Would luv to see you do a similar app with aider and claude 3.5, go through the same steps and see which one is faster/easier - less clicks, less copy paste. Also some background on token costs would be really helpful. Great videos btw ❤
Thank you for that @amritbanerjee! I thought the same thing after I finished this video - go back and build it with aider. I'll let you know what I find if I get to that. I added a note on my costs in the video description. I didn't track it, but it was under $1 U.S. for sure. That includes all 3 models I used.
Great content as always. Thank you for sharing such valuable and pioneering insides from this field of great interest, The AI Coding Assistants and Tooling Have you seen also Cursor's new multi-file Editing AI assistant , they called it "Cursor Composer" ?
Thank you so much for your kind words @georgeafrah4608! I've watched a couple of videos on that new Cursor feature and have talked with someone who's used it. It definitely seems super cool! I have Cursor on my list of new things to try soon. Have you used it yet? If so, how's it working for you?
@@CodingtheFuture-jg1he Cursor's Composer is really brand new, just came out few weeks ago. I've played with it in few test drive projects and i would say it's mind-blowing. Pretty similar to Aider features wise. But in my opinion Aider still wins here mainly because Aider is free Open Source Software while Cursor-composer is paid for Editor
Att 11:50 for the example of adding axios with npm, no need to copy and paste in the terminal, just click on the Play button in the chat and it will install automatically
Great tutorial, just a shame you stopped at the emulation of the code. Would have been great to see it end-to-end to show the app working standalone and deployed to an app store. Looking forward to the next tutorial! I don't subscribe to many channels so you're definitely creating unique valuable content.
Thank you for that @NaturalMelodicHarmonic! I hear ya. Sorry about that. It's always this dance between dragging out a tutorial too long and getting in everything I'd like to share 😆 But, if you read up on the instructions I linked to the video, you'll see that all you need to do to run it on your phone is to either install the Expo Go app on your phone and then either: a. Scan the QR code in your expo terminal; OR b. Link your phone to your PC via USB cable (I personally have not tried this method, but the Expo docs I linked say it works)
Hey @serafinmusic! I used Cody for a few weeks back in May-June. It performed pretty well. But after using Continue for a couple of weeks now, I prefer Continue at the moment. I've gotten a bit more reliable results from Continue, especially with Claude 3.5 (but I also use it with a local model for some tasks). One thing that Cody did (back then at least) that bugged me was to pretty consistently mess up in-line edits to an existing file. Maybe fixed by now. Seems to me that, feature-wise, they're mostly tracking very closely. For me personally, one of the biggest features I insist on in any assistant is the ability to choose which LLM I want - commercial/open, local/cloud - and I want the configuration and toggling process to be simple. I also want the tools I choose to be open source. That's very important to me. Both Continue and Cody are, but you do have to create an account with SourceGraph to start using Cody in reality. Probably my biggest reason for gravitating towards Continue right now is how really easy it is to understand what's going on, how to add/update its configuration, etc. I haven't run into anything I wanted to know or do with Continue so far that I haven't been able to quickly figure out. Although I've spent more hours using Cody, I still have a clearer picture of what's what with Continue. For me, getting that balance of power and simplicity is a key. I do not like black boxes when it comes to stuff like this😆
It is so frustrating that you have to constantly tell it to not provide snippets. Even if you create a project and specify how you want your code provided. It does not work. I find that the project files are mostly useless since it does not ready from them correctly. Even when files are in the knowledge base I still have to copy and paste to get it to use the correct file which uses up my messages very quickly. I cancelled until they fix it.
