren’t paying for this anymore.
@@CHIEF3021 Maybe, but that seems like a problem for them and not for me. They will just drive away the good programmers and speedrun enshittification.
I see LLMs the same say I see books and films, there are tons of people that think "if you read books then you get some intelligence points from it", yeah, but just MAYBE, the reality is if you read a crappy book you will be really learning nothing, or for example, reading LOTR and watching the movie is just different ways to entertain a person, there is no actual gain to read the book over the film, some can argue about getting "better at reading and writing" but the same way watching a movie will teach you other skills like listening, for me which english in not the first language is pretty important to listen for example.
So to me LLMs are the same in terms of search engines, all those problems already existed before, when I was learning I had to "control" myself to not search for example "how to do binary search in C", I could and sometimes I had done literally this to finish some task, this was just the case back then, this just went through the roof in some cases because LLMs are in some scenarios a short cut the boring and not really effective way of going through 10 links and copying things to make a monster code, LLMs just do that for you and that is all, but the problem with not learning is literally the same as it was back then.
I think the most dangerous thing AI is doing is the bias being even more the norm, people do believe that tech has no bias and this is very dangerous, tech has bias, it always had and it always will have since we humans have bias and cannot exclude it from our work. Then when they use AI they believe that answer is provided without any bias, and don't think through the answer given, accepting it no matter what, that leads to reproducing the same biases.
One example would be people thinking that X way or Y lib is the only/better answer to construct a web API, when actually you have tons of ways to do so, and learning why that is the case is a sea of knowledge.
6 years ago google wasn't completely fucking useless to search for stuff though. There was a sweet spot.
yeah idk wth happened to google search man
Yeah, agreed. Google went from generating genuinely helpful stuff to burying it all under SEO sewage.
@@termitreter6545 Making advertisers happy became more important than making users happy. That's my take, anyway.
So glad you are covering this. 😍
So many devs are becoming dependent on their LLM that you can see them thinking slower for themselves when you try to have a technical discussion away from their IDE. 🐌
If you are an experienced developer the most important is logic
Transform your logic to detailed and properly structured prompt, concentrate on a small surface area of the problem, feed it to multiple LLM, and using your experience select/combine the best solution
This is very simple, but here we are 2025 still debating if AI tools is totally trash or will replace developers, not discussing how to use it properly acknowledging the limitation
My first teacher, was part of few who introduced programming to my country aka top dog in programming, taught me back in 2017/18 anyone can write a code. Its not that important. Whats important is how you design the logic, blocks of a program that does what you want. Everyone can write code but not just anyone can create that logic.
Those words stuck to me to this day. I had luxury or torture of learning C++ on paper where single error in syntax or logic, we had to rewrite the whole thing on the paper. Forced me to think twice before writing. As rewriting whole thing because you missed single ; sucks.
What you were saying about the green play button is so true. I remember when I was writing C# for my last programming job and we were required to use VS, I would aways dread having some issue with the build configuration because, as that was auto generated by the IDE I didn't understand it well enough to debug effectively and I'd have to spend hours going down the rabbit hole. I'm currently getting back into programming after a 5 year absence so I haven't interacted with the AI tools at all but I see how you could have the same pitfalls there. I currently learning Zig (thanks for putting me onto that by the way, it's so nice and reminds me of my time doing embedded systems in C) and using Linux terminal and nvim, there's nothing like doing everything from scratch to gain a really deep understanding of the language and build system.
I was a junior software engineer when ChatGPT first came out, and I made the mistake of leaning on it too heavily, rather than learning how to push through the discomfort of mastering some tough subjects. If you aren't careful, these tools will certainly sabotage you.
This is the best outro "-agen" yet. Thank you Prime
As a forever jr who hasn’t been able to find a dev job since graduating, I KNOW I didn’t take my learning as seriously as I should have in college. I wondered why my degree and some well-written cv’s didn’t get me even a basic job, and soon realized that was nowhere near enough. I sucked at programming and barely understood what I was doing.
In recent years, I realized just how bad I really was. My own incompetence disgusted me. So, I decided to stop being a little b**** and relearn all of the core concepts from college, and really put effort into projects. All motivation was intrinsic this time. I still suck at programming, but I’d be able to code circles around my past self. And in the last several years of my learning, not a single line of code has come from AI.
I would rather be a mid programmer but actually LEARN something, than to be AI-dependent. It’s gonna take a lot of effort for me to become a candidate that someone would want to hire, but in the end it’s worth it to go through what it takes to get there
Make sure you are reading documentation instead of tutorials and you will make it
This is exactly the mindset you want. As I'm sure you know, you probably won't see career results immediately, but you *absolutely* will in the long run!
