E236 | 99% of Homework Is Written by AI: What Remains of College in the Eyes of Today's Elite Students?
Listen
问这期播客
会先在本集摘要、章节、转录和笔记里找答案。
TL;DR · AI Summary
Top university students now use AI for 99% of their assignments, shifting the core value of college from knowledge delivery to self-awareness and critical thinking.
Key Takeaways
- A Columbia student reports 99% of homework involves AI, calling traditional read
- AI tools like Dify and Variant are replacing ChatGPT as preferred choices for de
- Human understanding and judgment—not AI output—determine real academic and profe
Outline
Jump quickly between sections.
The first cohort of students who grew up with generative AI is entering graduation season, with AI now a standard learning tool.
Students from Tsinghua, NYU, and Columbia share how they use AI to complete coursework and enhance personalized learning.
As knowledge acquisition is automated by AI, college’s core value shifts to self-discovery and deep peer interaction.
ChatGPT is being replaced by faster, more specialized tools like Mercury 2 and Dify.
Human-specific skills like aesthetic judgment, coding intuition, and decision-making remain key differentiators.
Maintaining independent thought amid AI dependency is a defining challenge for this generation of graduates.
Mindmap
See how the topics connect at a glance.
查看大纲文本（无障碍 / 无 JS 友好）
- AI时代的大学教育
- 学习方式变革
- 99%作业含AI
- 个性化学习路径
- 大学价值重构
- 自我认知
- 同侪交流
- 批判性思维
- AI工具演进
- Dify / Variant
- Mercury 2提速
- ChatGPT被弃用
- 不可替代能力
- 审美判断
- 选择力
- 元认知
Highlights
Key sentences worth saving and sharing.
99% of my assignments involve AI—I can’t think of a single one that didn’t.
AI can make decisions (deciding), but only humans can make choices (choosing).
Reading textbooks won’t make me a better student—it’s just wasting my time.
Chapters
刚上大学就有了ChatGPT，预感打不过开始辅修AI
刚上大学就有了ChatGPT，预感打不过开始辅修AI
AI速通法律专业课：模拟教授出题，拿下4.0满绩
AI速通法律专业课：模拟教授出题，拿下4.0满绩
大学老师对AI态度发生转变，评估体系亟待升级
大学老师对AI态度发生转变，评估体系亟待升级
大学两大价值——认识自己 + 获得技能；后者被AI侵蚀
大学两大价值——认识自己 + 获得技能；后者被AI侵蚀
记忆类知识会过时，批判性思维和素养课程更值得投入
记忆类知识会过时，批判性思维和素养课程更值得投入
清华最重要的财富：同侪的深度交流、老师的解惑
清华最重要的财富：同侪的深度交流、老师的解惑
马斯克说对了，我上大学最重要的目的就是社交
马斯克说对了，我上大学最重要的目的就是社交
学历不再是知识的证明，但仍是社会信任的重要凭证
学历不再是知识的证明，但仍是社会信任的重要凭证
追寻美德、向往真理，才能更好地驾驭AI技术
追寻美德、向往真理，才能更好地驾驭AI技术
用AI在大学期间大量试错，而不是对中长期目标做死板的预设
用AI在大学期间大量试错，而不是对中长期目标做死板的预设
AI可以做出决策（deciding），但只有人能做出选择（choosing）
AI可以做出决策（deciding），但只有人能做出选择（choosing）
“自顶向下”学习法：先见证最高水平的卓越，再提问、理解、验证
“自顶向下”学习法：先见证最高水平的卓越，再提问、理解、验证
Transcript
刚上大学就有了ChatGPT，预感打不过开始辅修AI
AI速通法律专业课模拟教授出题，拿下4.0满绩
大学老师对AI态度发生转变，评估体系亟待升级
大学两大价值——认识自己 + 获得技能；后者被AI侵蚀
记忆类知识会过时，批判性思维和素养课程更值得投入
清华最重要的财富同侪的深度交流、老师的解惑
马斯克说对了，我上大学最重要的目的就是社交
学历不再是知识的证明，但仍是社会信任的重要凭证
追寻美德、向往真理，才能更好地驾驭AI技术
用AI在大学期间大量试错，而不是对中长期目标做死板的预设
AI可以做出决策（deciding），但只有人能做出选择（choosing）
“自顶向下”学习法先见证最高水平的卓越，再提问、理解、验证
99%作业由AI参与，已经想不到任何没有AI参与的作业了
翻书不会把我变成一个更好学生，只是在浪费我的时间
当大家都用AI做作业，真正PK的还是每个人的理解和判断
最近常用AI工具盘点
哪些AI工具正在被抛弃？首当其冲ChatGPT
适当关闭memory，避免模型被记忆污染形成信息茧房
设计师的审美，程序员对代码的感觉，仍然不可替代
AI比短视频更上瘾，失去AI就失去了执行力和灵感来源
完全抽离AI我们还能做什么？还能回到手写代码吗？
用AI的资深工程师反而慢了19%，初学者在AI时代可能反成优势
2026最迷茫就业季，这一代毕业生就业向何处去
硅谷高中生的反驳主播为什么要假设，我们不会离开AI了
这集是什么时候录的最好说一下，因为最近各家更新频率都很高，Codex明显有进步而claude也有过降智。我觉得 claude 很好用我也在掏钱，但不要把claude当成信仰～之前嘉宾也说了，AI 时代要保持不断地学习，你可以说 2026-03 Claude 降维打击其他模型，但你仍需要 follow up 其他模型的进展～ 从我听播客的经历来看，越是之前完全不懂编程的人，越容易神话 claude，觉得哪哪都比 Codex /其他模型强强，然而 Codex 训练得更严谨，就如 Peter Steinberger 说的，Codex 有时候会对一个复杂任务read for half an hour，然后才开干～以后最好改成我只用最好的模型，而不是我只用 Claude
哈哈哈现在每月都花 天天泡在一起
非常赞同
Show notes
E236｜99% of Homework Is Written by AI: What Remains of College in the Eyes of Today’s Top Students? - SV101 | Xiaoyuzhou FM – Listen to Podcasts on Xiaoyuzhou
Do housework. Morning hygiene.
Listen to podcasts, on Xiaoyuzhou! Click to download.
E236｜99% of Homework Is Written by AI: What Remains of College in the Eyes of Today’s Top Students?
