Presenters: Cathy Chase, Senior Research Scholar, Stanford Accelerator for Learning; Reuben Thiessen, Project Management Specialist, Stanford Accelerator for Learning

Recording of the Session

Central Questions:

• What strategies can be employed to prompt AI models for optimal feedback across various tasks?
• How can we ensure that AI-generated feedback is both meaningful and actionable, leading to real improvements in learning outcomes?

Key Quotes:

“Feedback is really a necessary part of any complex learning…There’s no instruction so flawless that people are going to learn something perfectly without feedback.” – Cathy Chase (3:43) 

“Effective feedback is specific, timely, understandable, non-threatening, and revisable.” – Cathy Chase (7:14)

“[ChatGPT] will tend to default to explanatory feedback. It often tries to sneak this in even if you don’t ask for it, so [watch out for] that.” – Reuben Thiessen (19:25)

Takeaways:

There are various types of feedback that are useful for different contexts:

• Reinforcement – Rewards positive behaviors
• Explanatory – explains what is good/bad and how to improve
• Model-based – provides examples to emulate
• Reflective – encourages learner self-assessment and reflection
• Discrepancy – highlights gap between performance and target

When prompting ChatGPT for feedback:

• Ask for the specific type of feedback you want, and make sure to define it
• Scope feedback to 1-2 focus areas
• Be explicit about quantity of feedback wanted
• Personalize for audience (age, background, etc.)