Presenter: Reuben Thiessen, Emerging Technology Lead, Stanford Accelerator for Learning
Recording of the Session
Download Chatbot Design Framework Worksheet
Central Questions:
- How could we build chatbots that are both flexible and predictable in teaching and learning contexts?
- Which techniques can make interactions feel unique and personal?
- What strategies can help your chatbot consistently deliver high-quality responses?
Key Quotes:
- “It’s not like Google search where you just kind of throw a couple of keywords in. But it’s more around being as specific as possible.” – Reuben (50:19)
- “You can model out an actual conversation as part of your system prompt, and it uses that as a [guide to understand], ‘okay, that’s how you want me to respond.'” – Reuben (58:19)
- “Continual refinement [is essential for creating a chatbot] that really suits your needs.” – Reuben (47:14)
Takeaways:
- A system prompt is the key part of every generative AI chatbot; it is a set of instructions that guides chatbot responses, ensuring interactions are relevant and aligned with your design goals.
- A carefully designed system prompt answers these questions:
- Who should the chatbot act like?
- Who are some potential users of this chatbot?
- What kind of posture and tone should the chatbot take?
- Can you constrain the scope, and place some limitations on the interactions?
- Are there any behavioral guardrails you’d like to put in place?
- User Start Messages are example sentences that are a great way to model to your users how they can most effectively interact with your chatbot.
- Hugging Face Assistants offers a free and easy way to get started designing a custom system prompt to power your own chatbot.