Data visualization is key to understanding complex information, turning raw numbers into meaningful insights. But creating effective charts often requires specialized skills. Could Large Language Models (LLMs) bridge this gap, allowing anyone to generate charts from simple text prompts? Recent research explores this exciting possibility, while also revealing the challenges in evaluating the quality of AI-generated visualizations. Traditional methods like human reviews are costly and subjective. Simply checking if the AI’s chart matches the original data isn't enough either—a chart might be technically accurate but visually confusing. This new research proposes a clever solution: using Visual Question Answering (VQA) to assess LLM-generated charts. VQA models are trained to answer questions about images, effectively “seeing” and interpreting visual data. By asking questions like “What trend does this chart show?” or “Which category has the highest value?”, researchers can gauge both the accuracy and clarity of the AI’s visualization. The study tested two leading LLMs, OpenAI's GPT-3.5 Turbo and Meta’s Llama 3.1, using established VQA benchmarks (ChartQA and PlotQA). The results? While LLMs show promise, they’re not quite human-level chart-makers yet. AI-generated charts often lagged behind human-created ones in accuracy and readability, especially with trickier visualizations requiring complex reasoning. Interestingly, the research found that “few-shot” prompting (giving the LLM a few example prompts) boosted performance, suggesting AI can learn to create better visualizations with more guidance. However, significant gaps remain. LLMs still struggle with details like axis labels, date formats, and avoiding overlapping elements, sometimes producing visually cluttered or misleading charts. Furthermore, evaluating charts designed for humans proved harder for VQA models than those designed for tests. This research provides a crucial first step towards automated evaluation of LLM-generated visualizations. As VQA models improve, this method could streamline the development of more sophisticated AI chart-makers, ultimately empowering anyone to visualize data effectively.