@Phoenixyin13: This kind of assessment method where students create questions that stump AI is indeed very innovative and highly forward-looking. Students need to explore the strengths and weaknesses of the three models: Claude, DeepSeek, and MiniMax. In this process, students no longer blindly trust AI outputs but learn to review AI responses with a critical and discerning eye, which...
Summary
This educational assessment method encourages students to explore the strengths and weaknesses of Claude, DeepSeek, and MiniMax, creating questions that defeat AI, thereby cultivating critical thinking and competitiveness needed in the AI era.
View Cached Full Text
Cached at: 07/01/26, 01:57 AM
This kind of student-generated challenge that stumps AI is indeed a novel and highly forward-looking assessment approach.
Students need to explore the strengths and weaknesses of three models: Claude, DeepSeek, and MiniMax. In this process, students no longer blindly trust AI outputs—they learn to review and verify AI responses with a critical eye. This is precisely the kind of competitive edge that is rare in the AI era.
In the future of education, the focus will no longer be on who possesses more known knowledge. https://t.co/vibnigJMVs
Similar Articles
@Fenng: AI little princes are shocked again... This problem has always existed. Using question banks like '作业帮' directly to search for answers (such tools have already evolved into AI), handing over thinking to tools, the inevitable result is a decline in one's own abilities. What's surprising about that?
This tweet discusses a study on students using AI to assist with homework, pointing out that while short-term grades improve, long-term exam scores decline, prompting reflection on the role of AI in education.
@jinchenma_ai: I think a very important approach to learning AI right now is to use AI to become your personal coach. Many people now try to use AI to help with work or answer questions. But going further, we can also use AI to observe ourselves in reverse and uncover more value for our own growth. Over the long term...
The author proposes using AI as a personal coach, discovering cognitive blind spots and strengths through regular review of interactions with AI, to achieve continuous growth.
@akokoi1: The first skill of this AI teaching tool, "3D Geometry Problem Solving," has been released, supporting text-based problem solving, image-based problem solving, and random question generation. After testing mainstream models, deepseek-v4-pro offers the best cost-performance ratio at 0.2-0.3 yuan per question, with accuracy on some questions even surpassing GPT5.5.
The first skill of the AI teaching tool, "3D Geometry Problem Solving," has been released, supporting text-based problem solving, image-based problem solving, and random question generation. Tests show that deepseek-v4-pro offers high cost-performance, with accuracy on some questions even surpassing GPT5.5.
@FudanUniversity: Final exam at Fudan: students don't answer questions. They write them — to stump AI. 51 students, 10 questions each, 3 …
Fudan University held a novel final exam where 51 students each wrote 10 questions designed to stump three AI models (Claude, DeepSeek, MiniMax), with grades based on how difficult the questions were for the AI.
@Aoyi21: https://x.com/Aoyi21/status/2064975015200829457
This article proposes that the most cost-effective way to learn AI is to deconstruct others' Skills. By analyzing their task definitions, trigger conditions, operation steps, prohibitions, and acceptance criteria, you can learn how experts think and train AI, rather than just using tools.