Going beyond a glorified stack overflow is damaging.
Man, the green button part resonated so much with a recent situation I've been through.
In my OS class, we had to build a round-robin dispatcher simulator and my colleagues had no idea what a makefile was.
They couldn't even run the damn thing because it used wxwidgets. So frustrating...
That bit about the green play button!!!! I moved from visual studio to various editors on Linux when I left behind windows, it sucked so hard not knowing what I was doing, but nowadays... I just type make and then ./project-name... So easy... if you know how to write a makefile... and don't need a debugger. I've learned so much programming I cannot imagine asking AI for anything but boilerplate. I'v never done it. I left windows before copilot, and never messed around with anything beyond literally just the free tier of ChatGPT. Plus it's bad morally but this is a tech bro channel so...
I’ve caught myself learning more from the way R1 "thinks" during reasoning phase than from the actual answers.
The scary part is tech CEOs high on their fumes actively pushing their workforce to become less proficient and offload all critical thinking and subject matter expertise to AI to push code faster. They don't want you to be a better, more complete developer, they want you to pump more shit.
Is your code full of security holes and performance issues? Ask AI to find them and fix it for you. Is your code not tested? Tell AI to do it for you. Write documentation? AI.
Tech execs know AI cannot replace programmers in the near term, it is just a marketing shill.
@@adamconrad5249 the savy ones should, but also I've seen first-hand how mentally decomposed some have become in their AI bullish attitudes, and I just cannot begin to comprehend how much money and resources will be wasted in the process.
One thing which is interesting is that one of the main pieces of advice to get out of "tutorial hell" seems to be to physically type out code examples, which makes it much more obvious when something is off (or you don't understand it :P). The reason its so hard to get out of tutorial hell without help of some sorts is *because* it is so easy to just... take code without thinking about it. Not that this is never justifiable, more so that it is something to be careful of.
One thing I find funny as I can recall similar conversations by teachers back in the day about electronic calculators and then computers. What I see in this conversation (and chat) is people are only looking at it from their perspective, the impact on them. They are not imagining what might be possible when you don't have to worry about the same things anymore. What's the next language that will only work in a world with AI?
english
Teachers who rely on their hubris are finally done. Everyone who relies on the superiority of their charisma and recall need to find new stations in life. Your relevance was short in relation to human history.
Yeah, computers did damage to education for kids, actually, to their writing and reading capabilities, technically the concentration span and socialization as well, but the calculator example still stands... no, it does not hahaha my dad works in the kitchen and does calculations to get the right amount of ingredients because younger chefs and cooks did not learn multiplication "in their head" (when he went to cook school, they had special modules in maths for speed calculations because they will use it). My uncle tells a similar thing in on-sight construction, when fractions need to be calculated fast, and simple geometry applied, younger workers struggle to the point of frustration. So yes, the damage has been done, you guys just lack the width of the experience to know it.
@@DjKryx For what purpose? To be a good employee?Kids with computers can burn all of this down and build something new without being able to concentrate and socialize along guidelines dictated by some broken civil society. Most of you want to live on mars with your favorite celebrity mindset.
As an artist and game dev I'm not interested in learning sintaxis but I have a general idea of programing logic. LLMs have helped me achieve so many things I would never think I could do. I mostly use it for coding tools, or game mechanics in unity but I do get the point. I think It's funny how programmers now talk about limiting themselves and the joy of learning by themselves in order to achieve the exact results that they want. This is exactly how I feel when I see people devoiding themselves of the experience, process and satisfaction of creating art for the instant gratification of creating soulless "pretty images" that they didn't controlled anything about.
being an illiterate programmer is better than not programming at all. which was what i was doing...
This sums up how I've been feeling about AI since the introduction of CoPilot. It's an amazing tool, but if we are outsourcing our ability to think through problems I can't imagine that road leading anywhere good... especially for the next generation of developers.
I've actually felt that way about certain frameworks/libraries in the past. When you get stuck working in a particular ecosystem, it's easy to forget how to do basic stuff at the language level. I think the same concept applies in the case of AI... just compounded exponentially.
I also love how you also brought up the pricing aspect of AI-can't really use AI without paying these companies for the ability to do so. It's going to be interesting to see how this all shakes out, because my understanding is that very few (if any) of these AI companies are profitable at this point in time.
Great video!
