.... Ctrl + c will copy, Ctrl + x will cut .... and if you setup your shortcuts in vscode ... Ctrl +d can duplicate a line ;) oh ... Ctrl + shift + v will past unformatted
@@yodasoja2011 "Or click the freaking "copy code" button! No highlighting necessary!" I am glad that this is an issue for other people. I was beginning to wonder if I was just a moron.
@@EricChiEricIt does, but also grabs everything from button text to page upload dates, etc. I've started just asking ChatGPT to scrape them for me... XD
@@Matt-jv4oj same I did last night it's so funny how we are all cresting the same things at the same time. I had the idea to do it a couple months ago but I just didn't believe it could be done until I saw that myself with the auto GPT thing
I'm impressed by what GPT 3.5 and 4 can do. And the initial effect is super dazzling, but I always feel like I'm talking to a data set, not a problem solver. Like it can alleviate menial tasks, but a lot of the cracks start to show as soon as you assume too much about its ability to solve problems that require more than one level of abstraction
@@pvanukoff pretty sure we’re approaching an asymptote. 3.5 -> 4 is the same model but with slightly different hyper parameters and throwing more hardware at the problem. While 5 looming and not much is known about it, I doubt it will be all that fundamentally different.
@@hamm8934 I'm of the same opinion, LLM's are a search/discovery centric technology not necessarily problem solving and I don't see these out of the box developing architecturally complex stuff.
I was literally just thinking about how I can structure ChatGPT prompts to teach myself new languages. However, your method of turning it into pseudocode is brilliant! I'd buy your AI Course in a heartbeat.
Nah, I prefer doing everything with the mouse. Typing with the mouse on a glide keyboard on my Android phone. Phones have the best IDEs. Joma Tech did a video where he tried to code a full app on his phone.
As intermediate Java and Python developer, I also really try to avoid javascript and php. So do not worry, at some point anybody wants to cry because of arrow operators, 200 word long lines and async await. At some point you accept it.
Wow, this masterclass on AI-driven development for programmers by Fireship is simply amazing! The instructor's expertise in the field is truly impressive, and the way they break down complex concepts into simple, easy-to-understand terms is remarkable. The examples used in the video are practical and relevant, and I feel like I've gained a whole new perspective on AI development after watching this. Thank you, Fireship, for this incredible tutorial - I can't wait to implement these concepts in my own work! (Thanks for the video!)
Been doing this since it came out, it’s insanely awesome. But for you beginners out there, go learn your basics before trying to fully use Ai to develop. You will find yourself more confused when it doesn’t run if you don’t have those basics down yet.
Is all this still worth learning for beginners ? I've been learning React for the last 3 month but recently feel like it's just a waste of time with all that AI thing developing.
I found that it helps if you let ChatGPT first explain concepts about code (relevant to your problem) and then ask it to either create or explain code.
GPT4 has up to 32k tokens of context, so technically you can just copy paste it let say 5 of your classes and ask it to create the 6th (damn complex) missing one :D
Maybe this is fixed in 4, but you forgot to add the part where the code doesn't work so you ask it to fix it in 5 different ways. Then ask it to write tests for it, tell it the codes fails its own tests, and then it still doesn't put out a solution that your senior devs will approve of during review. They will then come up with a much more elegant solution in 5 minutes, and you'll feel bad you wasted an hour talking to chatgpt.
You are right! I have been using this chatgpt for my daily coding work! I can’t understand the hype. This garbage hasn’t able to write a freaking regexp for me. Try to ask to write a freaking pattent to you to parse different kind of paths. It cant! It cant even add numbers correctly! This is the scam of the XXI. century!
No it is not fixed in 4, absolutely maddening, I see all these influencers showing it working perfectly when I'm getting unusable garbage multiple times per session
The word you're looking for is data-driven (not data-oriented) development, also called config-driven development in some industries. We use this in gamedev a fair bit, like for building procedural elements and menus or storing sequences of user actions. Some engines are wholly based on it. You store information in an easily parsed format like yaml, json, csv, etc. and let that information control the behavior or visual appearance of your program (or a part of it). The actual code is just a parser that implements it, and you could have multiple versions of this for different languages or build environments etc.
Ghost, not quite. Data-oriented programming is when you design the code in a way that is highly cpu-efficient, where all related data is stored together at runtime in plain arrays etc. Data-driven development says nothing about the runtime data structure or efficiency of the code, just that you initially describe it in a plaintext format and later convert it to an application / feature (not necessarily at runtime, though it can be). That conversion can be done in a way that creates highly cpu-efficient code if you choose to implement those features in your design and parser. But you can also generate more typical oop-structures with it or store data in inefficient ways etc.
What he said about complexity and efficiency is very true, the more I use chat GPT for basic code debugging i've learned not to overleverage it, for example if im having trouble with tree structures in C++ I know confidently chat GPT can help me organize it, however if I were to give it a larger scope of my program it starts to hallucinate on me.
@@walterlotte4215 Me too require multiple times typing questions but ot will correct and give finally what we need, Now i can focus on actual business logic and architecture part than mundane coding
if it ends up outputing too much, you can say "continue from where you left off at (the last thing it typed)" and it will finish it. But yea, just having it do one function is the way to go, and starting a new convo when you do somthin new so its not trying to look at old work. Sometiems ill start with an example of the format I want and thatw orks pretty good.
It's wild how I decided to become a developer in 2020 and only 3 years later now I have a job but the whole industry has been transformed so much by AI. Just when I thought I knew where I stood, now I'm completely perplexed about the future.
Don't worry, man. This only looks good on YT videos. If you are worried, just think of some big project you always wanted to make but you never had time, and try to do it with AI. I mean, if this gives you 10x, you should make it real fast, right? No. It can't really do that much programming. I tried to use it to write me some Rust code, and in many cases, the code it spits out won't even compile (I'm talking GPT-4 here). Not to mention, it can make up non-existing functions in libraries. Basically, it likes to make up a lot of stuff like not working logic which only looks like reasonable code, but it doesn't work as it should. On first interaction, I was amazed how good GPT-4 is, but the more I used it, the more I realized it's not that smart. Now, of course, this will get better and better with time, but the question is, can it really create complex projects on its own? I hear about self-driving cars for years, but somehow we are still not there yet. There is also a hardware limitation on how much context you can have at once. If AI will be able to do programming on its own without making too many bugs, then it will be able to do any other intellectual work. No need for lawyers, doctors, accountants, architects, and so on. If that happens, we will either see the world collapse or the economy will be completely redefined. TL;DR We are still far away from AI replacing people, and when it happens, everybody who does their work with their brains, not with their muscles, will be f***ed.
I personally think change GTP is not smart not even second grade smart it's just have anything from the internet and making blender like it has all the pictures on the internet so it's just altering them to fit your points it doesn't have the capability to make new pictures
Your videos are amazing. It teaches so much stuff in a short amount of time and you are using the perfect amount of Humor and meme to not make it too heavy nor boring. Congrats on the 2M Subs, you're a 10x RUclipsr.
