The Degree Thesis System Is Dead: Zhao Bin's First-Principles Takedown of the "AI-Detection Farce"
> The original purpose of the degree thesis was to evaluate independent research ability. In reality, it evaluates typing speed and formatting compliance. > In the AI era, the system itself rewards cheating.
1. Original Intent vs. Reality
The degree thesis system was born from a simple purpose: assessing whether a student possesses independent research ability. That purpose has never changed — what changed is how the system executes it. Longer theses are deemed better. Defenses have become a formality. Typing is equated with academic integrity. The system grows ever more refined in form and ever more hollow in substance.
Professor Zhao Bin of Fudan University dismantles this using Musk-style first principles: return to basic facts and reason upward.
The most basic facts:
- About 90% of the workload of an undergraduate thesis is meaningless. Students are not solving real problems; they are completing a set of formatting specifications — font size, line spacing, citation format, page numbering — consuming enormous time with no relation to research ability.
- Writing honestly requires 300 hours. With AI assistance, a thesis can be finished in 10 hours and pass more easily.
- A published paper
- An open-source codebase
- A completed entrepreneurship project
- A consulting report solving a real problem
- Even a failed experiment, if it proves the student went through a full research process and learned from it
- Universities know AI detection is unreliable — false positive rates are non-trivial and evasion methods keep emerging.
- Students know it — rephrasing lowers detection rates.
- Detection vendors know it — they sell reassurance, not truth.
- Zhao Bin's related commentary (specific source to be added by the user)
The system punishes the honest and rewards cheaters. This is a structural inversion.
2. Distortion One: Institutional Laziness
"Longer means better" is the classic lazy metric. The core of research ability is the capacity to pose and solve problems; thesis length has no necessary connection to it. Einstein's 1905 miracle-year papers were mostly short yet rewrote physics, while some modern doctoral theses run hundreds of pages with contributions amounting to swapping datasets in an existing framework. The system counts words because counting is easy: reviewers need not genuinely understand the research, only check page numbers and chapter structure. Systems gravitate toward quantifiable metrics even when those metrics are unrelated to research value.
Defenses as a formality. Ideally a defense is rigorous peer questioning; in practice committee members often lack time to read theses carefully. They skim abstracts, glance at tables of contents, and ask perfunctory questions while students recite prepared answers. The root cause is not reviewer attitude but institutional design: a professor may sit on dozens of defenses a year, and deep reading of each is unaffordable in time. The system mandates the defense without providing the resources — the inevitable result is formalism.
"Handwriting equals academic integrity" is the most absurd assumption. Some institutions require handwritten drafts or on-the-spot handwritten abstracts, reasoning that handwriting prevents AI ghostwriting and proves personal authorship. The flaw is obvious: handwriting has nothing to do with research ability. A good calligrapher is not necessarily a good researcher. The deeper problem: the system has made "detecting AI" an end in itself. Plagiarism rates, AI-detection rates, handwriting verification — these were meant to be means serving the goal of evaluating real ability. Instead the means have kidnapped the goal. Students now think not about making research more solid, but about passing detection.
3. Distortion Two: The System Rewards Cheating
Zhao Bin's arithmetic: honest writing takes 300 hours; AI-assisted writing takes 10 hours and passes more easily. That 30x gap is not a technology gap — it is a loophole in the system.
Honest students spend vast time on literature review, experiment design, data analysis, and revision. Students using AI type prompts and let the model generate everything — often producing more "standardized" results, since AI knows thesis templates and what reviewers like to read.
The paradox: the system claims to evaluate independent research ability while in practice rewarding non-independent completion. This is not a matter of student morality but of system structure. When a system is designed so cheating is more efficient and more likely to succeed than honesty, blaming individuals evades systemic responsibility.
Economics has the concept of incentive compatibility: good institutions align individual interest with social interest. The thesis system does the opposite — personal interest (graduating fast) diverges sharply from the stated goal (evaluating research ability).
4. Distortion Three: Abolish the Thesis, Adopt an Outcomes-Based System
Zhao Bin's conclusion: abolish undergraduate and master's thesis requirements and replace them with an outcomes-based system. Seemingly radical, it follows directly from first principles: if the thesis's purpose is to evaluate independent research ability, any form that demonstrates that ability should be accepted.
Under an outcomes-based system, graduation depends not on word count but on what problem you solved:
The core is "problem-driven": the student picks a real problem, solves it by any legitimate means, and presents the process and result. The fundamental difference: the thesis system is process-oriented (complete a fixed pipeline); the outcomes system is results-oriented (prove you have the ability).
5. The Purpose of Education in the AI Era
Zhao Bin's final thesis: the AI era does not need typists who can write theses; it needs creators who can solve problems.
The thesis system's flaw is not that it is old, but that its founding assumption no longer holds. The traditional assumption: writing ability = research ability. AI has broken this equation — AI writes structurally complete, fluent, properly cited theses, outperforming humans in format and speed. If a thesis's value lies mainly in "being written," AI has already won.
But AI cannot produce genuine problem consciousness. It handles existing problems but struggles to discover undefined ones; it optimizes known paths but rarely opens new directions. That is precisely what education should cultivate: the ability to define problems, choose methods, and move forward under uncertainty. These are evaluated not by "writing a thesis" but by "solving a real problem."
Zhao Bin's view is not anti-technology but a repositioning of it. AI is a tool, not a rival. A student who does not know the research purpose produces only refined garbage with AI; one who understands the problem can compress 300 hours to 10 and spend the saved time on more important thinking.
6. The Essence of the "AI-Detection Farce"
Zhao Bin calls the current AI-detection wave a "farce" — everyone knows it is a performance, yet everyone cooperates:
But the system mandates checking, so everyone checks. Afterwards, no one believes the results, yet everyone pretends to.
Behind this lies institutional inertia: facing the AI shock, the easiest response is "strengthen detection" — minimal design change, minimal training, minimal political risk, and maximal futility, since detection forever chases evasion. The harder but correct response is redesigning evaluation: if AI can write theses, "writing a thesis" should not be the core metric; if AI can generate code, "writing code" should not be either. The anchor of evaluation must move to what AI cannot do: problem definition, value judgment, ethical trade-offs, creative breakthroughs.
Conclusion
Zhao Bin's first-principles demolition yields a simple conclusion: return to the original purpose, abolish the formalism, replace theses with outcomes. The execution is complex — the system has run for decades, involving millions of students, tens of thousands of reviewers, and thousands of institutions. But "hard" does not mean "should not be done."
The basic facts do not disappear because change is difficult; they only worsen if nothing changes. AI-era education needs to cultivate people who can define problems, judge value, and move forward through uncertainty. Those are what the system should evaluate.
---
Reference