This is such an airhead statement. No one ever said that this was to do anything hard. If you are a programmer and not using LLMs, you are wasting time period. To be able to be in a file and just hit ctrl + k and have your IDE just automatically do things like adding dark mode styling to every element in your file or assisting with making your app more responsive with a sentence is such a value add for 20 dollars a month for the sheer amount of time this saves. I have been literally saving myself 4 to 20 hours a week when building and maintaining client work. Cursor, Aider or Replit are such valuable tools I feel like are so much worth the price you are paying. It's like the first time you type on a Kinesis keyboard after using a standard keyboard your entire life, when you hit buttons other than the space bar with your thumb, so much power so much precision, it feels almost godly in a way. It's a surreal experience and it will only get better from here.
@@CourageToGroww cheap insult man, not cool. Anyway, it's not so airhead. I mean imagine where the technology will be. I myself am not a programmer by trade. I've always struggled teaching myself, due to other commitments. But, with all this new technology my endless pipe dreams are a little more realistic. Me personally, I'm having trouble with big picture and architecture things. I have a big concept but can't quite put the pieces together yet. It will be helpful for me when these technologies can really reason over more complicated systems that allow me to simply be the director of features and functionality. Either way, there are a lot of people getting into software development, many of whom have domain specific knowledge that's not tech related. I look forward to seeing what is possible. Have a nice day 🎉
@@CourageToGroww The more you use AI, the more tech dept you create. It is not omnipotent. It's easy to produce simple code that works, but not code that will last. The bigger your codebase, the more annoying it is to prompt it correctly. It can save you time, but the more you use it, the more you begin to rely on it and the harder it is to code without it. You train your brain to think of prompts before code solutions. If all you do is create simple webpages and scripts then its fine, it does 90% of the work with ease. But, larger applications will fall apart as you spend more time prompting, editing, and later addressing the tech dept it created.
yeah it falls short for large codebase and async/multi processing logic. But we will soon see llms writing code that is concise,clean w/o technical debt
Cursor can also have codebase-wide context either in chat or editor mode(CMD+L or CMD+K). you just need to either type @codebase or use CMD+Enter when submitting your prompts. I like to use Aider to start projects or implement features that require editing or creating multiple files at once, then use Cursor chat with codebase-wide context to refine or troubleshoot. The paid membership allows the "apply" feature which makes the changes suggested by the LLM of my choice(Claude 3.5 Sonnet for now) automatically. I enjoy working with them very much.
Before even starting this video, I've become more self-aware about myself having this growing desperation to learn more about AI Coding since it feels like this area is going to be something integral to tech and possibly all things software-related as the AI models get better. I want to break down what skills I should develop to get better at AI Coding, in some sense it feels like a specialization of Prompt Engineering. I think obviously one skillset to develop is a basic or proficient understanding of the coding language your using, but in order to do bigger things than just have it generate code for me, like to ultimately create interesting Apps, I feel like I'm missing some skillsets or information I should be working on when I'm playing around with these models...
Hi Dan, thanks so much for your magic videos! I did the AIDER + Ollama, but I am stuck with Aider + Cursor + + OpenUI + Ollama models. My workflow is to create passion through this combination for other students. WITH THIS project they learn + see + make it work + deploy, and it drives insane passion to create. Most engineers are broke, that made 250k a year, imagine grads or newbies now, that have little money and time is ticking faster than money is becoming. Please help ALL, to help a new pathway for creation.code is not the purpose but ideas inside to come is the gold especially in younger minds. Thanks, the Creator
I've used Codium a completely free ide integrated copilot to do the things which are called innovative here long ago. I already have my code autocomplete for me while i'm writing and taking the context from my codebase directly, also can write a comment explaining what it needs to do and it will do it. Not as smart as the top models, but can be easily installed with no external API usage
Dan. Great video.. could you do a quick start video? Setup + tips, using a basic application. Include how to build your code dunctions by referencing other existing code from other folders or repos. That would be really useful.