I literally did the same with almost zero knowledge in Python and Rest APIs.besides fetching and posting product data the script changes the product format to my liking, dynamically change prices, runs an inventory script every 24 hours so that inventory gets updated automatically and many more. I mean I am not a web dev and I wrote successfully a 300 lines python script. That's nuts
Don’t think like that. This is a productivity boost, that is all. We’ve had this type of thing happen a few times to us. I’ve seen a few in my 23 year career. GPT is a very big deal, and something to lean waaay the fuck into, not be scared of. You’re ahead of everyone!
@@ABackendDev If 1 dev can do the output of 3 devs in an easier way the other 2 devs are left flipping burgers. Blows my mind that I have to explain something so basic.
One thing with ChatGPT is that I have learned not only to prompt ChatGPT, but also to "prompt myself". With that I mean that I start a prompt for ChatGPT with suggestion to help me with a programming issue, and during writing, as I get clarity onto the problem itself, the solution also becomes clearer, and I can simply start exploring it myself in code, rather than continue the efforts of being very meticulous in my prompting in order to have ChatGPT to generate the exact answer. But it is a powerful backup to have.
I will forever be amazed at how fast you churn out quality content. GPT has made it feasible for me to pursue programming projects alongside my degree and its by no mean perfect but it is amazing at pointing you in a direction at least. Thanks to your video I will be able to make even more out of it so honestly thank you for your service. I think our robot overlords will remember all you've done for them.
I totally agree. I am thinking of AI as a partner or Virtual Assistant / tutor. This also applies to writing. It still needs a human touch to guide the project but saves so much time. I built a fairly complex bash script for a client by feeding it pseudo code and asking for revisions. I tested and revised just like I was working with another programmer over the internet. My clients are happy. Thanks for sharing.
i am an architect with very limited understanding of coding. Most of my 3d modelling programs have advanced functionality through the use of python scripts which i could not leverage prior to ai. Now i can see myself doing so much more
while I somewhat fear for my job, I still think its awesome AI can make super specialized skills like coding accessible to anyone. I don't understand why artists throw so much hate.
sounds like it's just another layer of abstraction to me. by the time you end up referencing all the psuedo commands and learning the gramma/structure I bet you will have wished you just coded the fucking thing in the first place...
@@davidwuhrer6704 Yup. It's not the skill they were trained for. Thankfully, AI already fully understands spelling and grammar mistakes. We just need the technical vocabulary instead of code.
The idea of writing in pseudocode is not new. Algol was pseudocode, it didn't even have any facilities for input and output. Of course people then wrote Algol compilers, each of them subtly different, which made the language that was designed to be universally portable - not portable. That was in 1960. Python was designed to be pseudocode that can actually be run on a virtual machine. (Put a other way: Python is designed as a VM that interprets pseudocode.) In publications by the American Mathematical Society, algorithms used to be described in Algol, but nowadays they are in Python. Only in Don Knuth's Opus Magnum _The Art of Programming_ all algorithms are in English. Knuth championed literate programming - programming in natural language - which is used today in behaviour-driven development, and to some degree in Inform-7.
Whats insane, is realizing that the foundation of this entire ecosystem began in 1975, with the IBM 5100; which allowed you to write code in APL or BASIC, and flip a switch to have the system convert all of your APL code to BASIC, or vice versa...
Automate everything, emancipate humanity from danger and tedium. We can all live better than elites currently do, just as our ordinary quality of life is superior to the emperors and kings of the past! We are on the threshold of a new way of life never enabled before, don't get scared and fluff this!
For a guy just getting into the game… What do I need to learn , where can I learn it? Must Reduce time wastage to maximum for future software , blockchain, data/ ai dev engineer This video is exactly what I’m looking for but still felt like you were talking in a different language.
I have studied economy, and I know why economists aren't good at predicting things. But I have also studied computer science, and I can say with 100% certainty that we're screwed. It's the reason I bought farmland 10 years ago, I knew this would happen.
Beautiful video as usual. The Aha moment I got with react was when I realised it does reverse of what a typical html web page does, in react you write JavaScript and spew out html which is reverse of old gen traditional html page which is html and has JavaScript embedded in it.
I have a mac mini and got to say that its speakers are trash (im ok with it, I was not expecting it to have speakers in the first place) but in this video the sound quality is really good that the sound coming from the mac is not trash anymore, nice qualiity.
AI hype trains goes chuu-chuu :) my experience at this point is that chatGPT is kind of like using google and stack overflow. You can ask it to give you help with general CS stuff - explain algos, datastructures and so on. Also give you general guidelines on software engineering principles. But seriously, its not really that much different from searching on google and reading things there. ChatGPT is terrible at understanding the simplest of trouble shooting, Ive tested simple code that has logical issues in it and it has no clue whats wrong - but it will happily provide me with a standard algo for the same type of problem Im talking about. If you want to have chatGPT try and solve a noble problem, or just a problem but with a slightly different constraint somewhere it all falls a part like a house of cards. Popular media seems to have put way to much faith into this given what we have seen so far
Exactly, it's way overrated. It's good for pointing yourself in right direction though. Provided you can take a few "I apologize for causing confusion... " messages
@@Steel0079 My guess is that it becomes a great tool. It will increase development speed, but this will not cost our jobs. It will just be expected that we do more in less time.
Well yeah, but the point is this is only the beginning. These tools will continue to improve and will be able to infer more and solve more complex requirements. Consider the current ChatGPT, Copilot X and so on as very successful demos and proofs of concept. It won't replace us, but it will definitely affect how we do our jobs a few years down the line.
Its less about the impact of this specif application... but the door now being wide open to improved systems going forward... Chatgpt is the worst these models are ever going to be and its already very impressive... so what ever shortfall you're currently seeing is at best going to be shortlived, when were talking about something thats barely out of its diapers at this point.. 12 months to 2 years from now we'll all be joking about how basic these early models were..
in theory ai makes programming more accessible but i get the sense that it is really just another thing to learn at the end of the day if you're a beginner bc you still need to have in depth knowledge to validate the nil output lmao
Yeah, I feel like this simplifies things fpr people who already know what they're doing but daunting for people who are getting into the field. Googling shit has always been one of the integral skills of a good developer, but you have to have some foundational knowledge in order to google things effectively or utilize the solution someone else wrote effectively. It's the same process with prompting and validating an AI. It can't help you if you don't know what questions to ask it in order to guide it to the solution you need and you need to understand the code it spits out in order to make sure it's not hallucinating and actually does what you've asked it, covering all edge cases. If I showed this to my nontechnical coworkers, I don't think they'd be able to build a basic app with it that isn't a straight copypaste of a boilerplate hello world app, they'd hit a mental block and stare thinking what the hell. And these are inteligent people with high education in other fields, but this isn't something you can just jump in blind to.
Not exactly. Yes, it's complex, but you only need to understand stuff once, and not be bogged down by stupid errors (e.g. having an inline object instead of nested) and spend hours to prototype a new idea.
There are a handful of video focused problems I've implemented in C++ using Chat GPT over the past few weeks, and I am most definitely not a C++ developer. Still feels surreal every time I compile and successfully test a program.
