Question: Compare and contrast why human-level performance is a weaker baseline for tasks like stock prediction versus tasks where humans excel (like image labeling). Use the reasons given in the source in your explanation.

Sample answer: For tasks like stock prediction, humans are not good at the task itself, so their performance offers little useful benchmark. This creates three problems: labels are harder to obtain because human labelers cannot reliably identify the 'optimal' answer; human intuition is harder to rely on for diagnosing errors, since 'pretty much no one can predict the stock market'; and it is hard to know what error rate is optimal or reasonably achievable. In contrast, for tasks where humans excel, their performance gives a strong, informative benchmark for labels, intuition, and target error rates.

Key points:

• Labels are harder to obtain when humans aren't skilled at the task
• Human intuition is harder to count on for error diagnosis
• It is hard to know the optimal or reasonable desired error rate
• Contrast with tasks where humans perform well, giving a stronger benchmark

Rubric: Full credit: identifies all three problems (labels, intuition, optimal error rate) and explains why human skill at the task determines baseline usefulness. Partial credit: identifies one or two problems or lacks the comparison to human-strong tasks.