The AI-assisted coding paradigm
1. The AI-assisted coding paradigm
Hi there, welcome to the course! Together, we're going to explore how AI can transform the way we write and interact with code.2. Course goals
By the end of this course, you will: Understand how AI supports coding tasks like code completion, debugging, test generation, and documentation; Feel confident choosing the right model for your coding needs. Craft effective prompts for coding purposes; and build robust pipelines to boost your coding productivity.3. The AI-assisted coding paradigm
Let’s begin by setting the stage for how Large Language Models, or LLMs, became a key part of coding assistance. Since the release of ChatGPT, LLMs have started to be used across a wide range of applications. If you’ve already tried interacting with an LLM, I’m sure you’ve been surprised by its natural language capabilities.4. The AI-assisted coding paradigm
And if you regularly write code, chances are you’ve already tried delegating some work, right? You are certainly not alone. Interest in using LLMs for software development grew rapidly from the start!5. Coding models
One of the first models for code generation was OpenAI’s Codex, which powered the initial version of GitHub Copilot.6. Coding models
As shown in this example, this allowed to interact with the model within development environments, apply its suggestions directly in the editor, and choose from the available models.7. Coding models
It helped with both code completion, such as finishing the return statement in the square() function here, and also with code generation.8. Coding models
For example, generating entire blocks of code from a problem description as done here with the square function.9. Coding models
Following Codex, Salesforce released CodeGen, one of the first open-source code generation models. The following year, OpenAI introduced updated GPT models, general-purpose models with strong coding capabilities such as improved code synthesis, bug fixing, and code explanation.10. Coding models
Like this, two major trends began. First, general models that treat coding as a core capability. Examples include OpenAI’s GPT, Claude from Anthropic, and Gemini from Google. These are typically proprietary and closed-source models.11. Coding models
Second, code-specialized models, often derived from general-purpose ones. Notable open-source examples include CodeLLaMA, based on Meta’s LLaMA and DeepSeek-Coder from the DeepSeek model.12. Choosing the model
Among all these models, you might be wondering: which one is best for my task? It depends on the type of work.13. Choosing the model
Imagine you’re working on a complex project, like debugging across multiple files. Here, you’ll want a model with strong reasoning abilities — the kind trained to analyze and simulate human-like thinking.14. Choosing the model
Now suppose your task is simpler, like quickly fixing a small function or syntax error. In this case, speed becomes more important, so smaller models optimized for fast responses are a better fit.15. Choosing the model
Finally, for everyday coding tasks — writing functions, generating documentation, or reviewing errors — general-purpose models provide a solid balance of capability and reliability.16. Let's practice!
Now that we’ve reviewed several coding models, let’s take a moment to make sure everything is crystal clear!Create Your Free Account
or
By continuing, you accept our Terms of Use, our Privacy Policy and that your data is stored in the USA.