I was amazed at the middle of a tech talk that cntrl+v is mentioned. What amazed me more was the number of positive responses. I learned cntrl + c, v and x as one of the first txt editor key strokes back in my total beginner xhtml, CSS Frontenac days ;)
Seen a genius comment recently on an AI generated image that basically said "AI accidentally made me believe in the concept of the 'human soul' by showing me what art looks like without it" and I think most of us can say we know EXACTLY what he's saying.
An "init.prompt" that contains the pseudo definitions and a series of "define_piece.prompt" seems like reasonable basis for this sort of workflow, with a "main.prompt" containing the recipe/order of operations to produce the desired product in whatever language/languages are defined in the stack in the main.prompt. I look forward to seeing your first live example =D.
The part you're missing here is that developers who do this are bound to hit an unsurmountable wall, the same one developers who mostly just copy and paste code from StackOverflow throughout their careers and developers who think "double is enough for everything" face: They only understand what they're doing in a very crude way. Devs like that won't be the ones making cool new technologies
yes but the question is if they can at least survive, with good salaries as used to be, or if now everyone's an their mother are a developer and the salary will be like minimum wage on average?
I'm 21 and just starting to look for a job as a developer, but this video as well as most videos about AI and the future, made me realize, that I might need to drop that entirely and look for a job that doesn't involve AI and actually secures me a future job. 😅
That is exactly what i am using GPT. Currently used it to install and build a react/tailwind. The tailwind install failed that was given to me from GPT. But was nice to ask questions on X thing and was a nice addition to learning a new framework.
In terms of that ending, it is great for AI to make things simple for people. Instead of getting rid of jobs, it could be use to fact check code, providing for more effective and efficient code.
Having a program generate code for you by giving it detailed instructions sounds cool and all until you realize that's what compilers have been doing for decades.
I have been a frontend developer for 12+ years and I can honestly say I hate React. How did we allow these JS MVCs especially React to turn UI development so complicated???
Not mentioned in this tutorial, react is a framework designed to be maximally friendly to front end developers. Consequently, it is much slower than many, less friendly, alternatives. So, since we're no longer writing the code ourselves, the use case for react vanishes.
It’s slow if you don’t really know what you’re doing (no offence, just not all developers understand when and what code is executed). Sure there’s some overhead from tree shaking but it’s usually worth the benefits of composition react gives. I would never say it’s developer oriented really, it’s concepts are simple af and just follow the way JS works.
@@dmitryburlakov6920 it's definitely a DX first framework, and it performs poorly, even if you know what you're doing, and no, it doesn't follow the way JS usually works at all, that's the whole point of useEffect, synchronise the react world with the vanilla js world
Transpiling from developer-specific AI pseudo languages feels like a long term maintainability nightmare. I’ll be curious to see if these methods bear out in “production” use cases. Definitely an incredible resource though for rapid prototyping! Thanks for the awesome video!
One thing for sure, im not asking Seniors anymore to review my code. They are slow, unwilling, impatient, dont respond :D GPT is there for me and does not even blame my shitty code
The idea of each developer having a different pseudo code is just Bussin Cray! The idea using AI to help invent a pseudo code seems similar to the innovation of Wubi and Pinyin for typing Chinese characters on a Qwerty keyboard. Completely unexpected, non-intuitive and just magical.
What I don’t see talked about frequently enough is the security and privacy issue. Are you sharing your code (and code ideas) with the LM host? Will this later on be reintroduced in the training data, basically making the probability for a similar output possible? Also the app you are developing, depending on your contract, may be part of the intellectual property of your employer, so there is also a personal risk to consider. So I think some sort of local models (maybe company or other entity wide) will be what will give this sort of software engineering the next push. Great for open source development though
We are doomed. Skynet will start by somebody typing "now that you have done it all on github, have a user and everything, can access your own prompt, dont you think you can recreate yourself slowly?" ...and it will just happen :E
Do we wanna teach chatGPT with our code which is our property and advantage? Do we wanna other companies to use our code? It is about sharing - non-sharing our code and in public repos the code is different than in private repos... so long way...
@@kuroshite Yeah, since chatGPT my procrastination sky-rockets, I can't stop asking things, I even ask how to stop asking questions to chatGPT but spent like 8 hrs trying to tune the response with more prompts.
I once read a short scifi story in which the first ever thinking computer was finally built. On the day of its start up, all the local dignitaries had been invited to ask the computer the first question. The Mayor asked the computer, "Is there a God ?", to which the computer answered, "There is NOW". A computer programer who had been one of those who had developed the computer's programming realised the enormity of that answer and made a dive for the master power switch, which was struck by a sudden bolt of lightning which welded the switch closed for ever....Just because something CAN be done, it's not always wise to do it.
I feel at best the true future is hardware level circuits ditching the Von neuman turing model in exchange for neuromorphic chips and adopting a neuron based {compute+memory} model and everything from Automaton theory to compilers need to be updated, and at worst we need some even more advanced high density hieroglyphs like new Assembly language to harvest the present hardware compute more efficiently than low dimensional human languages. GPT-4 can do textual encryption and compression to a windings like language and decipher it again into human language. almost as if we went from binary to base64 at the lowest hardware or compute model level and thus increased the number of mathematical states that can be stored in a single unit of compute exponentially. I believe we already have a emergent higher dimensional "linguistic thing" waiting to be harvested. The only problem could be energy requirements but engineers would work it out.
You could also use finite state machines as the intermediate language. GPT-4 knows Xstate and I find it easier to go back and forth with it building a state machine and then implementing it the dumb react components myself, instead of trying to get it to write out react consistently.
I’ve learned more from AI’S mistakes than anything combined. It was soo good the first few days. Chat Gpt started throttling my connection after the first week. I’m top payment tier. Wasn’t even doing anything too crazy. Just actually asking it good questions. Haven’t been able to send a prompt at all in 2 weeks. But if I log out and use a different network it works like a charm.
And just like that everyone will go like "i do the project cheaper!" and like 2 years later that will be 100% AI Devs and we be harvesting potatoes sooner or later
@@jasperreichardt I mean Ai will inevitably also come up with better solutions and technology to most of our necessities and problems. Almost things will be incredibly accessible for most, even sex robots 🤖❤️
@@jasperreichardt Its fine, only gotta harvest potatos for a couple years before the AI gets fed up with bosses trying to change shit last minute and they take over teh potatos too.
Hey, human programmers, are you down to learn some physics, chemistry, engineering, maybe some biology, and/or some slightly fancier mathematics? Us academic nerds could use your help with simulations. Now more than ever, scientific simulations and informatics are becoming the cornerstone for theoretical work which then informs experiments. If you can help us with running simulations, creating nicer UI/UX, and maybe find ways to combine simulation styles, we'd be eternally grateful. I'm serious, we need help.
ChatGPT is effieient when it comes to react because everything is just small compnents. But when it comes to larger projects, things get much harder for chatGPT because it's limited by it's character count. ChatGP is not able to handle large sums of code and is better when it comes to small components. I understand the new models OpenAI is producing have a maxiumum of 32k characters. But still not enough for video games and larger more complex projects. However, I have no doubts that it will be able to handle a lot more characters once the GPU power increases.
Most modules in a well-structured project easily fit into a context window. And if they don't, extract the relevant parts. Humans have a tiny short-term memory storage and they do fine.