I use AI coding assistance every day. My biggest issue with them is that the models don't stay up to date as new languages and language enhancements are released. Do this experiment again but this time ask it to build the same app using Next.js 14. I can almost guarantee it will start building an API for form submission instead of just using Server Actions. It'll also probably build the components as client components by default, which is contrary to best practices for modern Next.js application development. All this leads to me spending time correcting these mistakes or crafting a very very very specific prompt for just a small portion of what I need a function or component to to do. I've had to fine-tune available coding models to include and use information about pretty much any language or frameworks built or enhanced within that last 2 years. This is time consuming because it requires sourcing the data, prepping it and then training the model. My system has a single 3080ti in it. It's a good system but it still takes a long time to train models.
I have tried to run this on my VSC and it won’t run. The aider command. Can you help brother. It is showing up the file. But is not linking up with a api key. I tried it with Gemini flash 1.5
@IndyDevDan I have a project that uses some githubs and then need to replace the libraries with JAX libraries, and obviously mutli-file uploading, it is complicated and have problems debugging. Would cursor or aider be best for to tackle this problem. Existing gpt4 can not debug. Thanks
I understand the concept of your warning, how do I get started? I dont understand the prompt you started with or any of the outputs. I want to invest an hour or two daily to master this skill. Thank you for your help!
the aider webui would be my goto but its still missing some feature and writes to files directly and forces commits .. which you shouldnt do as its not reliable enough for that .
Notice to navigators. I have been testing ‘cursor’ and its integration with the i.a. model is impressive. But I have to say that I have tried every possible way to make it generate uml dependency diagrams on a relatively simple code base and it has turned out to be anything but useful. I have tried actively and passively to make it understand what I wanted but the model invents what it wants according to what it thinks you are asking it, but it forgets the context and simply lies, contradicts itself and generates more distrust than anything else. It's a pity because it is very well integrated and promising, but as of today it lacks a lot, and I'm talking about gpt4. I think there is a lot of hype and a lot of desire for notoriety in all this, but time will put everyone in their place.
Retired hardware eng. started pgm on Z80 uPc with dip switches thru to today now with a few 'modern' IDE's. Very interested in Ai programming for ( hobby use only ). Watched your vid and TOTALLY lost. I guess we download Cursor, ok, will have to read the docs as learned nothing about it here. How the ??? do you even start to use aider. Are your vids for the prize guys that really already know about all these things, or am I missing something so basic, no pun. Many thks for the effort, but a 'Hello World' would have been just great, so totally lost here, is this maybe part x of a series I have hit on and I need the other parts to make sense of this ??. Kind Rgds.
Waiting for it to develop more. It's in the 'ai software engineer' category of ai coding tools which honestly, none of those tools are great for editing EXISTING codebases yet. They're good for new, toy applications. I"m keeping an eye on the 'ai software engineer' category of ai coding tools but they need more time, better LLMs (gpt5, claude-3.5 opus, gemini-1.5 ultra) to do real work. One of them will likely pop off in 2025 maybe it will be gpt engineer, maybe not.
It's incorrect comparison of performance just to use the amount of words for promt and generated code. You need some time to prepare this promt, seems it will take many tries.
You’re right but you’re ignoring the productivity gains of increasing your most valuable asset as an engineer: CODE. Think of the millions of lines of code you’ve written over your career. Now imagine if you can now do ALL that at 2-10x speeds!? That means you can ship MANY more of your hard problems in less time. Make sense?
Great, now spend more time solving those problems instead of writing them! Those tools work even better when you use pseudo code specific processes. You can tell it how to do things, how to access/use stuff etc.
aiderb doesn't work for me, I can use the command line though, anyone have this problem? I just have this "'aiderb' is not recognized as an internal or external command, operable program or batch file." error
Shytty non coder here🤣 Im learning, so im thinking this can assist me in getting better. Also, in a few years with AGI level foundational models, software development will probably be make a big jump.
@@clementita54 .That's because you don't know how to be resourceful. AI is an amplifier big time. I'm building scripts and learning key points way faster than with traditional methods. It's called having a tutor.u wanna stay in the stone age, be my guest.