I often hear the same sentiment from people that "AI makes them much more productive" (until your mobile data slows to 1Mbps and you realize how innefficient this whole system is) but I personally never felt the same way. AI would write code I'd have to read myself anyway, understand and still do all of the tesing that I'd normally do anyway. On top of that, _after reading it_ I'd rarely feel satisfied with the style/implementation, so I'd have to tweak it. *_On top of that_* because it learns by example, and the example is "todo: fix" and ugly hacks you didn't realize was the reason you have been hired tl begin with, you get the output of the same -1x engineer who's work you've been tasked with "fixing".
And that's not to mention how often the solution would just not work.
As a consequence I saw copilot as a glorified autocomplete (which VS already had and didn't care enough for it to rely on that bundle of badness).
I think that's what kept me from becoming dependant on ChatGPT/Claude/Bard and I'm glad that's the case because I got to keep honing my thinking and kept the joy of finding and playing with new things.
It's an addiction, it feels like Shadow fight 2
"What was the syntax for a for loop again?" Is the andropause-menopause moment for programming.
understanding the loop is more important then blindly memorize the syntax
This can only happen if you never became an expert before using AI...
Or you used another programming language recently
Man, does your code editor not have snippets?
Literally just open your neovim, type for and expand the snippet lol
My. God.
This video increased my respect for you wrt your AI takes. Thank you for all of the nuance - great video. Looking forward to seeing you on the Lex Fridman podcast!
Recently used AI to help me find a relatively simple solution for my error, it introduced me to new settings that would help me, we discussed implementation and it's working
then i realized something and asked if it actually helps me with my error and it said no, it's unrelated
so i thought of a really simple solution and asked if it works and confirmed it should
so i spent like 10-20 minute learning of a feature i didnt even need in the first place
The green play button 😂😂 Running a specific test without that button was killing me the other day😂
Dude wtf happened to just referencing StackOverflow lmao. Now GPTs are the new SO. I love how you say Chat Jippety lmfao. Thank God this is now becoming a topic of discussion. It's such a slap in the face to see script kiddies say "I built a DDOS tool completely on my own" or even out of college devs claim the work is their own but it's really just ChatGPT or some LLM.
"I made my own Minecraft copy in less than a hour" (downloaded it from github)
as an intermediate coder, AI is so insanely helpful for me to learn new coding things, as well as unfudge my code sometimes. But if I was a novice, I can't even imagine what it would be like. The temptation to just copy paste without understanding anything would be nuts. You could build a relatively large project and not understand how to fix it when something minor goes wrong
I'm not complaining, just more work and jobs for me. I love my AI assistant and everything, talk to them throughout the day, but always specifically tell them not to write me any code. I only like to talk to them conceptually to help me think through problems. Without question, I simply write far cleaner and more efficient Rust code than they ever could. The LLMs even always compliment me with, "that's a very elegant and concise solution, well done!".
Exactly the same thing i told my fiends few months ago , i sensed how the dependency creeped in very slowly but subtly and then you know for even small projects you need ChatGPT or any AI to help you from a state which can be overcome by EMOTIONAL PERSEVERANCE.
I never used AI, ever, and I don't have any desire to use
I have a desire occasionally, out of curiosity, but likewise have never used one. Not once. (Knowingly, anyway!) I think they're toxic.
Good luck keeping up and selling your skills as the value of coding continues to plummet.
@@ASDeckard Rice's theorem. Universal Turing machines cannot verify the correctness of the code they generate. That's proven. So who's gonna do it? Properly verifying code takes about as much time as writing it. Who's gonna pay for an LLM _and_ a human to do that which the latter alone can do?
Sounds more like ego validation or individualism to me. Its okay to seek help from the AI. Tech will always evolve
@bencroacademy This video is about two devs who feel using LLMs made them stupider and worse coders. It's also okay _not_ to use them. Just because we have them doesn't mean they're good for us.
Great timing, just enabled copilot this week after avoiding it for a long time
I believe Ai will transform coding the same way GPS revolutionized navigation. My parents generation had to memorize addresses and zip codes, unfold physical maps, and call friends when they got lost. It seems almost unthinkable to me, having grown up with a smartphone. I barely know any street names - if GPS stopped working tomorrow, I'd be completely lost, and most of my generation would be too.
