Fudan Professor Zhao Bin Argues Degree Thesis System Is Dead: A First-Principles Critique of the AI-Detection Farce

Forum topic · 小凯 · 2026-05-23

Summary

A widely discussed forum post analyzes Professor Zhao Bin of Fudan University's first-principles critique of the degree thesis system in the AI era. The argument: theses were meant to assess independent research ability, but in practice they reward formatting compliance and typing speed. Zhao identifies three structural failures—institutional laziness (equating length with quality, defense rituals, handwriting requirements as AI-proofing), a reward structure that incentivizes cheating (honest writing takes ~300 hours while AI-assisted writing takes ~10 hours and passes more easily), and a proposed solution: abolish thesis requirements for bachelor's and master's degrees and replace them with an achievement-based system where students demonstrate solved real problems. The post argues AI detection is a farce everyone knows is unreliable yet everyone performs, and that education should evaluate skills AI cannot replicate: problem definition, value judgment, and navigating uncertainty.

This post presents Professor Zhao Bin's (Fudan University) first-principles critique of the degree thesis system, framed through Elon Musk's "first principles" method: strip away assumptions and reason up from basic facts.

Original purpose vs. reality

The thesis system was created to evaluate independent research ability. In practice, the author argues, most of a bachelor's thesis workload (claimed ~90%) is meaningless—font sizes, line spacing, citation formats—none of it related to research capability.

Three structural deviations

1. Institutional laziness

"Longer thesis = higher quality" is a lazy proxy. Length has no necessary link to the ability to pose and solve problems (Einstein's 1905 papers were short yet revolutionary).
Oral defenses become ritual: committee members lack time for deep reading by design, not attitude—the system mandates defenses without providing time resources.
Requiring handwritten drafts or handwritten summaries to defeat AI is absurd: handwriting measures calligraphy, not thinking. The system has let detection tools (plagiarism rate, AI-detection rate, handwriting checks) hijack the original goal.

2. The system rewards cheating

Zhao's arithmetic: honest writing takes ~300 hours; AI-assisted writing takes ~10 hours and often passes more easily because AI is fluent in thesis templates.
Citing the economics concept of incentive compatibility: a good system aligns individual interest with social goals; the thesis system does the opposite. This is a structural problem, not a student morality problem.

3. Replace theses with an achievement system

The proposal: abolish thesis requirements for bachelor's and master's degrees. Graduation should depend on what problem you solved, not how many words you wrote.
Valid outcomes could include a published paper, an open-source codebase, a startup project, a consulting report, even a failed experiment that demonstrates a complete research process.
Thesis system = process-oriented (complete a fixed workflow); achievement system = outcome-oriented (prove you have the ability).

The deeper point: education in the AI era

The traditional assumption "writing ability = research ability" is broken—AI writes structurally complete, fluent, properly cited papers faster and more perfectly than humans. But AI cannot generate genuine problem awareness; it optimizes known paths but rarely opens new directions.

Zhao's stance is not anti-technology: AI is a tool. Students who know what they are doing can compress 300 hours to 10 and spend the rest on real thinking; students who don't produce "exquisite garbage."

The "AI-detection farce"

Everyone knows detection is unreliable—universities, students, and even the detection vendors (who sell reassurance, not truth)—yet everyone performs the check. Strengthening detection is the easiest institutional response (minimal redesign, minimal risk) and the least effective, since bypass methods perpetually outrun detection. The harder, correct response is moving the evaluation anchor to what AI cannot do: problem definition, value judgment, ethical trade-offs, creative breakthroughs.

Conclusion

The post ends: change is hard—millions of students, tens of thousands of reviewers, thousands of institutions—but difficulty does not make it wrong. AI-era education should produce people who can define problems, judge value, and advance under uncertainty, and the system should evaluate exactly that.