Thank you for watching and commenting @OldGamr! I can see you're frustrated. I'd love to help. Unfortunately, I'm not clear on the exact issue you're experiencing. Did you follow the installation & setup instructions and the prompts to the letter? And still having issues? I ask because I understand there are a lot of details and it can get overwhelming until you've done this stuff enough repetitions to get a good feel for it. There's always a chance I left out some critical detail, so would like to know that. I did repeat the tutorial steps several times before publishing and I got consistent results, because I don't want you all dealing with unnecessary issues. I did have one person tell me for another video that they decided to swap out the LLM I used in the video and things went awry. There's a reason I never say "But, feel free to change this or that and you'll get the same results". Having said all that, as I point out in some other videos, we are dealing with generative AI, which by its very nature is non-deterministic. So, there is a possibility you're doing everything correctly and still getting different results than I am. Please don't give up on this stuff. It takes time and patience. But, once you got it down, you'll be glad you took the time 😀
@@CodingtheFuture-jg1he Hello. Yes, I did follow the instructions. I am working on a more advanced application which uses the JUCE framework. I also uploaded the entire JUCE document library and Claude verified it could read my uploads. I studied several different methods for setting up the custom instructions and I am confident it was correct. The primary issues are simple, Claude does not follow instructions. I was very specific in that I did not want code snippets or placeholder code. As long as I added this requirement to each message it would usually follow my instructions. If I did add it to the chat it would immediately add placeholder code and snippets. It also would make things up and add code that I did not request. That code would have errors since it was not related to the project. Claude would then apologized and provide the correct code. At one point, there were errors in my code and Claude kept posting what it said were fixes. Same error kept popping up. I compared the code, no changes were made at all. I actually asked Claude if it gave me the exact same code that was broken. it said yes, and apologized for not actually providing a resolution. Claude is great and I think it will become a very valuable tool. As long as you are creating more basic applications it has fewer issues. ChatGPT does the exact same thing in the same way. They both are amazing when it comes to Python and I have built some pretty complete Python apps with fewer problems. The only area that ChatGPT beats Claude is with code continuation and the length of time you can use it. I can work for hours before using up my limit and then I only have to wait for about 20 minutes. Claudes message limit gets hit very fast and then you have to wait over 3 hours. They also both suffer when your chat gets very long. Also, neither one carries chat sessions across to other chat sessions and they both state this. You have to start over each time. I usually have it create a prompt with the project to make the transition less painful. Worlds longest RUclips comment..Sorry
This is crazy.🤯 What a world am I living in!🔥 Great video! Appreciate if you could share the cost also. Looking forward for more ❤
Thank you so much @vishnuitsrocking! Please check the video description. I put a note about costs somewhere in there.
@@CodingtheFuture-jg1he cool. Thank you 👍🏼
Would luv to see you do a similar app with aider and claude 3.5, go through the same steps and see which one is faster/easier - less clicks, less copy paste. Also some background on token costs would be really helpful. Great videos btw ❤
Thank you for that @amritbanerjee! I thought the same thing after I finished this video - go back and build it with aider. I'll let you know what I find if I get to that. I added a note on my costs in the video description. I didn't track it, but it was under $1 U.S. for sure. That includes all 3 models I used.
Great content as always. Thank you for sharing such valuable and pioneering insides from this field of great interest, The AI Coding Assistants and Tooling
Have you seen also Cursor's new multi-file Editing AI assistant , they called it "Cursor Composer" ?
Thank you so much for your kind words @georgeafrah4608!
I've watched a couple of videos on that new Cursor feature and have talked with someone who's used it. It definitely seems super cool! I have Cursor on my list of new things to try soon. Have you used it yet? If so, how's it working for you?
@@CodingtheFuture-jg1he Cursor's Composer is really brand new, just came out few weeks ago. I've played with it in few test drive projects and i would say it's mind-blowing. Pretty similar to Aider features wise. But in my opinion Aider still wins here mainly because Aider is free Open Source Software while Cursor-composer is paid for Editor
Att 11:50
for the example of adding axios with npm, no need to copy and paste in the terminal, just click on the Play button in the chat and it will install automatically
Thank you so much for pointing that out @ginocote! I totally missed on that one. That's clearly the way to go.
Great video!
Great tutorial, just a shame you stopped at the emulation of the code. Would have been great to see it end-to-end to show the app working standalone and deployed to an app store.
Looking forward to the next tutorial! I don't subscribe to many channels so you're definitely creating unique valuable content.
Thank you for that @NaturalMelodicHarmonic!
I hear ya. Sorry about that. It's always this dance between dragging out a tutorial too long and getting in everything I'd like to share 😆
But, if you read up on the instructions I linked to the video, you'll see that all you need to do to run it on your phone is to either install the Expo Go app on your phone and then either:
a. Scan the QR code in your expo terminal; OR
b. Link your phone to your PC via USB cable (I personally have not tried this method, but the Expo docs I linked say it works)
Great tutorial, but when i use expo go i cant select an image in my phone, the button not working,
do you have any experience with groq?