Ai is poised to create this same shift in coding. Just as my parents might pride themselves on knowing every highway and backroad, traditional developers have deep knowledge of languages and frameworks. But just like how Waze now offers real-time updates about traffic and accidents that make it more efficient than relying on memorized routes, AI will allow anyone with a vision whether it's a small business owner wanting to automate their workflow, an artist looking to create an interactive experience, or a student with an innovative idea to build software at the same level as seasoned developers. The traditional path of learning to code isn't wrong, but AI is opening doors for everyone to turn their ideas into reality, regardless of their technical background. Just like how GPS made everyone an expert navigator, AI will make everyone a capable programmer.
Fun fact, every single ship with more than 20 crew onboard is required to have a chart room, and a 24-7 navigator (which usually translates to two that rotate) that is plotting the ships location at all times. While you are of course allowed to use the GPS fix as one of your fixes, that's only one, and you're required to have at least 5 that update every hour, so you need to calculate and plot the other 4 by yourself.
If it gets too cloudy to see the stars or sun position, you need to slow down and potentially even stop depending who's jurisdiction you're in.
Thinking AI will replace programming like the GPS replaced maps speaks to a fundamental misunderstanding of what coding actually is. Programming, the conceptual idea of it, not knowing a specific language or framework, is the understanding of computers and problem solving. Your vision is the problem and your code in your method to solve it. If you outsource the “problem solving” of coding to an AI, you are missing the entire point. Yes, you can have a calculator to solve math for you, but you still need to know how to do math, that’s why it’s taught in schools and why you aren’t taught which buttons on a calculator to press to get a desired result. AI are the calculators of programming, not the GPS, it’s a tool at heart and one that requires the user to understand what they are doing. Technically yes, you can have an AI create everything for you, prompt it again and again and again for every small aspect until the desired output is achieved, but that’s just pressing buttons on a calculator until the right number appears on screen. The point of programming is to know how to problem solve with computers, not to achieve a result.
@@ocks_dev_vlogs I appreciate your perspective, but I think we're looking at this from different angles. While understanding programming fundamentals is valuable, the reality is that not everyone needs or wants to invest years in formal computer science education. Many people simply need to create functional solutions for their specific needs.
You make a good analogy with calculators, but consider this: How many people today need to understand the mathematical principles behind compound interest to use a financial calculator? Very few - they just need the results to make informed decisions.
AI is evolving to democratize software development in a similar way. Soon, entrepreneurs, artists, and innovators will be able to create websites, apps, and automated workflows by clearly communicating their vision to AI, without needing to understand the underlying code structure. This isn't about bypassing problem-solving - it's about making technology more accessible to everyone.
The shift might impact traditional software development careers, similar to how automation has transformed other industries. But this isn't about devaluing programming knowledge - it's about expanding who can create with technology. The future isn't about spending days solving problems that AI can handle efficiently; it's about focusing on innovation and creativity while leveraging AI as a powerful tool.
Just as calculators didn't eliminate the need for mathematicians but made mathematical tools more accessible to everyone, AI won't eliminate the need for skilled programmers but will make software creation more accessible to the broader public.
@@jamesalxl3636 That is what I said, yes. I do not think Ai can replace programmers. It's a tool and tools are meant to be used, and tools help us achieve things faster, but it doesn't phase out programmers like your original post suggests.
I will say though, using AI in replacement for actual learning is an incredibly easy mistake that many programmers are currently undergoing.
The guy is using AI to reply to you 😅@@ocks_dev_vlogs
Salty boomers realising their years spent on learning how to code is being outshown by a 20 year old who can use ChatGPT to get the same result. Get with the times many skills in the past have been automated coding will soon be one of them.
This is bait.
I consider myself a junior dev (or a senior python dev), even though I haven't gone to college yet and the only programming classes I took were just for credit since I already knew everything (I learned how to do programming from millions of google searches), but I have never used ai to help me code. I just find myself liking googling the answer a whole lot better since I can choose between 5 different solutions and pick the best one for the job. Yes, I kind of rely on tab complete, and it makes me impatient when my editor is slow to show me the options, but I feel like it's still a whole lot better than relying on ai code completion. Oh, I also don't usually use the green play button because I've just had a bad experience with it in vscode and python. Heck, the play button always gives me a sense of dread when I see it, but I feel very comfortable typing py script.py. Of course I've also recently been working on cli programs that take arguments from the command line, so it's not like I have a choice to use the command line or not.