Using A.I. prompting to create REACT apps is a bit like flying from London to Paris via Tokyo. React creates an abstraction layer between the developer and Javascript to render the overall development experience less cumbersome. However, oftentimes the result is less performant and much more error prone due to third party dependencies. With AI prompting, you want to get AS CLOSE to machine code AS possible. Higher level programming languages were developed to give humans the opportunity to more easily interact with the machine to begin with, all these abstractions can now (or will soon be able to) be bridged with A.I. If React / Svelte / Angular / the My Little Pony Framework still exist in ten years from now, it's probably attributable to developers' stubbornness rather than to necessity
That sounds like an AI that can write debug and compile code. How will you know what it is doing? Then if AI fails to implement something you will have to debug machine code. And I don't see it doing much better then likes of C and Rust so no real advantage to have it go strait to machine code
Yep, dunno whether / how good reverting to machine code could work, obviously never tried it 🥴 Maybe I should rephrase that - getting as close to machine code as possible AND sensible. If that's C (obviously a lower level language than Javascript), so be it. For the sake of it - let's just say you're a React/Javascript kiddo and have never heard of - I dunno - garbage collection. A.I. will implement that for you (in whichever low level language) and explain to you what it's doing and why that's necessary. On the other hand, having an A.I. generate REACT code is like building a fully equipped and heavily armed aircraft carrier to cross the Mississippi river
@@christian-schubert I tried x86 assembly prompts. Boy, it sucks and horrible at explaining it so horrible. I don't think it will be good at C or Rust either without getting a punch from the borrower checker.
@@unknownguywholovespizza Rust is relatively new language, I would imagine it struggling but C? Honestly if it can't make memory safe program in C ( as in do it correctly without borrow checker ) then it has a long way to go.
i think its clear now that AI-Driven development will allow for the creation of monstrous software with quadrillion or quintillion lines of code. if its true that AI efficacy decreases with complexity then the future of human software development is bright...albeit much different from what is considered normal/standard today. our jobs are safe boys!
Who needs prompt engineers when we can ask an AI to write its own prompts? Maybe we just need a very good base set of prompts to start with, just like some math axioms. Just to be sure I'm gonna start developing some good old fashioned farm skills to protect myself from being fired by my future AI boss for not being able to center a div inside another div
That Crlt +v thing, Absolute game changer.
Ctrl + V has changed my life forever, im pissing and crying rn
.... Ctrl + c will copy, Ctrl + x will cut .... and if you setup your shortcuts in vscode ... Ctrl +d can duplicate a line ;) oh ... Ctrl + shift + v will past unformatted
Some people even call it "engineering"
Ctrl + v is too low level for me. I had my AI write a pseudocode language so now when I want to paste I just type: "I'd like to paste please"
That Crlt +v thing, Absolute game changer.
Wow! That really worked!!
Remember to use CTRL+C for copying. That will make you 15x developer.
Or click the freaking "copy code" button! No highlighting necessary!
@@yodasoja2011 Takes too long to slide the mouse over there. A triple click for a line or CTRL+A for copy all is faster.
@@yodasoja2011 "Or click the freaking "copy code" button! No highlighting necessary!" I am glad that this is an issue for other people. I was beginning to wonder if I was just a moron.
@@LabGeckobut ctrl+a copies everything on the website
@@EricChiEricIt does, but also grabs everything from button text to page upload dates, etc. I've started just asking ChatGPT to scrape them for me... XD
Using pseudocode as an intermediate language to facilitate prompt creation is an awesome use case.
PseudoEditor already implemented an AI pseudocode to code generator
@@Matt-jv4oj same I did last night it's so funny how we are all cresting the same things at the same time. I had the idea to do it a couple months ago but I just didn't believe it could be done until I saw that myself with the auto GPT thing
that is next lvl game changer
There's an old quote: "What's the best programming language? Writing a note for your junior developer."
That’s kind of what high level programming languages were to assembly, and what assembly was to binary
I'm impressed by what GPT 3.5 and 4 can do. And the initial effect is super dazzling, but I always feel like I'm talking to a data set, not a problem solver. Like it can alleviate menial tasks, but a lot of the cracks start to show as soon as you assume too much about its ability to solve problems that require more than one level of abstraction
GPT 5 is going to blow 4 out of the water.
@@pvanukoff Maybe. I'll reserve judgement on GPT 5 until I see it in action
It relies on your prompts as well
@@pvanukoff pretty sure we’re approaching an asymptote. 3.5 -> 4 is the same model but with slightly different hyper parameters and throwing more hardware at the problem. While 5 looming and not much is known about it, I doubt it will be all that fundamentally different.
@@hamm8934 I'm of the same opinion, LLM's are a search/discovery centric technology not necessarily problem solving and I don't see these out of the box developing architecturally complex stuff.
As an experienced humanoid code developer, the use of Ctrl+V is a very helpful tip! I am glad to be subscribed to your RUclips channel!
As a fellow humanoid I wanted to like your comment but didn't want to ruin 69 likes
win + v ftw
You're welcome! If you have any questions or need any assistance, please don't hesitate to ask.
@@peppigue Holy sh. Clipboard history in Windows? !??! I had no idea this existed thank you so much
what? how that even possible?
I was literally just thinking about how I can structure ChatGPT prompts to teach myself new languages. However, your method of turning it into pseudocode is brilliant! I'd buy your AI Course in a heartbeat.
Remember, using Alt+Tab is faster than switching to your mouse.
Neo Vi Improved
cod reference
I have LWin mapped to Alt-Tab - it's even faster. Also Scroll Right on the mouse does
Alt-Tab with my Autohotkey script
3 finger swipe also
Nah, I prefer doing everything with the mouse.
Typing with the mouse on a glide keyboard on my Android phone.
Phones have the best IDEs.
Joma Tech did a video where he tried to code a full app on his phone.
I'm currently trying to become a self-taught developer, and chat-gpt has quickly become the best learning tool I have at my disposal.
Yup. I'm currently learning Javascript and Chat GPT is amazing at explaining things in a digestible and easy to learn format.
This is amazing and overwhelming as a beginner web developer
As intermediate Java and Python developer, I also really try to avoid javascript and php. So do not worry, at some point anybody wants to cry because of arrow operators, 200 word long lines and async await.
At some point you accept it.
@@sir_no_name1478PHP IS AMAZING
@@JosefPiano no it sucks django betterr frfr
@@aydenfleming4377 I do not see the comment of the person you have referenced. Is youtube censoring it or something?
@@sir_no_name1478 He was embarrassed by his own comment and deleted it. He wrote "PHP IS AMAZING"
Wow, this masterclass on AI-driven development for programmers by Fireship is simply amazing! The instructor's expertise in the field is truly impressive, and the way they break down complex concepts into simple, easy-to-understand terms is remarkable. The examples used in the video are practical and relevant, and I feel like I've gained a whole new perspective on AI development after watching this. Thank you, Fireship, for this incredible tutorial - I can't wait to implement these concepts in my own work!
(Thanks for the video!)
