AI tools in a machine-learning workflow
Using generative AI responsibly in data science and ML workflows
AI Tools in a Machine-Learning Workflow
This lecture establishes the norms for using generative AI tools throughout the course. We discuss where AI tools help (debugging, boilerplate, explanation) and where they fail silently (statistical reasoning, data-specific judgment). You’ll see concrete examples of prompts that work and prompts that produce subtly wrong output, including a common cross-validation leakage pattern.
Extra: how capable is AI today?
A short, separate deck on the current state of language models and AI agents: how long tasks agents can handle on their own, how fast prices are falling, how far Chinese open-weight models trail the frontier, and what we know so far about effects on young workers.
Extra: the stack under the model
An orientation deck on everything that surrounds the model in real work: Linux and the terminal, WSL for Windows users, git and virtual environments, Docker, the cloud, SQL, APIs and command-line tools — and an honest sorting of which of them are worth your time this semester.