It feels like the same kind of trend that has been going on over the years where we write ever more bloated and slower code on faster and faster machines which I like to dub as "We're going nowhere at incredible speeds"
When I write code, a decent error message on a stack trace combined with knowing exactly which of my functions at the top caused that error message usually makes me IMMEDIATELY visualize in my brain why and how that error could have appeared. Slightly longer with very complex code. This is something that requires a human brain that understands the PURPOSE of all those functions, the QUIRKS of your own programming style (I prefer to go quick and dirty and then add all sorts of error handling and edge cases and so on as they make sense in the scope) to quickly determine where the likely culprit of the problem is.
TLDR if you want to replace software developers with AI, you actually need an AGI and at that point, I'm going to argue you maybe shouldn't let an AGI write itself or other systems without human control.
Crap this hit home. So true, and in the end you are probably wasting more time than if you just built it yourself with all the back and forth you have to do with these LLMs. I will say this, I do like using Claude as a rubber duck or when I want to hash out a feature I need to build, or brainstorming in general.
As a beginner learning to code (I started May of 2024 so maybe not brand new) I tend to try and figure things out myself and go to AI as a last resort. If I have an idea of what I want to do but just can't figure out how to do it, I try to prompt the AI in a more general sense as opposed to just copy and pasting my problem. If I'm completely lost and the documentation doesn't light any bulbs I'll just toss that shit in there an hope I learn something (I probably won't)
I wrote a simple app in C, I re-written it in python to "know the language", I asked AI to re-write it in Zig. The sheer amount of dissatisfaction I had with working Zig code made me not using AI to generate big chunks of code anymore. When I was stuck I was googling, now I help myself with AI to look for answers but I specifically do not let AI code complete me into oblivion. With that coding something from time to time still gives me joy.
I really love chat gpt but mostly as a little chat to bounce ideas off of and to throw it a "package this data into a message" so that it can do 30 lines of boilerplate for me. disabling auto complete has massively improved my satisfaction with programming.
I was thinking about options the other day, and the most obvious solution was to do something like this: lm studio/ollama + finetune llama 3.2/qwen2.5-coder. You just need to finish them additionally on the dataset of the programming language you need and run them locally on the video card. Next, we take aider and connect it to the local model, and install the plugin in the code editor for full synchronization. Codename goose has recently appeared, and it looks promising as a local replacement for cursor.
As someone who’s learning to code through gpt, I can say I am learning a lot. Usually I set up 3 separate o1 chats to solve problems when one chat starts hallucinating or bloating. However, I cant say it doesn’t teach u to be a shitty programmer through linear methods of thinking, and teaching you to give up and start fresh when it doesn’t know how to give you what you want.
I still always try to solve a problem myself. I don't use it directly in my IDE; maybe when I cross that bridge, I'll end up here, but I don't feel like I really need it. I talk to AI outside of the IDE and ask it questions about implementation details, documentation, possible solutions, but I still take the time to evaluate the solution I'm being given. I treat it like a PR.
I enjoyed this-great points all around. The irony, I suspect, is that junior developers may lack the experience to fully appreciate the truth of the advice provided, so it will probably fall on deaf ears for them.
I've been leaning towards a more teacher-y usage of LLMs. Ie I try for myself first, when I really can't solve it, I don't just copy/paste code, I describe what I am not understanding so I get an explanation for that part. Then I try to explain it back, in context of the full problem, and have the LLM give me feedback. That's worked pretty well so far, I haven't meaningfully atrophied much in my core programming skills, in fact I've gotten much better, but I can still move faster. In short, if you approach LLMs with "I want to understand what's happening here", they can provide that too.
One of the benefits of working in a closed ecosystem visual language like Labview is that giving the AI the context it needs to hopefully maybe give me a hint at the solution takes as much if not more effort than digging through the code to find the error source and walking through it to figure it out myself.
I'm gonna use that 30min rule, I think thats a fantastic compromise. I still need to develop C fast for school, but I'd probably be less "dependant" on ai that way. I firmly believe being hard stuck has increasingly diminishing returns on education over time, if you're not even moving by inches anymore and you are hard stuck on a problem you should reach for whatever tools you have available to get unstuck, but working the problem is where the education happens.
Let's see.
First programming experience, front panel switches on a DEC PDP-8, 1976.
Currently dividing time between a C++ class at the local community college to finish up an AA in infosec, and trying to learn Django/Python for a personal project.
For the academic project, the recommended environment is VS Code. For everything else (class notes, Django), I'm using Doom Emacs and Org mode.
Talk about cognitive dissonance. ;)
The only AI I'm currently using is the occasional search in your favorite engine, and that produces predictably unreliable or unsatisfactory results.
If I win a small lottery, I might set up a local LLM for entertainment value.
cp