Visual Instruction can involve which model types?

• A. Watch a single expert
• B. Watch a novice
• C. Watch neither
• D. Watch an expert, a novice, or both

Answer: (D)

Visual Instruction learns from demonstrations shown in visual form. It can use demonstrations from an expert, a novice, or both. Using only an expert shows the best way to perform tasks, but it can miss the variety and mistakes you’d see in real-world use. Using only a novice reveals common errors, but it can teach suboptimal strategies. Bringing together both sources gives the model a richer training signal: it learns to imitate high-quality behavior while also understanding common mistakes and variations, which improves generalization and robustness across different scenarios.