New Year Reflection: Learning from GPT

I have used GPT Pro fairly extensively over the past year for different things. When I asked GPT to summarize my usage patterns, more than a dozen came up. Most notable among them are curriculum design, lab development, technical explanation, and critique. The sheer volume of output made it impossible to pretend I was just using GPT for little edits but as a drafting engine, a reviewer, and in many cases an argument partner.

Early on, I made predictable mistakes. I initially mistook fluent text for finished thinking, and was happy with merely editing generated drafts. Fortunately, I quickly noticed how my voice was drifting toward generic professional writing even with careful editing. As I realized that I must audit AI output, I began making changes in my working habits with GPT. Most importantly, I have learned to be very intentional in my usage and to ensure that I use GPT to improve my productivity rather than to replace my thinking.

This essay reflects on my more significant usage patterns and how they have gradually shaped my working habits.

My GPT usage

I used GPT for a lot of things. These are the most notable.

Lab and assignment designer: This category has the highest estimated usage. No, I did not use this to guard against students’ AI-generated submission. Instead, I have used GPT to do the following:

1. Augment existing assignments through identifying potential logical loopholes and cleaning up ambiguity in assignment descriptions.
2. Create new hands-on labs with concrete steps and deliverables, then extended them into follow-up versions to create progression instead of one-off demos.

For (2), the prompts have to be explicit regarding the number of steps, acceptable tools/libraries to be used in the lab. Once the lab is generated, it must be customized to fit with the course’s computing environment. Descriptions of steps will also need to be elaborated and have more details added.

Curriculum architect: I also have GPT analyze and provide feedback on my existing and upcoming syllabi. I want to ensure that current courses have relevant and updated contents while new courses are built upon solid big ideas. At the same time, these contents must be crafted into a teachable course structure, including weekly topics, learning outcomes scaffolding, assessments, and readings. GPT can provide general suggestions, but I need to make sure that everything fits together not only within a course but also across existing prerequisites across our entire curriculum.

Critical reviewer: I have GPT act like that skeptical second reviewer that calls out weak logic, missing evidence, unclear claims, and any other constructive criticisms. This helps me with rewriting and/or restructuring my papers so the argument holds up under unfriendly reading. I actually like this a lot. To this extent, I also use the free versions of other AI providers (Gemini, DeepSeek, Grok, etc.) to cross check my contents against GPT’s reviews.

General utility: I use GPT as a Swiss Army knife in this setting. One day I could be discussing values and identity questions. Another day I would ask GPT to explore new ideas and to find out whether others have worked on these ideas before. In many cases, I have asked GPT to work on mundane things like making sure that forms and documents follow proper templates and institutional structures.

Changes in my writing habits

When I first started using GPT extensively, I was quite fascinated with its ability to generate lengthy written text with complicated words. However, eventually I have come to the realization that it feels like eating cotton candy! It smells great and looks delicious, but once you bite down, it is mainly air.

Simpler writing: The very first thing I learned from working with GPT is how to simplify my writing. I try to use shorter sentences and simpler words. If I have to write long sentences or use big words, I need to be sure that I really mean it.

Longer writing: I write more now because I’ve seen how effortlessly AI can generate competent, generic prose. Once you’ve watched a machine produce a dozen reasonable writing alternatives in a few minutes, you can’t help but feeling worried and compelled to become better. Hence more writing and longer writing. I sincerely believe that after one year my writing has improved, and a big part of this comes from my motivation to be better and more personal than my AI companions. I am still running my final draft through GPT for critiques, but the writings are now truly mine.

Changes in my coding habits

I have seen plenty of vibe coding success stories online where people building apps with almost no manual coding. My own experience with GPT-generated code has been more mixed, especially for large or complex projects. I have devised my own approach as follows.

Generate what I know: This is purely for productivity. I use GPT to generate codes that I already know how to do. GPT is great for simple and lengthy boilerplates codes targeting well understood problems and templates. Due to my own understanding, I am confident in how accurate my prompts can describe the problems and my ability to debug the final generated products.

Vibe what I don’t know (to learn): I vibe code not to generate product but to learn. I’m using this process to skip the slow ramp-up and get straight into the part where learning actually happens: diagnosing problems, fixing edge cases, customizing structure, and making decisions under constraints. In other words, I don’t learn by watching the tool perform but wrestling with the tool’s mistakes and adjusting the system to meet my standards. I will still need to go back and go through the fundamentals, of course, but revisiting the fundamentals after already knowing some working aspects is always more productive, at least for me.

Forward looking statement!

This essay listed (most) of the ways I have used AI and reflect on what that did to my habits. I don’t think any of these lessons are permanent, because the tools are evolving fast and I’m still recalibrating to find a balance between productivity and originality.