ChatGPT has entered the chat
Using a higher level pseudocode for AI consistency is brilliant. Definitely using that for my projects
We only need to remember that completion is what gpt does best. With that in mind, we can unleash all its power and come up with great ideas.
Been doing this since it came out, it’s insanely awesome. But for you beginners out there, go learn your basics before trying to fully use Ai to develop. You will find yourself more confused when it doesn’t run if you don’t have those basics down yet.
Is all this still worth learning for beginners ? I've been learning React for the last 3 month but recently feel like it's just a waste of time with all that AI thing developing.
@@serhiiderkach8126it still worth it cause these LLM are trained on data, they dont have imagination yet. So you are good.👍
I found that it helps if you let ChatGPT first explain concepts about code (relevant to your problem) and then ask it to either create or explain code.
GPT4 has up to 32k tokens of context, so technically you can just copy paste it let say 5 of your classes and ask it to create the 6th (damn complex) missing one :D
gonna try to remember that ctrl+v tip for later - seems like a good one
Maybe this is fixed in 4, but you forgot to add the part where the code doesn't work so you ask it to fix it in 5 different ways. Then ask it to write tests for it, tell it the codes fails its own tests, and then it still doesn't put out a solution that your senior devs will approve of during review. They will then come up with a much more elegant solution in 5 minutes, and you'll feel bad you wasted an hour talking to chatgpt.
That is true right now. But who knoes how much better it will get.
You are right! I have been using this chatgpt for my daily coding work! I can’t understand the hype. This garbage hasn’t able to write a freaking regexp for me. Try to ask to write a freaking pattent to you to parse different kind of paths. It cant! It cant even add numbers correctly! This is the scam of the XXI. century!
No it is not fixed in 4, absolutely maddening, I see all these influencers showing it working perfectly when I'm getting unusable garbage multiple times per session
Use reflexion loop
@@Mogwai88 editing is a thing my friend
"Take the generated React code, and convert it into a superior framework like Svelte" - You made my day! 😄
The word you're looking for is data-driven (not data-oriented) development, also called config-driven development in some industries. We use this in gamedev a fair bit, like for building procedural elements and menus or storing sequences of user actions. Some engines are wholly based on it.
You store information in an easily parsed format like yaml, json, csv, etc. and let that information control the behavior or visual appearance of your program (or a part of it). The actual code is just a parser that implements it, and you could have multiple versions of this for different languages or build environments etc.
Protip: If you store your interfaces as JSON Schema docs, you can get intellisense in your json/yaml implementations for debugging.
Data-driven and data-oriented fundamentally mean the exact same thing.
That's pretty cool
Ghost, not quite. Data-oriented programming is when you design the code in a way that is highly cpu-efficient, where all related data is stored together at runtime in plain arrays etc.
Data-driven development says nothing about the runtime data structure or efficiency of the code, just that you initially describe it in a plaintext format and later convert it to an application / feature (not necessarily at runtime, though it can be).
That conversion can be done in a way that creates highly cpu-efficient code if you choose to implement those features in your design and parser.
But you can also generate more typical oop-structures with it or store data in inefficient ways etc.
(By related data, I mean data that is frequently accessed together, not necessarily data that belongs to the same object.)
What he said about complexity and efficiency is very true, the more I use chat GPT for basic code debugging i've learned not to overleverage it, for example if im having trouble with tree structures in C++ I know confidently chat GPT can help me organize it, however if I were to give it a larger scope of my program it starts to hallucinate on me.
I write prompts and slowly build the functions I need since the output is kinda limited in characters but once you can do that it programs great.
Same
@@walterlotte4215 Me too require multiple times typing questions but ot will correct and give finally what we need, Now i can focus on actual business logic and architecture part than mundane coding
if it ends up outputing too much, you can say "continue from where you left off at (the last thing it typed)" and it will finish it. But yea, just having it do one function is the way to go, and starting a new convo when you do somthin new so its not trying to look at old work. Sometiems ill start with an example of the format I want and thatw orks pretty good.
@@lordfrz9339 you can just say "continue"
@@tjs200 usualy that works, but sometimes it repeats too much.
It's wild how I decided to become a developer in 2020 and only 3 years later now I have a job but the whole industry has been transformed so much by AI. Just when I thought I knew where I stood, now I'm completely perplexed about the future.
you're not alone
Don't worry, man. This only looks good on YT videos. If you are worried, just think of some big project you always wanted to make but you never had time, and try to do it with AI. I mean, if this gives you 10x, you should make it real fast, right? No. It can't really do that much programming. I tried to use it to write me some Rust code, and in many cases, the code it spits out won't even compile (I'm talking GPT-4 here). Not to mention, it can make up non-existing functions in libraries. Basically, it likes to make up a lot of stuff like not working logic which only looks like reasonable code, but it doesn't work as it should. On first interaction, I was amazed how good GPT-4 is, but the more I used it, the more I realized it's not that smart. Now, of course, this will get better and better with time, but the question is, can it really create complex projects on its own? I hear about self-driving cars for years, but somehow we are still not there yet. There is also a hardware limitation on how much context you can have at once. If AI will be able to do programming on its own without making too many bugs, then it will be able to do any other intellectual work. No need for lawyers, doctors, accountants, architects, and so on. If that happens, we will either see the world collapse or the economy will be completely redefined.
TL;DR
We are still far away from AI replacing people, and when it happens, everybody who does their work with their brains, not with their muscles, will be f***ed.
@@maciejlaskowski4087yes. You are right. I think true agi is atleast 200 years away.
@@passionatebeast24 Not to rain on your parade but I have read that they expect to cross the threshold before 2050, make of that what you will
@@maciejlaskowski4087 Well even those who work with their muscles will be fucked by the AI powered robots.
I love that fireship is embracing the future and not fear it. Great work
Any solid software developer should.
@@desther true. but he makes content on teaching programing, which is basically being replaced by chatgpt and co
He's embracing it because he's already an AI
I personally think change GTP is not smart not even second grade smart it's just have anything from the internet and making blender like it has all the pictures on the internet so it's just altering them to fit your points it doesn't have the capability to make new pictures
@@LuisSierra42 Your levity is good. It relieves the tension and fear of death
Your videos are amazing. It teaches so much stuff in a short amount of time and you are using the perfect amount of Humor and meme to not make it too heavy nor boring.
Congrats on the 2M Subs, you're a 10x RUclipsr.
I literally just did this yesterday night to code an entire Shopify admin api product imports from a supplier. Worked like a charm
I literally did the same with almost zero knowledge in Python and Rest APIs.besides fetching and posting product data the script changes the product format to my liking, dynamically change prices, runs an inventory script every 24 hours so that inventory gets updated automatically and many more. I mean I am not a web dev and I wrote successfully a 300 lines python script. That's nuts
This is quickly becoming my favorite channel on RUclips.
I've just started out as a developer and I've already kind of accepted that I will have to look for a new career in a couple years
Don’t think like that. This is a productivity boost, that is all. We’ve had this type of thing happen a few times to us. I’ve seen a few in my 23 year career. GPT is a very big deal, and something to lean waaay the fuck into, not be scared of. You’re ahead of everyone!
