How can we use AI
(responsibly)
in Data Science?
HANDS-ON TUTORIAL WITH INSIGHTS
Department of Statistical Sciences Workshop Series
March 5, 2026
Preliminaries
What is a Large Language Model (LLM)?
- A type of AI trained on massive text datasets
- Uses deep learning (transformers) to understand and generate language
- Can answer questions, summarize, and write code
- Learns statistical patterns, not facts or reasoning
- Outputs are probabilistic, not deterministic
- Powers tools like ChatGPT, Gemini, Copilot, and more
It all starts with statistics!
Source: https://shiny.posit.co/blog/posts/shiny-side-of-llms-part-1/
Intended audience
- Statisticians using R and/or Python
- You’ve used LLMs via ChatGPT, Gemini, or similar
- You haven’t used LLMs integrated into an IDE
- You haven’t used LLMs from code
- You haven’t performed EDA supported by an LLM
Initial disclaimers
- When it comes to data science, today’s large language models are remarkably capable, yet they are not infallible
- However, the promise of AI for data science is too great to ignore
Aknowlegments
- Much of what you will see is inspired by notes, videos, slides etc
- This is my personal and biased view, based on experience
- The field is evolving rapidly, so information may quickly become incomplete or outdated
- Non-exhaustive list of very interesting resources
Getting Started
Sign up for a developer account with Anthropic or OpenAI and grab an API key (requires credit card)
- Just do it already 🙄
Sign up for GitHub Education (free for faculty!) to get GitHub Copilot Pro at no charge
Install an LLM client package
- For R:
install.packages("ellmer") - For Python:
pip install chatlas
- For R:
A warning
- On their own, LLMs are terrible at analyzing data
- They cannot do simple arithmetic reliably
- They cannot even count rows reliably
- Having LLMs write code that analyzes data (and sometimes, even execute it) is the way