I'm a use any tool that helps you kind of person, so I like GPT-4o and Claude for coding tasks, but open models are getting much better. Problem is models like GPT and Claude require an internet connection. Open models like LLaMA 3.2 does not, which confers benefits for employees at companies looking to increase their productivity.
Here are several Leaderboards measuring code model performance:
1. https://prollm.toqan.ai/leaderboard/coding-assistant
2. https://aider.chat/docs/leaderboards/
3. https://evalplus.github.io/leaderboard.html
4. https://huggingface.co/spaces/bigcode/bigcode-models-leaderboard
5. https://huggingface.co/spaces/mike-ravkine/can-ai-code-results
Side by Side Model Comparison: https://lmarena.ai/
The amazing thing here is that you can use models using ollama and Continue for VSCode. That is private local inference for code tasks whenever, wherever. Given policies that ban leaking "secrets" private local inference is a way to get past those policies and adopt AI models for everyday work or homework 😇.
Obviously learning how to use AI is important, but learning about what it is generating is also VERY important. Copy pasting can work, but not always. Doing code review or review in general of whatever it spits out is your job or you will be bitten like the many other fools (lawyers, students, programmers, authors, etc.) who did not. REVIEW IS THE LEAST JOB YOU CAN DO. Using AI as an assistant and being the EDITOR, CURATOR is the best job you can do. Don't be a fool.
Ollama:Â https://github.com/ollama/ollama
Open WebUI for Easy LLM Interaction: https://www.openwebui.com
Open models keep closed models pushing ahead which feeds back like the neural network architecture of the models. 🚀