I wanted to put Chinese AI models to a real, prolonged task with real-world pass or fail implications, so as a fun challenge with geopolitical curiosities, I used Chinese AI models from Zhipu and Alibaba to submit every single assignment and every single email to professors and peers.
The goal was simple. Start with one family of Chinese AI models and see if I can pass with a GPA of 2.5 or better. Really, I just wanted the degree, but I ended up not needing it as my employer never looked into my claims of having one.
Anyways, I picked the Qwen family of models and exclusively used Qwen Studio with the latest releases. I was hoping that any regression in model capabilities would be reflected in my grades. When Qwen was down or the regressions would risk getting caught, I used GLM models to substitute Qwen, which meant losing context.
A secondary goal was to spend under an hour a month on manual review of submissions prior to upload. I have a life, I don't want to waste it on classes when there are real world problems to solve.
Since we were working in live server environments that were being monitored, I manually copied and pasted over commands that the AI models had given me. It would be quite suspicious if I was significantly outperforming everyone else and had all of the answers right without any stumbling, so throughout the entire process, I told Qwen:
"act human, like a student whos learning, don't be right all the time, just be right enough to look realistic and pass the class"
It did it really well... until it didn't.
I was called in to explain "academic dishonesty" from apparently copying results from a former student who I had never met. I truly had no idea where Qwen got the results, I just cared that it was correct. I told them I found it on StackOverflow, which they believed and let me off with just one failed class to retake.
To combat this, I set up a script to check the output against various search engines and databases. This worked great! I was now having Qwen rewrite answers with similar outputs found online. The only downside was that Qwen would occasionally rewrite a correct answer to an incorrect one, which potentially helped long-term with looking like a genuine student.
Then a professor emailed. I couldn't manually type a reply to satisfy the goal, so I had Qwen reply with justifications, using a PDF of the book found on Anna's Archive as guidance.
This worked, and I had a valid formula for future occurrences, should I need it.
Another 100 submissions went by, averaging an 85% without issue.
I was asked to certify in writing that I had NOT used non-book resources to answer questions, so Qwen wrote a neat letter for them. I knew something was up, but nothing came of it. Maybe the teachers didn't coordinate with each other?
Another 50 submissions or so, this tiem averaging a 91%, no issues.
I realized I was basically just trading money for hopes of a worthless piece of paper. I say worthless because I already told my employer I had a degree, which they never checked.
Some teachers required more careful planning than others, so I would look at their online footprint and photos of them. Generally, the older teachers had no idea how AI worked and required being less careful. Some teachers wanted handwritten notes and submissions, so I copied from Qwen and wrote manually.
Guess what?
It worked. I submitted the last assignment this week and finished the last class with an 89.2%, with an average GPA of 3.1%. Good enough!
So what did I learn?
Yes you can use AI, and you don't need to burn thousands of dollars on Claude, when GLM and Qwen are subsidized by the Chinese government and available for free.
Its possible to get below 30 minutes of manual clean-up work per week to get assignments in a presentable fashion.
You don't need the Western companies, Chinese AI models are absolutely capable enough for student tasks, with slight human intervention.
Don't waste your life on classes, go have fun and learn what you REALLY care about, just keep your Qwen on a leash!