Large language models have proven to be highly effective programming assistants, and GPT-4 has achieved human-level performance on coding challenges. I will present some analyses of GPT-4 coding performance in the context of data science and statistics problems, and demonstrate its potential utility to improve programmer effectiveness as well as its shortcomings. I will also discuss ethical and educational issues surrounding the use of these models for coding and highlight a road forward towards effective use of these tools in service of rigorous research.