Claude 3.5 Sonnet is uncanny in many ways. Code is outstanding. Even just brainstorming projects is enjoyable with Claude. Really, it's like talking with another human. Thanks for the video, this is the type of video that is very informative. You don't see it on other channels.
Claude Opus 3 and Sonnet 3.5 have superior reasoning ability which makes them superior at performing the most difficult tasks in any domain. DeepSeek Coder V2 is comparably good but only in the coding domain. All DeepSeek models have in common that they were trained using the "chain of thought" prompting technique. When using any DeepSeek model, one should therefore adhere to this paradigm and should not use DeepSeek models for creative writing because that goes against the chain of thought paradigm.
@@amerrashed6287 I trust the benchmark by the creators of the "Aider" programming assistant. They use evaluation data that is not contained in any major benchmark you can read about and Aider's benchmark uses "medium" to "hard" programming tasks. There are two other benchmarks (BigBench Hard, LiveCodeBench) that also avoid biased results by using test data that is not contained in training data or test data of LLMs. They achieve this by careful test data selection and test data randomization. By these methods, Aider, BigBench Hard and LiveCodeBench prevent LLMs from overfitting to specific benchmark data. The results of these 3 benchmarks show much lower numbers than the widely reported standard benchmarks to which many LLMs are overfitted. Aider's benchmark shows that Claude Sonnet 3.5 and GPT-4o are very smiliar, and Sonnet 3.5 is best used for the most complex tasks because it's reasoning ability is superior.
@@amerrashed6287 I prefer Claude over ChatGPT of any type. Every day I spend at least an hour or more brainstorming and none can brainstorm as good as Claude 3.5
Awesome video! Please do a video where you do the same project (task app) using Aider + Mistral Large 2, Aider + Claude 3.5, Aider + Chatgpt 4o. The critrria is 20 prompts to be production ready. One to two features minimum, but the goal is to get it ship ready and hosted on vercel. Hopefully that will show how many models get you to that level. Share your experience on each. You can the video a bit RAW with less effect as I know the editing takes more time to be your level of polished. Great work King!
Open Interpreter + Groq + Llama 3.1 + n8n + Gorilla AI = Lightning speed 100% autonomous agent that automates all workflows with a simple prompt, all open source and free accept inference, access to over 1600 API's.
Nice video, man! My short impression about AI coding is: 4o is drastically failing on large projects, Sonnet is also drastically failing on large projects. There is no real consistency on the AI models right now. They are like chickens without head! Not even Aider, or other tools like it help in that matter. I have tested them a lot and spent a lot of money on the APIs, but no... Not working. Even with context they tend to forget what have they done previously and change the code chaotically - like they don't remember. The tools like Aider are great but unfortunately the problems relies on the models. They maybe will get better in time but for now, we are still the brain and memory.
Claude 3.5 Sonnet is uncanny in many ways. Code is outstanding. Even just brainstorming projects is enjoyable with Claude. Really, it's like talking with another human.
Thanks for the video, this is the type of video that is very informative. You don't see it on other channels.
Honestly i love claude
Claude Opus 3 and Sonnet 3.5 have superior reasoning ability which makes them superior at performing the most difficult tasks in any domain. DeepSeek Coder V2 is comparably good but only in the coding domain. All DeepSeek models have in common that they were trained using the "chain of thought" prompting technique. When using any DeepSeek model, one should therefore adhere to this paradigm and should not use DeepSeek models for creative writing because that goes against the chain of thought paradigm.
Is it better than chatgpt 4o?!
@@amerrashed6287 I trust the benchmark by the creators of the "Aider" programming assistant. They use evaluation data that is not contained in any major benchmark you can read about and Aider's benchmark uses "medium" to "hard" programming tasks. There are two other benchmarks (BigBench Hard, LiveCodeBench) that also avoid biased results by using test data that is not contained in training data or test data of LLMs. They achieve this by careful test data selection and test data randomization. By these methods, Aider, BigBench Hard and LiveCodeBench prevent LLMs from overfitting to specific benchmark data. The results of these 3 benchmarks show much lower numbers than the widely reported standard benchmarks to which many LLMs are overfitted.