What are your thoughts on Cody vs Continue?
Hey @serafinmusic! I used Cody for a few weeks back in May-June. It performed pretty well. But after using Continue for a couple of weeks now, I prefer Continue at the moment. I've gotten a bit more reliable results from Continue, especially with Claude 3.5 (but I also use it with a local model for some tasks). One thing that Cody did (back then at least) that bugged me was to pretty consistently mess up in-line edits to an existing file. Maybe fixed by now.
Seems to me that, feature-wise, they're mostly tracking very closely.
For me personally, one of the biggest features I insist on in any assistant is the ability to choose which LLM I want - commercial/open, local/cloud - and I want the configuration and toggling process to be simple. I also want the tools I choose to be open source. That's very important to me. Both Continue and Cody are, but you do have to create an account with SourceGraph to start using Cody in reality.
Probably my biggest reason for gravitating towards Continue right now is how really easy it is to understand what's going on, how to add/update its configuration, etc. I haven't run into anything I wanted to know or do with Continue so far that I haven't been able to quickly figure out. Although I've spent more hours using Cody, I still have a clearer picture of what's what with Continue. For me, getting that balance of power and simplicity is a key. I do not like black boxes when it comes to stuff like this😆
This was a very helpful tutorial.
Thank you so much @Techpeasy!
Does it work with flutter ??
Yes but flutter uses dart language and AI's not that good with newer languages , gemini is only good with dart
Hi @lawwong2319! I agree with what @xelerator2398 said below.
It is so frustrating that you have to constantly tell it to not provide snippets. Even if you create a project and specify how you want your code provided. It does not work. I find that the project files are mostly useless since it does not ready from them correctly. Even when files are in the knowledge base I still have to copy and paste to get it to use the correct file which uses up my messages very quickly. I cancelled until they fix it.
Thank you for watching and commenting @OldGamr!
I can see you're frustrated. I'd love to help. Unfortunately, I'm not clear on the exact issue you're experiencing. Did you follow the installation & setup instructions and the prompts to the letter? And still having issues?
I ask because I understand there are a lot of details and it can get overwhelming until you've done this stuff enough repetitions to get a good feel for it. There's always a chance I left out some critical detail, so would like to know that. I did repeat the tutorial steps several times before publishing and I got consistent results, because I don't want you all dealing with unnecessary issues.
I did have one person tell me for another video that they decided to swap out the LLM I used in the video and things went awry. There's a reason I never say "But, feel free to change this or that and you'll get the same results".
Having said all that, as I point out in some other videos, we are dealing with generative AI, which by its very nature is non-deterministic. So, there is a possibility you're doing everything correctly and still getting different results than I am.
Please don't give up on this stuff. It takes time and patience. But, once you got it down, you'll be glad you took the time 😀
@@CodingtheFuture-jg1he Hello. Yes, I did follow the instructions. I am working on a more advanced application which uses the JUCE framework. I also uploaded the entire JUCE document library and Claude verified it could read my uploads. I studied several different methods for setting up the custom instructions and I am confident it was correct. The primary issues are simple, Claude does not follow instructions. I was very specific in that I did not want code snippets or placeholder code. As long as I added this requirement to each message it would usually follow my instructions. If I did add it to the chat it would immediately add placeholder code and snippets. It also would make things up and add code that I did not request. That code would have errors since it was not related to the project. Claude would then apologized and provide the correct code. At one point, there were errors in my code and Claude kept posting what it said were fixes. Same error kept popping up. I compared the code, no changes were made at all. I actually asked Claude if it gave me the exact same code that was broken. it said yes, and apologized for not actually providing a resolution. Claude is great and I think it will become a very valuable tool. As long as you are creating more basic applications it has fewer issues. ChatGPT does the exact same thing in the same way. They both are amazing when it comes to Python and I have built some pretty complete Python apps with fewer problems. The only area that ChatGPT beats Claude is with code continuation and the length of time you can use it. I can work for hours before using up my limit and then I only have to wait for about 20 minutes. Claudes message limit gets hit very fast and then you have to wait over 3 hours. They also both suffer when your chat gets very long. Also, neither one carries chat sessions across to other chat sessions and they both state this. You have to start over each time. I usually have it create a prompt with the project to make the transition less painful. Worlds longest RUclips comment..Sorry