SV101
82 minutes · 2 hours ago
2185
·
9
The first generation of students who grew up alongside generative AI is now entering graduation season.
When we spoke with college students about AI last year, discussions were still centered on “whether to use AI” and “how professors feel about it.” But just one year later, AI has truly become a productivity tool. This graduating cohort may realize: whether you're studying knowledge-based courses or practical skills, no field is immune to AI disruption. To stand out, one must become someone who "uses AI better."
When knowledge can be retrieved by AI and skills replaced by tools, what unique value can universities still offer? In the face of AI disruption, which abilities remain timeless—what are the “meta-competencies” that won’t be reshuffled by the times?
In this episode, we invite three recent elite graduates—Alfred, Kolento, and Jack—to discuss how they used AI to achieve highly personalized learning within standardized university education; where they draw intellectual nourishment from, and what capabilities matter most to them; and as they stand at the threshold of graduation, how they evaluate the value of college and what choices they’ve made for their futures.
This is also a self-rescue guide for all college students: the abilities AI cannot replace are the true confidence education leaves us with.
【Host】
Hongjun, Founder & Producer of SV101
【Guests】
Lin Tongyu (Alfred), Tsinghua University, Law (International), Class of 2026 Hou Taiyu (Kolento), New York University, Applied Psychology, Class of 2026 Rao Jiewu (Jack), Columbia University, M.S. in Management Science & Engineering, Class of 2026; Founder & CTO of Emergences Labs
【You’ll Hear】
Alfred
03:28 Got ChatGPT right when starting college—felt defeat was inevitable, so began minoring in AI
09:40 Used AI to speed-run law coursework: simulated professor-style exam questions, achieved a perfect 4.0 GPA
12:15 Professors’ attitudes toward AI are shifting—evaluation systems urgently need upgrading
13:43 Two core values of college—knowing yourself + acquiring skills; the latter being eroded by AI
15:49 Memorization-based knowledge becomes outdated—critical thinking and literacy courses deserve more investment
20:11 Tsinghua’s greatest asset: deep peer exchanges and teachers who clarify doubts
Kolento
23:05 Musk was right—the main purpose of college for me is socializing
26:30 Degrees are no longer proof of knowledge, but remain key credentials for social trust
28:35 Only by pursuing virtue and aspiring to truth can we better master AI technology
29:55 Use AI to experiment and make mistakes extensively during college, rather than rigidly predefining long-term goals
32:32 AI can make decisions (deciding), but only humans can make choices (choosing)
36:38 “Top-down” learning method: witness excellence at its highest level first, then question, understand, and verify
Jack
39:23 99% of assignments involved AI—I can't even recall any assignment done without AI
43:14 Flipping through textbooks doesn’t make me a better student—it’s just wasting my time
44:44 When everyone uses AI for homework, what’s really being tested is each person’s understanding and judgment
Evolving AI Tools
46:51 Review of recently used AI tools
52:37 Which AI tools are being abandoned? First on the list: ChatGPT
58:46 Temporarily disable memory to avoid model contamination and information bubbles
Meta-Cognition Accumulation and AI Addiction
01:01:33 A designer’s aesthetic sense, a programmer’s intuition for code—still irreplaceable
01:04:43 AI is more addictive than short videos—losing access to AI means losing both execution power and inspiration
01:06:28 What could we do if completely detached from AI? Could we go back to writing code by hand?