Aider's benchmark shows that Claude Sonnet 3.5 and GPT-4o are very smiliar, and Sonnet 3.5 is best used for the most complex tasks because it's reasoning ability is superior.
@@amerrashed6287 I prefer Claude over ChatGPT of any type. Every day I spend at least an hour or more brainstorming and none can brainstorm as good as Claude 3.5
Can you try to create this same app in claude 3 and deepseek coder v2???
Also tell us in how many prompts you maked.
Awesome video! Please do a video where you do the same project (task app) using Aider + Mistral Large 2, Aider + Claude 3.5, Aider + Chatgpt 4o. The critrria is 20 prompts to be production ready. One to two features minimum, but the goal is to get it ship ready and hosted on vercel. Hopefully that will show how many models get you to that level. Share your experience on each. You can the video a bit RAW with less effect as I know the editing takes more time to be your level of polished. Great work King!
Ooo you should try fine tuning llama 3.1 8b for coding
Would love to see how to do this!! @AICODEKING
Bro, I love your videos. Continue.
Agree.. 6:40
Claude is more better than Llama 405b. i hope deepseek will launch better model for coding after this llama 405b version release.
Open Interpreter + Groq + Llama 3.1 + n8n + Gorilla AI = Lightning speed 100% autonomous agent that automates all workflows with a simple prompt, all open source and free accept inference, access to over 1600 API's.
Any chance we can learn more about this?
@@zheng2hao first check out Open Interpreter
i am begining to like Claude 3.5 Sonnet . It brainstormed ideas on how to troubleshoot an error. Also it sounds more humble while chatting.
I love your content and style
Thanks for the cool video! I'm curious regarding the differences between maestro and Aider. How about make a video comparing those two tools?
Nice video, man! My short impression about AI coding is: 4o is drastically failing on large projects, Sonnet is also drastically failing on large projects. There is no real consistency on the AI models right now. They are like chickens without head! Not even Aider, or other tools like it help in that matter. I have tested them a lot and spent a lot of money on the APIs, but no... Not working. Even with context they tend to forget what have they done previously and change the code chaotically - like they don't remember. The tools like Aider are great but unfortunately the problems relies on the models. They maybe will get better in time but for now, we are still the brain and memory.
As a dev i agree with you 100%
maybe it's because from the ollama, it's just Q4 models, so f16 version may have different result, I think (?!)
I wonder if fine tuning it with the most recent Next.js documentation, examples and relevant blog posts would help.
Bro can u make video (aider - deep seek - nextjs ) to complete application
Could you test and see which model is better at C++ please brother king. Add in JUCE framework for happy time
Supabase ssr, not all models trained on the new complicated ssr method
Can i know the asset you use to run large language models like 3.1 405b?? Please share with specifications and configuration
woooaw you fast bro
Try complex buildingb apps but you are the best also I think deepseek 2 is better in coding and Claude
Have you tried Aider with Flutter?
Good Job thanks. "Make it more cool" :P
Can you do one recreating partiful app?
Can you use aider for android development for axample with flutter or react native
Thank you 🙏
Pleaseeee write your prompts.. Please... Thank you very much. Talent Management / Talenst Screener with AIDER + OLLAMA/OPENAI + SUPABASE... thank youuu🎉
This is not for windows users because i didn't see how to setup
Have you messed with Pretzel AI for Jupyter notebooks?
Not yet.
it's easy mkay...
amazing!! great work😍
Which ai we can use for visionOS
What do you mean? Any of them if you connect the vision to a mac
@@gorgeuos claude and chatgpt answers are not useful for visionOS. Getting lots of errors
@@keyunsimejiya4945 how do you use it ?
This is the issue with llama it makes a mess of everything.
Cool, but can you make it more cool?