@@mrd6869 🤣 No way a BEGINNER, try to tell a WORKING ENGINEER how to learn and how to be good at coding ...... Go on, keep using your method and be ready to feel stupid in front of peoples when they talk and brainstorm about algorithms, code and projects. I also hope for you that every company you will work for have a very small and permissive security team and they let you spread corporate data to an unknown source. Good luck kid
This has a lot going on for sure, not the best example of where to begin. Start simple with learning basic HTML/CSS, then move onto Python. Python is great for beginners and the primary language of all things AI/data currently. Get a sense for what you’re interested in building, and learn the languages/frameworks in that area. Could be front-end, iOS, ML, etc. Build your ideas and let curiosity guide you. Consistency x reps will help you achieve exit velocity.
I've been trying to use cursor all day and wasting time going around in circles getting angry with it. I will stick to co-pilot for short snippets for now. I feel like its been trained on too many hacky codebases
Using cursor and really like it a lot! Having a hard time when working with Github-Copilot then. Also used aider but since cursor has the "Interpreter mode" in the chat i am only using cursor now. Everything is so well integrated. @IndyDevDan have you tried out cursors interpreter mode (ruclips.net/video/r3BSULZNWBI/видео.html). Fun fact: From what i saw on first-use-imports seems to be using aider under the hood.
I just started with Aider after being underwhelmed with Cursor. I might be wrong but Cursor only has knowledge of the files you provide to the context explicitly or because they're open. Aider maps the entire project as a complete repository, so it has access to every file you haven't ignored. I'm sure I don't understand that completely and am happy to be corrected. Aider also automatically edits multiple files without intervention and commits all changes to the repo, which helps blunt the fear of an AI changing my codebase with wild abandon.
You're missing a point here, more lines doesn't equal better software. Performance of AI code tends to degrade quickly with length and complexity, especially if you want to build applications that are more advanced than a demo. Still, quite amazing how good LLMs have gotten, but the metric we measure shouldn't be the length of code produced
There is a development tool called an IDE, which is like a fancy editor for your code that does the colors and gives you error correction suggestions and some autocomplete capabilities. You'll have to search for an IDE for your preferred programming language.
I mean in general - it's a hot take but I don't think we should spend much time 'chunking' data outside of brute filtering especially when we have large context windows (and they will continue to increase). How do you feel about rag?
I'm not a coder, In fact I'm a a builder by trade. By using LLM's I've managed to build a 20,000 + line code app for my business completely from scratch, using AI 100%. Half the time I have no idea the specifics of the code it is typing, but I know what I want it to do, I know the what the functions should do and I work through debugging using AI. It's game changing. I know a lot of devs will bash me, "You need to learn how to code first before you can use AI". No, I need an app to do what I want and this is the tool I am using to build it. Just the same as I use a nail gun to bang together rafters. Game changer
Do these two tools also work as great if used with the newest open source LLMs that specialize in coding? - Or isn't it worth the time trying? I don't see any case in which these tools could be used in a professional company environment if they are only good if used with the huge LLMs of the leading AI companies. At least in our company is passing code or any other project related data to other companies strictly forbidden for obvious reasons.
@@henrischomacker6097you could host open source LLMs locally with Ollama, then use Aider and select Ollama as the model provider. The best open source coding LLM is DeepSeek Coder V2(as good as GPT 4O), Mistral Coder Nemo is also very good if your machine can't run DeepSeek Coder V2. Aider's documentation has all the information you need including what LLMs they recommend.
I've not seen 1 example online where someone does changes in context of a gigantic project with 10-15 files open that are all messy and include weird business logic. All AI examples are projects starting from 0 with crappy CRUD examples and snake games.
Code base will become full of bug and impossible to maintain, no one will be expert of his features and will be able to tell how to fix an issue without even looking at the code base
8x this 10x that This is not productivity, its just an indicator to how little brainwaves activity humanity is gonna have by 2040, congratz on making your AI write your boilerplate
It's mid, last time I checked it's missing the selection prompt. See selection prompt: ruclips.net/video/2j_fgMPJGM0/видео.htmlsi=g-ch254mCEy_xVmi&t=310
The hard part of coding isn’t writing boilerplate though. We need more demos showing off thoughtful changes to large codebases.
