PromptDecipher: AI Tool for Editable Tutor Chatbot Authoring
A new system called PromptDecipher has been created by researchers to streamline the development of AI tutoring chatbots. This innovation enables educators to modify simulated conversations instead of crafting abstract prompts. A preliminary study indicated that nearly all teachers do not systematically evaluate their bots prior to launch. PromptDecipher transforms the authoring process through direct interaction: educators can edit undesirable chatbot responses in a live chat preview, while an automated system analyzes these corrections, suggests specific prompt adjustments, and verifies changes against established scenarios. This solution bridges the gap between the ideal roles of educators as designers and the reality of their limited involvement in these capacities. The findings are detailed in arXiv paper 2605.16605.
Key facts
- PromptDecipher is a system for authoring AI tutoring chatbots.
- It uses direct correction-based interaction instead of writing system prompts.
- A formative study found that virtually no teachers systematically tested their bots.
- Teachers edit undesirable bot responses in a live chat preview.
- An automated pipeline proposes targeted system prompt rewrites.
- The system validates changes across predefined scenarios.
- The research is published on arXiv with ID 2605.16605.
- The work addresses the gap in educator roles for chatbot authorship.
Entities
Institutions
- arXiv