Design teachable ChatGPT sessions with purposeful errors or lower-level models so learners practice error correction¶

Strategy · All strategies
Evidence · no claims cited

Description¶

For error-correction skill, the article recommends that "incorporating scenarios where learners must identify and correct errors, such as purposeful mistakes by the AI, can be beneficial", and that designers focus "more on lower-level ChatGPT, which may introduce more opportunities for students to correct errors". Pedagogical design should target the learning processes stimulated, not solely the accuracy of AI-generated content.

Design Implications¶

Context¶

Requirements¶

• Deliberately imperfect agent responses that learners can detect and correct during code generation

Constraints¶

• Derived from a null correctness result; the suggestion that GPT-3.5 would work better than GPT-4 is the authors' speculation, not a tested comparison

Target Learners¶

• university students learning programming

Target Learning Goals¶

• error-correction and debugging skill
• judging the correctness of generated code

Related Strategies¶

• Apply error mitigation such as self-consistency before deploying LLM-generated help, and frame unmitigated LLM feedback as an imperfect source
• Introduce prompt-engineering training before students use Gen AI for structured self-study techniques

Examples¶

-

Key Sources¶

• Angxuan Chen, Yuang Wei, Huixiao Le, Yan Zhang. (2024). Learning-by-Teaching with ChatGPT: The Effect of Teachable ChatGPT Agent on Programming Education. arXiv. https://arxiv.org/abs/2412.15226