Generative AI: A Tool Beyond Human Cognition
"The emergence of ChatGPT completely astounded me. Upon testing it, I found it exceeded my expectations—operating at a level far beyond what I had conceived AI to be. This is a tool we must embrace."
When ChatGPT debuted in November 2022, its unprecedented capabilities made Professor Lan realize that embracing the technology was essential. He immediately encouraged his students to adopt it, even reviewing their prompt histories during lab meetings to diagnose why certain queries yielded inadequate responses.
Beyond coursework, Professor Lan asks students to utilize ChatGPT in thesis writing. He notes that while the core research process requires human insight and rigor, ChatGPT’s Chinese prose often surpasses that of student writing. Through experimentation, he also discovered that constraining ChatGPT within specific parameters effectively mitigates hallucinations and controls generation quality.
Exploring Diverse Platforms for Specialized Applications
"Every generative AI platform has its own strengths and weaknesses, so it’s impractical to rely exclusively on just one."
Driven by a penchant for exploring new technology, Professor Lan has experimented with a wide array of tools beyond ChatGPT, including NotebookLM, Claude, Gamma, Grok, Gemini, and Cursor. While he does not maintain paid subscriptions for every service, prolonged usage across platforms has enabled him to anticipate output quality and understand their respective traits.
In highlighting these nuances, Professor Lan points out that while ChatGPT excels at flexible data analysis and broad contextual synthesizing, it lacks long-term context retention and requires repeated input. Conversely, NotebookLM offers meticulous analysis of user-provided sources but remains strictly bounded by those documents.
Understanding these trade-offs allows him to deploy each tool strategically: using NotebookLM for literature reviews, Grok for deep Web search, and Claude for code generation.
Instructional Integration: Enriching Curricula Through Human-AI Dialogue
"Ideally, I would love to build a course curriculum designed from the ground up to co-exist with ChatGPT... Guiding students through a three-hour session on how to effectively leverage ChatGPT equips them with skills essential for their future careers."
Citing his course Data Analysis Methods, Professor Lan uploads course materials, handouts, and slides into ChatGPT before submitting assignment and exam questions for evaluation, requiring the AI to substantiate its feedback. The platform highlights missing information and suggests real-world problem scenarios. This iterative dialogue uncovers novel perspectives, enriching the curriculum. Occasionally, he compiles these AI interactions into supplementary reading materials for students.
However, Professor Lan notes that theoretical courses involving pure mathematical derivations are less suitable for AI integration. Beyond the standardized nature of such content, AI often fails to resolve novel or non-existent proofs. Furthermore, students currently lack the domain expertise required to verify the accuracy of AI-generated derivations, making direct professor-student interaction indispensable.
Professor Lan previously experimented with testing students on prompt engineering during exams, but concluded that single-instance assessments were unfair. Effective prompting is rarely a one-shot endeavor; it demands continuous iteration and refinement. To hone his own practice, he maintains a dedicated prompt repository, continuously revising his queries to achieve higher precision.