This is such an airhead statement. No one ever said that this was to do anything hard. If you are a programmer and not using LLMs, you are wasting time period. To be able to be in a file and just hit ctrl + k and have your IDE just automatically do things like adding dark mode styling to every element in your file or assisting with making your app more responsive with a sentence is such a value add for 20 dollars a month for the sheer amount of time this saves. I have been literally saving myself 4 to 20 hours a week when building and maintaining client work. Cursor, Aider or Replit are such valuable tools I feel like are so much worth the price you are paying. It's like the first time you type on a Kinesis keyboard after using a standard keyboard your entire life, when you hit buttons other than the space bar with your thumb, so much power so much precision, it feels almost godly in a way. It's a surreal experience and it will only get better from here.
@@CourageToGroww cheap insult man, not cool. Anyway, it's not so airhead. I mean imagine where the technology will be. I myself am not a programmer by trade. I've always struggled teaching myself, due to other commitments. But, with all this new technology my endless pipe dreams are a little more realistic. Me personally, I'm having trouble with big picture and architecture things. I have a big concept but can't quite put the pieces together yet. It will be helpful for me when these technologies can really reason over more complicated systems that allow me to simply be the director of features and functionality. Either way, there are a lot of people getting into software development, many of whom have domain specific knowledge that's not tech related. I look forward to seeing what is possible. Have a nice day 🎉
@@CourageToGroww The more you use AI, the more tech dept you create. It is not omnipotent. It's easy to produce simple code that works, but not code that will last. The bigger your codebase, the more annoying it is to prompt it correctly.
It can save you time, but the more you use it, the more you begin to rely on it and the harder it is to code without it. You train your brain to think of prompts before code solutions. If all you do is create simple webpages and scripts then its fine, it does 90% of the work with ease. But, larger applications will fall apart as you spend more time prompting, editing, and later addressing the tech dept it created.
yeah it falls short for large codebase and async/multi processing logic. But we will soon see llms writing code that is concise,clean w/o technical debt
@@alextasarov1341 If you rely on it then you are not a programmer.
Watching at x2 to achieve 16x productivity boost
Cursor can also have codebase-wide context either in chat or editor mode(CMD+L or CMD+K). you just need to either type @codebase or use CMD+Enter when submitting your prompts. I like to use Aider to start projects or implement features that require editing or creating multiple files at once, then use Cursor chat with codebase-wide context to refine or troubleshoot. The paid membership allows the "apply" feature which makes the changes suggested by the LLM of my choice(Claude 3.5 Sonnet for now) automatically. I enjoy working with them very much.
Before even starting this video, I've become more self-aware about myself having this growing desperation to learn more about AI Coding since it feels like this area is going to be something integral to tech and possibly all things software-related as the AI models get better. I want to break down what skills I should develop to get better at AI Coding, in some sense it feels like a specialization of Prompt Engineering. I think obviously one skillset to develop is a basic or proficient understanding of the coding language your using, but in order to do bigger things than just have it generate code for me, like to ultimately create interesting Apps, I feel like I'm missing some skillsets or information I should be working on when I'm playing around with these models...
Hi Dan, thanks so much for your magic videos! I did the AIDER + Ollama, but I am stuck with Aider + Cursor + + OpenUI + Ollama models. My workflow is to create passion through this combination for other students. WITH THIS project they learn + see + make it work + deploy, and it drives insane passion to create. Most engineers are broke, that made 250k a year, imagine grads or newbies now, that have little money and time is ticking faster than money is becoming. Please help ALL, to help a new pathway for creation.code is not the purpose but ideas inside to come is the gold especially in younger minds. Thanks, the Creator
Only having 13.7k subscribers is a crime. This channel is top notch.
🙏 Thank you brother.
totally agree, this has helped change my entire process.
I've used Codium a completely free ide integrated copilot to do the things which are called innovative here long ago. I already have my code autocomplete for me while i'm writing and taking the context from my codebase directly, also can write a comment explaining what it needs to do and it will do it. Not as smart as the top models, but can be easily installed with no external API usage
Your voice is EXCELLENT! I really like your no bs attitude. I wish I could chill in person!