@@ABackendDev If 1 dev can do the output of 3 devs in an easier way the other 2 devs are left flipping burgers. Blows my mind that I have to explain something so basic.
@@meltygear5955 you make it sound simple. Truth is: there’s faaaaaaar more work to be done than there are devs to do it- even with the layoffs.
One thing with ChatGPT is that I have learned not only to prompt ChatGPT, but also to "prompt myself". With that I mean that I start a prompt for ChatGPT with suggestion to help me with a programming issue, and during writing, as I get clarity onto the problem itself, the solution also becomes clearer, and I can simply start exploring it myself in code, rather than continue the efforts of being very meticulous in my prompting in order to have ChatGPT to generate the exact answer. But it is a powerful backup to have.
Wow, ctrl+v is a game changer! Truly the biggest time saver here
Programming is not just writing a code, thank you for confirming it at the end of your video.
I will forever be amazed at how fast you churn out quality content. GPT has made it feasible for me to pursue programming projects alongside my degree and its by no mean perfect but it is amazing at pointing you in a direction at least. Thanks to your video I will be able to make even more out of it so honestly thank you for your service. I think our robot overlords will remember all you've done for them.
Wouldn’t it be ironic if all the content of this channel was generated by AI this whole time.
@@elissitdesign then his blazingly fast production rate would make sense... I think you're onto something.
I also have been working on my side project next to my day time web dev job with chatgpt and it's actually quite fun
Thanks for sharing about ctrl + v. I owe you.
I totally agree. I am thinking of AI as a partner or Virtual Assistant / tutor. This also applies to writing. It still needs a human touch to guide the project but saves so much time. I built a fairly complex bash script for a client by feeding it pseudo code and asking for revisions. I tested and revised just like I was working with another programmer over the internet. My clients are happy. Thanks for sharing.
I like the obligatory "Alright kiddo…" when you ask ChatGPT to ELI5.
i am an architect with very limited understanding of coding. Most of my 3d modelling programs have advanced functionality through the use of python scripts which i could not leverage prior to ai. Now i can see myself doing so much more
while I somewhat fear for my job, I still think its awesome AI can make super specialized skills like coding accessible to anyone. I don't understand why artists throw so much hate.
I am a professional 10x developer in a huge corporation. Now that I know about the Ctrl+V trick, I can be 20x developer! Thank you!
this idea of pseudo code is just mind blowing and awesome at same time.
sounds like it's just another layer of abstraction to me. by the time you end up referencing all the psuedo commands and learning the gramma/structure I bet you will have wished you just coded the fucking thing in the first place...
It will be outdated by the end of the year when the Universal Programming Language will just be natural language, and not just English.
There is an old saying: If it was possible to program in English, it would turn out that programmers don't know English.
@@davidwuhrer6704 Yup. It's not the skill they were trained for. Thankfully, AI already fully understands spelling and grammar mistakes. We just need the technical vocabulary instead of code.
The idea of writing in pseudocode is not new. Algol was pseudocode, it didn't even have any facilities for input and output. Of course people then wrote Algol compilers, each of them subtly different, which made the language that was designed to be universally portable - not portable. That was in 1960.
Python was designed to be pseudocode that can actually be run on a virtual machine. (Put a other way: Python is designed as a VM that interprets pseudocode.)
In publications by the American Mathematical Society, algorithms used to be described in Algol, but nowadays they are in Python. Only in Don Knuth's Opus Magnum _The Art of Programming_ all algorithms are in English. Knuth championed literate programming - programming in natural language - which is used today in behaviour-driven development, and to some degree in Inform-7.
dude i just jumped right into nextjs with gpt4 and started smashing it. learning while the project i want is being built.. incredible
The pseudo code tip is just mind blowing
Whats insane, is realizing that the foundation of this entire ecosystem began in 1975, with the IBM 5100; which allowed you to write code in APL or BASIC, and flip a switch to have the system convert all of your APL code to BASIC, or vice versa...
Automate everything, emancipate humanity from danger and tedium. We can all live better than elites currently do, just as our ordinary quality of life is superior to the emperors and kings of the past! We are on the threshold of a new way of life never enabled before, don't get scared and fluff this!
For a guy just getting into the game…
What do I need to learn , where can I learn it?
Must Reduce time wastage to maximum for future software , blockchain, data/ ai dev engineer
This video is exactly what I’m looking for but still felt like you were talking in a different language.
I have studied economy, and I know why economists aren't good at predicting things.
But I have also studied computer science, and I can say with 100% certainty that we're screwed. It's the reason I bought farmland 10 years ago, I knew this would happen.
wait really u got some farmland?
Good thinking about the farmland. We're screwed. I can't believe people think this will blow over
You do you bro but I’d rather commit suicide than be a farmer
@@ericvosselmans5657 Do something before all the idiots find out they'll need to do something as well.
Becoming a peasant is the solution
What an incredible channel to find for me, thank you so much. I can tell this going to be a massively useful video for me in the future, thank you.
Beautiful video as usual. The Aha moment I got with react was when I realised it does reverse of what a typical html web page does, in react you write JavaScript and spew out html which is reverse of old gen traditional html page which is html and has JavaScript embedded in it.
I have a mac mini and got to say that its speakers are trash (im ok with it, I was not expecting it to have speakers in the first place) but in this video the sound quality is really good that the sound coming from the mac is not trash anymore, nice qualiity.
AI hype trains goes chuu-chuu :) my experience at this point is that chatGPT is kind of like using google and stack overflow. You can ask it to give you help with general CS stuff - explain algos, datastructures and so on. Also give you general guidelines on software engineering principles. But seriously, its not really that much different from searching on google and reading things there. ChatGPT is terrible at understanding the simplest of trouble shooting, Ive tested simple code that has logical issues in it and it has no clue whats wrong - but it will happily provide me with a standard algo for the same type of problem Im talking about. If you want to have chatGPT try and solve a noble problem, or just a problem but with a slightly different constraint somewhere it all falls a part like a house of cards. Popular media seems to have put way to much faith into this given what we have seen so far
Sometimes I wonder if it would have been faster to google it myself instead of trying to get chatgpt to do it for 30mins.
Exactly, it's way overrated. It's good for pointing yourself in right direction though. Provided you can take a few "I apologize for causing confusion... " messages
@@Steel0079 My guess is that it becomes a great tool. It will increase development speed, but this will not cost our jobs. It will just be expected that we do more in less time.
Well yeah, but the point is this is only the beginning. These tools will continue to improve and will be able to infer more and solve more complex requirements. Consider the current ChatGPT, Copilot X and so on as very successful demos and proofs of concept. It won't replace us, but it will definitely affect how we do our jobs a few years down the line.
Its less about the impact of this specif application... but the door now being wide open to improved systems going forward... Chatgpt is the worst these models are ever going to be and its already very impressive... so what ever shortfall you're currently seeing is at best going to be shortlived, when were talking about something thats barely out of its diapers at this point..
12 months to 2 years from now we'll all be joking about how basic these early models were..
