Few-Shot Chain-of-Thought prompting achieves 78.2% accuracy on chart question answering, outperforming other strategies across GPT models.

Researchers systematically evaluated four prompting strategies (Zero-Shot, Few-Shot, Zero-Shot CoT, and Few-Shot CoT) on chart-based QA tasks

Testing across GPT-3.5, GPT-4, and GPT-4o models on 1,200 ChartQA samples revealed Few-Shot Chain-of-Thought consistently delivers highest performance

Few-Shot Chain-of-Thought excels on reasoning-intensive questions while Few-Shot improves format adherence in responses

Zero-Shot prompting only performs adequately with high-capacity models on simpler tasks, suggesting limited effectiveness for complex chart analysis