This would have to be the best AI channel I’ve found… and I’m subbed to like 30 of them 🙏🏽
Dan. Great video.. could you do a quick start video? Setup + tips, using a basic application. Include how to build your code dunctions by referencing other existing code from other folders or repos.
That would be really useful.
I use AI coding assistance every day. My biggest issue with them is that the models don't stay up to date as new languages and language enhancements are released. Do this experiment again but this time ask it to build the same app using Next.js 14. I can almost guarantee it will start building an API for form submission instead of just using Server Actions. It'll also probably build the components as client components by default, which is contrary to best practices for modern Next.js application development. All this leads to me spending time correcting these mistakes or crafting a very very very specific prompt for just a small portion of what I need a function or component to to do.
I've had to fine-tune available coding models to include and use information about pretty much any language or frameworks built or enhanced within that last 2 years. This is time consuming because it requires sourcing the data, prepping it and then training the model. My system has a single 3080ti in it. It's a good system but it still takes a long time to train models.
When do we get something good for java its just all python and a bit of javascript
Just discovered your channel. Awesome value.
I am one year in to coding and loving it.
I have tried to run this on my VSC and it won’t run. The aider command. Can you help brother.
It is showing up the file. But is not linking up with a api key. I tried it with Gemini flash 1.5
@IndyDevDan I have a project that uses some githubs and then need to replace the libraries with JAX libraries, and obviously mutli-file uploading, it is complicated and have problems debugging. Would cursor or aider be best for to tackle this problem. Existing gpt4 can not debug. Thanks
"AI is the top tool to crank out as much code as possible"
Absolutely agreed
I think your channel is my new favorite. You are real and don’t whore yourself out. Well done brother.
I understand the concept of your warning, how do I get started? I dont understand the prompt you started with or any of the outputs. I want to invest an hour or two daily to master this skill. Thank you for your help!
the aider webui would be my goto but its still missing some feature and writes to files directly and forces commits .. which you shouldnt do as its not reliable enough for that .
Notice to navigators. I have been testing ‘cursor’ and its integration with the i.a. model is impressive. But I have to say that I have tried every possible way to make it generate uml dependency diagrams on a relatively simple code base and it has turned out to be anything but useful. I have tried actively and passively to make it understand what I wanted but the model invents what it wants according to what it thinks you are asking it, but it forgets the context and simply lies, contradicts itself and generates more distrust than anything else. It's a pity because it is very well integrated and promising, but as of today it lacks a lot, and I'm talking about gpt4. I think there is a lot of hype and a lot of desire for notoriety in all this, but time will put everyone in their place.
I like how you talk my boy you’re very positive
Devin + GPT5 = App Dev unlocked
Cursor is my dna now
Nice, never gets tired of hearing the click clacks
Thanks for the video, why use cursor if aider can do all and more ? I didnt tested cursor yet but i used aider and works nice with claude . Thanks you
You earned my like & subscribe. Thank you very much for introducing me to both of these honestly incredible tools.
Same here. I am definitely switching over to Cursor
Hi. Subbed and liked. Great content. One question, how do you get the tokens to show bottom right of UI?
Search for the vscode extension named: 'Token Count in Status Bar...'
Link: marketplace.visualstudio.com/items?itemName=marxism.ai-token-count
@@indydevdan Thank you
Which colorscheme is used in the editor
Retired hardware eng. started pgm on Z80 uPc with dip switches thru to today now with a few 'modern' IDE's. Very interested in Ai programming for ( hobby use only ). Watched your vid and TOTALLY lost. I guess we download Cursor, ok, will have to read the docs as learned nothing about it here. How the ??? do you even start to use aider. Are your vids for the prize guys that really already know about all these things, or am I missing something so basic, no pun. Many thks for the effort, but a 'Hello World' would have been just great, so totally lost here, is this maybe part x of a series I have hit on and I need the other parts to make sense of this ??. Kind Rgds.
How much do you spend on ai for actual work? Ball prk?
How do we get the Aider browser plugin?
