I first try free models as gemini and others, both cline and roo-cline do wrong code diff placements sometimes, especially with big code files. But I have found a program which does it good with free trial. Windsurf and cursor should do it good but paid. All the point is to find the good diff code change. All the programs can make something in the first request. But then it's the purpose
I find roo cline loops out and ignores instructions quicker than cline. I’ve also never been able to get the results with anything running deepseek that I can with Claude. It appears better than 4o but a good bit below Claude still.
Anyone know which ai agent coding platform can be automated to create multiple apps without my intervention overnight? Give it a list of apps to create and it does all one at a time.
@@PubgSpeed-wl8yo I am making a big business management with an integrated AI agent. With frontend and backend, double database based on electron+vite+react. I use any model based on claude for the frontend and deepseek v3 for the backend. Cursor + Roo Cline. I am sorry to hear many people say such things, maybe they lack experience, I have tried dozens of combinations and models and I have found this balance, sometimes I also draw from GitHub Copilot when I need to debug the code, but basically you can have the best claude but if the fine-tuning sucks it is useless to use claude. Claude does not work well everywhere, it is not a constant, it depends on the fine-tuning. So friend, you are wrong.
When i want to ask any Ai coder to translate my projects i always use "change my project to Arabic and make it RTL" perfect results every single time, then you can ask for option or a button in the navigation bar to switch between the two languages
I tried to use it for a dotnet 9 application and the message i get from deepseek is, I am DeepSeek-V3, an upgraded version of the DeepSeek model. My training data includes information up until October 2023 so it has no knowledge for anything after October 23
On the site deepseek doesn't give many times good code for what asked. Maybe because it needs doc context. claude 3.5 on the site works. The issue could be cline that doesn't add the code in the right format. I found good programs which do . Cursor is paid but maybe should do it right. There are other
@@Codewello I use the claude for the frontend and deepseek 3 for the backend. As you say in the video, you are right. When I run commands to modify the frontend, any template always succeeds. But when it comes to backend deepseek is unbeatable, I am using Zustand, SQLite3 and Firestore in the same project. I am making a whole big Ai business management with Cursor and Roo Cline (with deepseek v3) and github copilot (vscode), so I am not making websites or todo apps. Do not switch to Windsurf, after a while the model slows down until it stops completely becoming useless. I immediately abandoned windsurf, sorry but Cursor is even stronger. Do not look at the 5 dollars difference. Even if you exhaust the 500 quick requests I noticed that if it is not overloaded it lets you go fast with claude and then in reality it is not a long wait, for example github copilot is eternal sometimes and makes many errors despite having models like o1. I found my balance like this, as long as the deepseek tokens are on sale I will stock up.
You should compare Cline to Roo Cline in regards to coding / code dif editing perfomance
yeah I'm still trying to figure out what is different about them
I first try free models as gemini and others, both cline and roo-cline do wrong code diff placements sometimes, especially with big code files. But I have found a program which does it good with free trial. Windsurf and cursor should do it good but paid. All the point is to find the good diff code change. All the programs can make something in the first request. But then it's the purpose
I can make a video on that. Thanks for the idea!
I find roo cline loops out and ignores instructions quicker than cline. I’ve also never been able to get the results with anything running deepseek that I can with Claude. It appears better than 4o but a good bit below Claude still.
What is the difference between Cline and the Roo Cline plugin? Which one do you think I should use?
Roo-Cline is Cline, but with additional features.
Anyone know which ai agent coding platform can be automated to create multiple apps without my intervention overnight? Give it a list of apps to create and it does all one at a time.
Come back in a year, it's still too early
wait for 6 -12 months
What you're asking for is not available yet, and it will be exciting for a while.
I'm using it with Roo-cline building with deepseek directly API (don't know why do you use hf). And changing from Claude 3.5.& Deepseek.
Claude, much better, have you done anything serious on Deepseek?
the model is really good but not that good as sonnet 3.5 new more like older version
@@PubgSpeed-wl8yo I am making a big business management with an integrated AI agent. With frontend and backend, double database based on electron+vite+react. I use any model based on claude for the frontend and deepseek v3 for the backend. Cursor + Roo Cline. I am sorry to hear many people say such things, maybe they lack experience, I have tried dozens of combinations and models and I have found this balance, sometimes I also draw from GitHub Copilot when I need to debug the code, but basically you can have the best claude but if the fine-tuning sucks it is useless to use claude. Claude does not work well everywhere, it is not a constant, it depends on the fine-tuning. So friend, you are wrong.
can't imagine where we will be in 5 years, what a time to be alive
I think still the Claude is the edge!
yes it's still the best in general
Claude is good for coding ?
@@custim9188 claude has no competitors in code. deepseek and gemini can be near but maybe they need documentation for the task
Hit = data already available, fast response.
Miss = data not available, processing required.
thank you so much
Lovely review
Glad you enjoyed it!
When i want to ask any Ai coder to translate my projects i always use "change my project to Arabic and make it RTL" perfect results every single time, then you can ask for option or a button in the navigation bar to switch between the two languages
Cline + deepseek, not free but cheap, best for me.
I'm sure hf in glhf stands for Hugging Face
So, it's not 'Good Luck, Have Fun'? That sounds weird if it's 'Good Luck, Hugging Face.
@@Codewello hehehe. Maybe it's not about good luck either?
What do you guys think about this model? Have you ever used it? 😎
I tried to use it for a dotnet 9 application and the message i get from deepseek is,
I am DeepSeek-V3, an upgraded version of the DeepSeek model. My training data includes information up until October 2023
so it has no knowledge for anything after October 23
but sometimes stoping
the api ?
I like you Egyptian English accent
thanks brother 😁
Cline or roocline + deepseek : cheap but too slow.
On the site deepseek doesn't give many times good code for what asked. Maybe because it needs doc context. claude 3.5 on the site works. The issue could be cline that doesn't add the code in the right format. I found good programs which do . Cursor is paid but maybe should do it right. There are other
I'm really thinking about switching to windsurfing just to save money.
@@Codewello cline too many requests. glhf not give credits for users at moment. There are alternatives
@@Codewello I use the claude for the frontend and deepseek 3 for the backend. As you say in the video, you are right. When I run commands to modify the frontend, any template always succeeds. But when it comes to backend deepseek is unbeatable, I am using Zustand, SQLite3 and Firestore in the same project. I am making a whole big Ai business management with Cursor and Roo Cline (with deepseek v3) and github copilot (vscode), so I am not making websites or todo apps. Do not switch to Windsurf, after a while the model slows down until it stops completely becoming useless. I immediately abandoned windsurf, sorry but Cursor is even stronger. Do not look at the 5 dollars difference. Even if you exhaust the 500 quick requests I noticed that if it is not overloaded it lets you go fast with claude and then in reality it is not a long wait, for example github copilot is eternal sometimes and makes many errors despite having models like o1. I found my balance like this, as long as the deepseek tokens are on sale I will stock up.