Designing course materials is very boring. When I felt tired, I suddenly thought: Why not play a game with DeepSeek?
My game rules were simple: Ask DeepSeek to send me an English sentence each time. I would identify the core action (the verb/predicate). The winner would be decided by three rounds, best two out of three.
Pay attention now—here’s where things get interesting.
After I told DeepSeek my request, it agreed immediately. Of course, it still “deeply thought” for 20 seconds. During those 20 seconds, it analyzed my intentions and possible mistakes I might make. Finally, it even reminded itself:
“Don’t design sentences that are too hard. Avoid structures that could trick the user.”
Seeing its “deep thinking” process, I almost laughed. This was either extreme confidence—like playing with cards on the table—or just trying to make users happy. Does DeepSeek’s programming have performance reviews?
I decided to ignore my curiosity and see what sentences it would give me.
Round 1: It threw out a simple sentence, roughly CEFR A1 level. Probably testing my grammar first before raising the difficulty:
The curious child carefully examined the ancient map under the dim light.
This couldn’t stump me. Strangely, even though my answer was correct, DeepSeek insisted I had a spelling error. I checked twice—my spelling was fine.
DeepSeek must have hallucinated.
Round 2:
“Despite the heavy rain, the determined hikers continued their journey toward the mountain peak.”
Look at this sentence—the difficulty didn’t increase at all.
Round 3:
“Under the guidance of her mentor, the young scientist successfully synthesized a groundbreaking compound in the lab.”
Slightly harder, reaching CEFR A2-B1 level. But any high school student could handle this.
Finally, DeepSeek congratulated me for winning. But I felt its questions were too casual, and it cared too much about my feelings.
So I challenged it: “Dare to give me a super hard sentence? Same rules. One round decides the winner. Don’t show your thinking process. Don’t worry about my feelings.”
This time, I truly laughed. Guess what DeepSeek did during its “deep thinking”? It accidentally listed the answer to the sentence it planned to give me.
I was speechless. This shows English might be relatively simple. Of course, maybe that’s a hasty conclusion—I didn’t ask for “hell-level difficulty” sentences. I’ll try that next time.
But before starting “hell-level” English tests, I checked DeepSeek’s Chinese language analysis skills. The details will be in my next blog, but here’s a spoiler:
Its Chinese semantic analysis is already at high school level.