Its command `aider --browser`
@@CurlabsAdmin 🤩 THANK YOU
@IndyDevDan What do you think of GPT Engineer?
Waiting for it to develop more. It's in the 'ai software engineer' category of ai coding tools which honestly, none of those tools are great for editing EXISTING codebases yet. They're good for new, toy applications.
I"m keeping an eye on the 'ai software engineer' category of ai coding tools but they need more time, better LLMs (gpt5, claude-3.5 opus, gemini-1.5 ultra) to do real work. One of them will likely pop off in 2025 maybe it will be gpt engineer, maybe not.
It's incorrect comparison of performance just to use the amount of words for promt and generated code. You need some time to prepare this promt, seems it will take many tries.
What about Codium AI and their new framework CodiumAlpha?
My productivity isn’t measured in words written. It’s measured in hard problems solved.
I don't understand this mentality, brother is telling you that you will be able to solve more problems in less time.
You’re right but you’re ignoring the productivity gains of increasing your most valuable asset as an engineer: CODE.
Think of the millions of lines of code you’ve written over your career. Now imagine if you can now do ALL that at 2-10x speeds!? That means you can ship MANY more of your hard problems in less time.
Make sense?
@@indydevdan you must be working on some upwork style place pumping out shoddy code that you never have to maintain
Great, now spend more time solving those problems instead of writing them!
Those tools work even better when you use pseudo code specific processes.
You can tell it how to do things, how to access/use stuff etc.
@@indydevdan no, because the time spent physically writing the code is easily less than 5% of the time it takes to solve the problem.
aiderb doesn't work for me, I can use the command line though, anyone have this problem?
I just have this "'aiderb' is not recognized as an internal or external command,
operable program or batch file." error
aider --browser
Shytty non coder here🤣
Im learning, so im thinking this can assist me in getting better.
Also, in a few years with AGI level foundational models, software development will
probably be make a big jump.
This wont make you better, repetition and logic makes you better, this is just help when you ALREADY KNOW how to do it
@@clementita54 .That's because you don't know how to be resourceful. AI is an amplifier big time. I'm building scripts and learning key points way faster than with traditional methods. It's called having a tutor.u wanna stay in the stone age, be my guest.
@@mrd6869 🤣 No way a BEGINNER, try to tell a WORKING ENGINEER how to learn and how to be good at coding ......
Go on, keep using your method and be ready to feel stupid in front of peoples when they talk and brainstorm about algorithms, code and projects. I also hope for you that every company you will work for have a very small and permissive security team and they let you spread corporate data to an unknown source.
Good luck kid
if anyone finds Cursor too expensive, especially with using the higher token models, reply back to me and I'll blow your mind...
I want to learn about coding and programming but I'm not sure where to start. There's so much going on.
This has a lot going on for sure, not the best example of where to begin.
Start simple with learning basic HTML/CSS, then move onto Python. Python is great for beginners and the primary language of all things AI/data currently.
Get a sense for what you’re interested in building, and learn the languages/frameworks in that area. Could be front-end, iOS, ML, etc.
Build your ideas and let curiosity guide you. Consistency x reps will help you achieve exit velocity.
amazing work. Thanks for all your great info.
Please make a video about GPT Pilot (Pyhtagora)
I've been trying to use cursor all day and wasting time going around in circles getting angry with it. I will stick to co-pilot for short snippets for now. I feel like its been trained on too many hacky codebases
Uhm why you dont use the Cursor Chat or add the files with @ ? That way Cursor has also the context of the files
ya.. I don't get what aider is doing that you can't do with cursor (I only use cursor)
Dude fr I noticed no difference except extra steps and more money for no reason
Using cursor and really like it a lot! Having a hard time when working with Github-Copilot then.
Also used aider but since cursor has the "Interpreter mode" in the chat i am only using cursor now. Everything is so well integrated.
@IndyDevDan have you tried out cursors interpreter mode (ruclips.net/video/r3BSULZNWBI/видео.html).
Fun fact: From what i saw on first-use-imports seems to be using aider under the hood.
