Introduction to AI Tools for Data Analysis: ChatGPT, Claude, and Beyond - A Short Course

ChatGPT is a particularly powerful tool for data analysis because it acts as an intelligent research assistant, translating your instructions into Python code, executing it using libraries (pandas, Matplotlib), and presenting results in an accessible format. Research shows that ChatGPT can match or exceed the performance of junior data analysts across various analytical tasks (Cheng, Li, and Bing, 2023). In fact, on DSBench, a benchmark designed to evaluate agents on realistic data science tasks spanning data analysis and modeling, the ChatGPT agent already surpasses human performance by a significant margin.

You will need access to ChatGPT and Claude for this course. A ChatGPT Plus (pricing here) and Claude Pro (pricing here) subscription is recommended, as the free versions offer more limited functionality for data analysis.

ChatGPT Plus has a message limit, so please avoid heavy ChatGPT usage before each session.

The course is designed for both newcomers to data analysis and experienced researchers interested in adding AI tools to their workflow. The emphasis will be on practical ways to use ChatGPT and other leading AI tools to explore, understand, and work with your data more efficiently.

1. Foundations of LLMs

• • Short intro to LLMs
  • Current capabilities of AI models
  • Benchmarks & pace of development
  • Comparison of leading LLMs

2. AI for data analysis

• • Introduction to ChatGPT for data analysis
  • Working with popular data formats (CSV, Excel, PDF, JSON)
  • Python libraries and execution environment
  • Comparison with traditional statistical software

3. Limitations & best practices

• • Types of AI errors
  • Verification strategies
  • Privacy and security considerations

4. Working with data

• • Importing different file formats
  • Handling missing values and outliers
  • Data cleaning and preprocessing
  • Creating summary tables and reports

5. Statistical analysis and visualization

• • Descriptive statistics
    • Means, medians, standard deviations
    • Frequency distributions
    • Group summaries
  • Data visualization
    • Histograms and density plots
    • Box plots and violin plots
    • Scatter plots and line graphs
    • Heat maps and correlation matrices
  • Advanced techniques
    • Linear regression analysis
    • Time series analysis
    • Factor analysis
    • Event studies
    • Propensity score matching

6. Real-world applications

• • Survey data analysis
  • Price prediction models
  • Cross-country comparisons
  • Time series trends
  • Event analysis
  • Customer segmentation

“I enjoyed how Nikolaev demonstrated applications, possibilities, and getting us acquainted with different platforms and their capabilities. I particularly liked the hands-on approach with data analysis, along with broader conversations around the practicality of the use and limitations of AI.”
  Ankur Srivastava, University of North Carolina

“Boris Nikolaev is a superb teacher. His pacing, the case studies, and the attention he pays to the needs of his students are great. All contribute to an educational experience that I rarely encounter. In my career, I’ve sat through many hours of graduate and medical school lectures. I am a retiree now, but I remain engaged in science by being a managing editor and working with junior colleagues to publish peer-reviewed papers. Boris has helped me understand what AI is and is not.”
  Toni Miles, University of Georgia

“Boris is an amazing instructor! I never felt like a minute was wasted. I enjoyed his discussion about AI’s impact on employment and industries.”
  Martín Jacinto, California State University

“The instructor showed many things that ChatGPT can do with data, especially with data visualizations. His suggestion to try them across more than one model of ChatGPT and other AI platforms to catch glitches was great.”
  Catherine Walker, Union College

“I enjoyed that we had a good space for questions. The course was held in a very fun and light way that inspired us. Also, I believe that having the datasets to try the analyses ourselves was a great approach.”
  Sarah Warkentin, Barcelona Institute for Global Health