01:12:53 Senior engineers using AI are actually 19% slower—beginners might have an advantage in the AI era
01:17:07 The most uncertain job market in 2026—where should this generation of graduates go?
01:20:27 A rebuttal from a Silicon Valley high schooler: Why assume we won’t ever leave AI?
【AI Tools Mentioned in This Episode】
Variant: An AI design tool that generates dozens of high-fidelity web and app design concepts within seconds from a single-sentence prompt.
Typeless: An AI voice dictation tool that turns natural speech into polished, ready-to-send text as if spoken by a real person.
Dify: An open-source platform enabling users to quickly build and deploy AI applications based on large models via drag-and-drop and visual interfaces.
Raycast: A Mac efficiency hub that unifies control over your entire computer via keyboard, allowing rapid launch of apps and services.
Suno: An AI music creation platform that supports personal voice cloning and one-click generation of complete original songs with vocals and instrumentation.
Gemini Dynamic View: Google Gemini’s dynamic visualization feature that generates interactive 3D models and custom UI interfaces in real-time based on natural language prompts.
Mercury 2: An ultra-fast inference large language model based on diffusion architecture, generating over 1,000 tokens per second—5 to 10 times faster than mainstream models.
【Executive Producer】
Hongjun
【Post-Production】
Amei
【Operations】
Zhu Jie
【BGM】
Between Moments - Amber Glow Call Waiting - Future Joust Pulse Runner - Helmut Schenker First Flakes - Philip Ayers
【Find Us Here】
WeChat Official Account: SV101 Listening Platforms: Apple Podcasts｜Spotify｜Xiaoyuzhou FM｜Ximalaya｜Qingting FM｜Lizhi FM｜NetEase Cloud Music｜QQ Music Other Platforms: YouTube｜Bilibili search “SV101 Podcast” Contact Us: [podcast@sv101.net](mailto:podcast@sv101.net)
Special Guests: Kolento, Lin Tongyu, and Rao Jiewu.
Expand Show Notes
1 hour ago
Education feels severely delayed in the AI era. The new era should start from problems, not learn atomic concepts bottom-up. College seniors today are under immense pressure.
2 minutes ago
Can't stand the pretentiousness after six minutes
Justin0087: Leaving, leaving Justin0087: Or is the male speaker actually AI-generated?
24 minutes ago
It's truly impressive that these young people have such insights. I still remember when we were in college twenty years ago—most of us were just低头 working on assignments, never really thinking about what we genuinely liked or why we were even attending university. The younger generation is truly formidable!
1 hour ago
The interviewees aren't representative enough—what do students with a CS background think about the remaining value of university? How does the perspective change when science includes extensive lab work?
2 minutes ago
54:40 It would be better to mention when this episode was recorded, since update frequencies across platforms have been very high recently. Codex has clearly improved, while Claude has also experienced periods of degraded performance. I find Claude very useful and am paying for it myself, but let's not treat Claude like a religion. As previous guests have said, in the AI era we must keep learning continuously. You might say that by March 2026, Claude will outclass all other models—but you still need to keep up with developments in other models. From my podcast listening experience, the people who previously knew nothing about programming are most likely to mythologize Claude, believing it's superior to Codex and other models in every way. Yet Codex is trained more rigorously—as Peter Steinberger put it, Codex might spend half an hour reading and understanding a complex task before even starting to work. In the future, it'd be better to say "I only use the best model," rather than "I only use Claude."
4 minutes ago
06:38 Hahaha, now spending monthly and hanging out together every day
35 minutes ago
36:01 Fully agree
00:00
00:00