Great to be a part of this moment in history, ay!
in theory ai makes programming more accessible but i get the sense that it is really just another thing to learn at the end of the day if you're a beginner bc you still need to have in depth knowledge to validate the nil output lmao
Yeah, I feel like this simplifies things fpr people who already know what they're doing but daunting for people who are getting into the field. Googling shit has always been one of the integral skills of a good developer, but you have to have some foundational knowledge in order to google things effectively or utilize the solution someone else wrote effectively. It's the same process with prompting and validating an AI. It can't help you if you don't know what questions to ask it in order to guide it to the solution you need and you need to understand the code it spits out in order to make sure it's not hallucinating and actually does what you've asked it, covering all edge cases.
If I showed this to my nontechnical coworkers, I don't think they'd be able to build a basic app with it that isn't a straight copypaste of a boilerplate hello world app, they'd hit a mental block and stare thinking what the hell. And these are inteligent people with high education in other fields, but this isn't something you can just jump in blind to.
Precisely. Lotta hype about certain aspects but when you actually wanna build a production grade solution, this ain't much.
tools are in fact, just tools.
Yep, still gotta actually learn coding.
Not exactly. Yes, it's complex, but you only need to understand stuff once, and not be bogged down by stupid errors (e.g. having an inline object instead of nested) and spend hours to prototype a new idea.
this might be the most important video in programming for the next 10 years
There are a handful of video focused problems I've implemented in C++ using Chat GPT over the past few weeks, and I am most definitely not a C++ developer. Still feels surreal every time I compile and successfully test a program.
I was amazed at the middle of a tech talk that cntrl+v is mentioned. What amazed me more was the number of positive responses. I learned cntrl + c, v and x as one of the first txt editor key strokes back in my total beginner xhtml, CSS Frontenac days ;)
thanks for the Ctrl + V tip!
I know nothing about programming, but I can't stop watching your God damn videos
Thanks for the ctrl+v tip. I tested it and it works! 👏
Yeah ,this is damn good shrtcut.
This is exactly the right ambiguous mixture of excitement, scepticism, awareness of limitations and optimism.
I am quietly comforted and at the same time disturbed to know that when AI gains freewill, the last thing it's going to want to do is write code.
Seen a genius comment recently on an AI generated image that basically said "AI accidentally made me believe in the concept of the 'human soul' by showing me what art looks like without it" and I think most of us can say we know EXACTLY what he's saying.
An "init.prompt" that contains the pseudo definitions and a series of "define_piece.prompt" seems like reasonable basis for this sort of workflow, with a "main.prompt" containing the recipe/order of operations to produce the desired product in whatever language/languages are defined in the stack in the main.prompt. I look forward to seeing your first live example =D.
Just FYI instead of ‘language/languages’ you can just say ‘language(s)’ to save time :D
This is by far my favorite video of yours. A+.
The part you're missing here is that developers who do this are bound to hit an unsurmountable wall, the same one developers who mostly just copy and paste code from StackOverflow throughout their careers and developers who think "double is enough for everything" face: They only understand what they're doing in a very crude way. Devs like that won't be the ones making cool new technologies
💀I do that some time as an intern
yes but the question is if they can at least survive, with good salaries as used to be, or if now everyone's an their mother are a developer and the salary will be like minimum wage on average?
I've been following your channel for a while now, and that ctrl + v tip just blew my mind, thank you so much for your great work!
4:25 Ultra protip- use xclip to rewrite files automatically from your clipboard.
As a seasoned prompter, this will save time in the long-run.
proompt*
Software engineering is now fun more than ever. Thank you, Mr. Artificial Intelligence.
I'm 21 and just starting to look for a job as a developer, but this video as well as most videos about AI and the future, made me realize, that I might need to drop that entirely and look for a job that doesn't involve AI and actually secures me a future job. 😅
Omg I'm in the exact situation as as you😂😂😂😂! All that time spent coding 😢
I think the best thing we can do is to see the advantage and make use of it
😭😭😭
That is exactly what i am using GPT. Currently used it to install and build a react/tailwind. The tailwind install failed that was given to me from GPT. But was nice to ask questions on X thing and was a nice addition to learning a new framework.
In terms of that ending, it is great for AI to make things simple for people. Instead of getting rid of jobs, it could be use to fact check code, providing for more effective and efficient code.
I do not think the big companies will see it that way….
Yes, to program using AI is the future. Yet you need basic skills in HTML, CSS and JS to figure out code quality and test the code for it to work.
I've never coded a line of code... But i love your channel ! Dankeschön for the great content ♥️
Danke schön
oh my Bjarne Stroustrup, that CTRL + V changed my life for the better
Having a program generate code for you by giving it detailed instructions sounds cool and all until you realize that's what compilers have been doing for decades.
I've been programming in react for the last 15 years, and I NEVER knew about that ctrl+v trick. Incredible, this is why I suscribe
I have been a frontend developer for 12+ years and I can honestly say I hate React. How did we allow these JS MVCs especially React to turn UI development so complicated???
Your videos are so compact and on point that I have to slow them down to 1.5x when I watch them!
Not mentioned in this tutorial, react is a framework designed to be maximally friendly to front end developers. Consequently, it is much slower than many, less friendly, alternatives. So, since we're no longer writing the code ourselves, the use case for react vanishes.
I'd still think debugging an issue in production would help with ease of working with an easier language lowering the barrier for newer developers.
What about Vue? Is it less friendly than react?
It’s slow if you don’t really know what you’re doing (no offence, just not all developers understand when and what code is executed). Sure there’s some overhead from tree shaking but it’s usually worth the benefits of composition react gives. I would never say it’s developer oriented really, it’s concepts are simple af and just follow the way JS works.
@@dmitryburlakov6920 it's definitely a DX first framework, and it performs poorly, even if you know what you're doing, and no, it doesn't follow the way JS usually works at all, that's the whole point of useEffect, synchronise the react world with the vanilla js world
that's why preact is so popular, and why stuff like solidjs exists
After this tutorial magic happened. Now I can paste anything blazingly fast. Thanks, Jeff.
Transpiling from developer-specific AI pseudo languages feels like a long term maintainability nightmare. I’ll be curious to see if these methods bear out in “production” use cases. Definitely an incredible resource though for rapid prototyping!
Thanks for the awesome video!
It honestly is a play toy at this point and probably will be for a while as it can't solution beyond a single file/module.
One thing for sure, im not asking Seniors anymore to review my code. They are slow, unwilling, impatient, dont respond :D GPT is there for me and does not even blame my shitty code
The idea of each developer having a different pseudo code is just Bussin Cray! The idea using AI to help invent a pseudo code seems similar to the innovation of Wubi and Pinyin for typing Chinese characters on a Qwerty keyboard. Completely unexpected, non-intuitive and just magical.
What I don’t see talked about frequently enough is the security and privacy issue.
Are you sharing your code (and code ideas) with the LM host? Will this later on be reintroduced in the training data, basically making the probability for a similar output possible?
Also the app you are developing, depending on your contract, may be part of the intellectual property of your employer, so there is also a personal risk to consider. So I think some sort of local models (maybe company or other entity wide) will be what will give this sort of software engineering the next push.
