Large Language Models (LLMs) are making waves, but can they truly handle the complexities of data science? A new study puts four leading AI assistants—Microsoft Copilot (GPT-4 Turbo), ChatGPT, Claude, and Perplexity Labs—to the test, challenging them with 100 diverse data science coding problems from the real-world interview question platform Stratascratch. These problems ranged from analytical and algorithmic challenges to visualization tasks, each designed to assess the AI's ability to generate correct, efficient, and high-quality code. The results reveal a fascinating landscape of AI capabilities. While all models performed above random chance, only ChatGPT and Claude consistently cleared a 60% success rate, demonstrating their proficiency in handling typical coding challenges. However, even these top performers stumbled when faced with more complex problems, none reaching a 70% success rate. This highlights a crucial limitation: while AI can generate impressive code, it still grapples with the nuances and deeper reasoning required for truly advanced data science. Interestingly, ChatGPT shone brightest on the hardest problems, suggesting a potential for tackling complex challenges that other models missed. The study also examined how efficiently the models generated code, looking specifically at execution times. While no statistically significant differences emerged, Claude generally produced faster-running code, while ChatGPT tended to be slower and more variable. For visualization tasks, ChatGPT again led the pack, generating visual outputs that were most similar to the expected results. This suggests a stronger capability in generating accurate and insightful visuals from data. This research provides a valuable benchmark for understanding the current state of AI in data science. While LLMs show promise for automating certain coding tasks, there's still a gap between AI-generated code and the sophisticated problem-solving skills of human data scientists. This points towards exciting future research directions, including exploring more complex real-world tasks, expanding model diversity, and refining evaluation metrics to better capture the nuances of AI-driven code generation. As AI continues to evolve, studies like this are essential for understanding its strengths, limitations, and ultimate potential to transform the landscape of data science.