I just started with Aider after being underwhelmed with Cursor. I might be wrong but Cursor only has knowledge of the files you provide to the context explicitly or because they're open. Aider maps the entire project as a complete repository, so it has access to every file you haven't ignored. I'm sure I don't understand that completely and am happy to be corrected. Aider also automatically edits multiple files without intervention and commits all changes to the repo, which helps blunt the fear of an AI changing my codebase with wild abandon.
@@ontheruntonowhere wrong. Cursor figures out the files to use. Anything that can’t do this I don’t bother
You're missing a point here, more lines doesn't equal better software. Performance of AI code tends to degrade quickly with length and complexity, especially if you want to build applications that are more advanced than a demo. Still, quite amazing how good LLMs have gotten, but the metric we measure shouldn't be the length of code produced
I learned many things, thank you
Please check the audio. I guess it's you keyboard typing in the background. With Airpods Pro it's extremely annoying and I can't finish the video.
How do you make your codes so colorful and stylish? Is there a way I can do that on my windows pc
There is a development tool called an IDE, which is like a fancy editor for your code that does the colors and gives you error correction suggestions and some autocomplete capabilities. You'll have to search for an IDE for your preferred programming language.
He was talking about the actual theme @@Jumpyfoot
@@chillgames823 Oh, I see. Well, you have a good day.
When you say you're very anti-RAG, do you mean in general, or only as it pertains to code generation?
I mean in general - it's a hot take but I don't think we should spend much time 'chunking' data outside of brute filtering especially when we have large context windows (and they will continue to increase).
How do you feel about rag?
Awesome 🚀 content! Thanks so much.
I'm not a coder, In fact I'm a a builder by trade. By using LLM's I've managed to build a 20,000 + line code app for my business completely from scratch, using AI 100%. Half the time I have no idea the specifics of the code it is typing, but I know what I want it to do, I know the what the functions should do and I work through debugging using AI. It's game changing. I know a lot of devs will bash me, "You need to learn how to code first before you can use AI". No, I need an app to do what I want and this is the tool I am using to build it. Just the same as I use a nail gun to bang together rafters. Game changer
Do these two tools also work as great if used with the newest open source LLMs that specialize in coding? - Or isn't it worth the time trying?
I don't see any case in which these tools could be used in a professional company environment if they are only good if used with the huge LLMs of the leading AI companies.
At least in our company is passing code or any other project related data to other companies strictly forbidden for obvious reasons.
If you're looking for open source support I recommend Aider: aider.chat/docs/llms.html
@@indydevdan Many thanks!
@@henrischomacker6097you could host open source LLMs locally with Ollama, then use Aider and select Ollama as the model provider. The best open source coding LLM is DeepSeek Coder V2(as good as GPT 4O), Mistral Coder Nemo is also very good if your machine can't run DeepSeek Coder V2. Aider's documentation has all the information you need including what LLMs they recommend.
I've not seen 1 example online where someone does changes in context of a gigantic project with 10-15 files open that are all messy and include weird business logic. All AI examples are projects starting from 0 with crappy CRUD examples and snake games.
Agreed - I've noticed this as well. I'll address and show off AI Coding on a 'real' codebase (minimum 100 files)
Like & Sub, incredible tutorial. Thank you
Heavenly sent
I goon to your videos
The typing sounds hurts my brain.
What about plandex?
Code base will become full of bug and impossible to maintain, no one will be expert of his features and will be able to tell how to fix an issue without even looking at the code base
Gpt 3.5 I can write longer codes with one prompt. Rookie
8x this 10x that
This is not productivity, its just an indicator to how little brainwaves activity humanity is gonna have by 2040, congratz on making your AI write your boilerplate
git mad
i use cursor is great:)
Amateur coder in action.
keep talking non-sense, better fill those 20 mins showtime for youtube algo, you're whole point could have been said in less than 4 mins you dofus
git mad
What do you think of Cody? Have you tried using that one?
It's mid, last time I checked it's missing the selection prompt.
See selection prompt: ruclips.net/video/2j_fgMPJGM0/видео.htmlsi=g-ch254mCEy_xVmi&t=310
The hard part of coding isn’t writing boilerplate though. We need more demos showing off thoughtful changes to large codebases.