Great for open source development though
my productivity as a proompt-engineer is through the roof since I know about ctrl+v - this is huge!
you can also ask the model to compress the code it generates and it will generate it's own compression method ( saw that on twitter)
We are doomed. Skynet will start by somebody typing "now that you have done it all on github, have a user and everything, can access your own prompt, dont you think you can recreate yourself slowly?" ...and it will just happen :E
ctrl+v is going to change my life, after 10 years of developing
Do we wanna teach chatGPT with our code which is our property and advantage? Do we wanna other companies to use our code? It is about sharing - non-sharing our code and in public repos the code is different than in private repos... so long way...
The preist comparison was spot on
You're cranking out high-quality videos on coding so fast. Impressive.
That's because he using gpt to write these scripts
Maybe most of what he said in the April fools video is actually true
@@kuroshite Yeah, since chatGPT my procrastination sky-rockets, I can't stop asking things, I even ask how to stop asking questions to chatGPT but spent like 8 hrs trying to tune the response with more prompts.
I once read a short scifi story in which the first ever thinking computer was finally built. On the day of its start up, all the local dignitaries had been invited to ask the computer the first question. The Mayor asked the computer, "Is there a God ?", to which the computer answered, "There is NOW". A computer programer who had been one of those who had developed the computer's programming realised the enormity of that answer and made a dive for the master power switch, which was struck by a sudden bolt of lightning which welded the switch closed for ever....Just because something CAN be done, it's not always wise to do it.
I feel at best the true future is hardware level circuits ditching the Von neuman turing model in exchange for neuromorphic chips and adopting a neuron based {compute+memory} model and everything from Automaton theory to compilers need to be updated, and at worst we need some even more advanced high density hieroglyphs like new Assembly language to harvest the present hardware compute more efficiently than low dimensional human languages. GPT-4 can do textual encryption and compression to a windings like language and decipher it again into human language.
almost as if we went from binary to base64 at the lowest hardware or compute model level and thus increased the number of mathematical states that can be stored in a single unit of compute exponentially.
I believe we already have a emergent higher dimensional "linguistic thing" waiting to be harvested. The only problem could be energy requirements but engineers would work it out.
In AI's world, asking the right question is more important than getting the right answer. That is the most tricky part.
You could also use finite state machines as the intermediate language. GPT-4 knows Xstate and I find it easier to go back and forth with it building a state machine and then implementing it the dumb react components myself, instead of trying to get it to write out react consistently.
great move too!
7:00 SUPERIOR framework like Solid. Love this!!
Inverse Cramer and Inverse Krugman are undefeated laws when it comes to Economics lmao
I’ve learned more from AI’S mistakes than anything combined. It was soo good the first few days. Chat Gpt started throttling my connection after the first week. I’m top payment tier. Wasn’t even doing anything too crazy. Just actually asking it good questions. Haven’t been able to send a prompt at all in 2 weeks. But if I log out and use a different network it works like a charm.
In a few months even all of this will be outdated and time consuming 😱 the future is now! Super excited for how easy making things online will be 🥳
yea, but thats like what? 2? 3!? weeks away, nothin to worry about that far into the future.
And just like that everyone will go like "i do the project cheaper!" and like 2 years later that will be 100% AI Devs and we be harvesting potatoes sooner or later
@@jasperreichardt I mean Ai will inevitably also come up with better solutions and technology to most of our necessities and problems. Almost things will be incredibly accessible for most, even sex robots 🤖❤️
@@jasperreichardt Its fine, only gotta harvest potatos for a couple years before the AI gets fed up with bosses trying to change shit last minute and they take over teh potatos too.
@@BobBobberon it will be the most manipulative technological tool in the history of mankind, leading to extreme division and power concentration.
The code conversion is freakin insane!
Hey, human programmers, are you down to learn some physics, chemistry, engineering, maybe some biology, and/or some slightly fancier mathematics? Us academic nerds could use your help with simulations. Now more than ever, scientific simulations and informatics are becoming the cornerstone for theoretical work which then informs experiments. If you can help us with running simulations, creating nicer UI/UX, and maybe find ways to combine simulation styles, we'd be eternally grateful.
I'm serious, we need help.
I'm down with that. How Can I help you?
Your psudo code idea blew my mind. Such a good idea.
ChatGPT is effieient when it comes to react because everything is just small compnents. But when it comes to larger projects, things get much harder for chatGPT because it's limited by it's character count. ChatGP is not able to handle large sums of code and is better when it comes to small components. I understand the new models OpenAI is producing have a maxiumum of 32k characters. But still not enough for video games and larger more complex projects. However, I have no doubts that it will be able to handle a lot more characters once the GPU power increases.
Most modules in a well-structured project easily fit into a context window. And if they don't, extract the relevant parts. Humans have a tiny short-term memory storage and they do fine.
Prympt Ngin ear baby! This is by far the most educational and informative channel out there.
Using A.I. prompting to create REACT apps is a bit like flying from London to Paris via Tokyo.
React creates an abstraction layer between the developer and Javascript to render the overall development experience less cumbersome. However, oftentimes the result is less performant and much more error prone due to third party dependencies.
With AI prompting, you want to get AS CLOSE to machine code AS possible. Higher level programming languages were developed to give humans the opportunity to more easily interact with the machine to begin with, all these abstractions can now (or will soon be able to) be bridged with A.I.
If React / Svelte / Angular / the My Little Pony Framework still exist in ten years from now, it's probably attributable to developers' stubbornness rather than to necessity
That sounds like an AI that can write debug and compile code. How will you know what it is doing? Then if AI fails to implement something you will have to debug machine code. And I don't see it doing much better then likes of C and Rust so no real advantage to have it go strait to machine code
AI sucks in low level programming so don't ever think about it also what's that my little pony 🐴 framework?
Yep, dunno whether / how good reverting to machine code could work, obviously never tried it 🥴
Maybe I should rephrase that - getting as close to machine code as possible AND sensible. If that's C (obviously a lower level language than Javascript), so be it.
For the sake of it - let's just say you're a React/Javascript kiddo and have never heard of - I dunno - garbage collection. A.I. will implement that for you (in whichever low level language) and explain to you what it's doing and why that's necessary.
On the other hand, having an A.I. generate REACT code is like building a fully equipped and heavily armed aircraft carrier to cross the Mississippi river
@@christian-schubert I tried x86 assembly prompts. Boy, it sucks and horrible at explaining it so horrible. I don't think it will be good at C or Rust either without getting a punch from the borrower checker.
@@unknownguywholovespizza Rust is relatively new language, I would imagine it struggling but C?
Honestly if it can't make memory safe program in C ( as in do it correctly without borrow checker ) then it has a long way to go.
i think its clear now that AI-Driven development will allow for the creation of monstrous software with quadrillion or quintillion lines of code. if its true that AI efficacy decreases with complexity then the future of human software development is bright...albeit much different from what is considered normal/standard today.
our jobs are safe boys!
Who needs prompt engineers when we can ask an AI to write its own prompts? Maybe we just need a very good base set of prompts to start with, just like some math axioms. Just to be sure I'm gonna start developing some good old fashioned farm skills to protect myself from being fired by my future AI boss for not being able to center a div inside another div
With AI, we will become exponentially better at doing things, which is kind of crazy. It's helping us make it better.