Abundance in Learning Materials

Human time is the ultimate bottleneck

When I was an undergraduate, It was not easy to find economics textbooks in English. To read James Hamilton’s Time Series Analysis, the big, chuck book, I reserved it in the library system, picked it from the front desk, returned it before a tight deadline when it was recalled by another user, and played the game to get it back by submitting another reservation. It was the time of scarcity.

As internet became widespread, I was amazed that so many pirate e-copies were easily available. With serendipity, one could search virtually any mainstream textbooks of various scanning quality and download them. There was even a market of pirate books (in PDF) on CENET (China Economics Education and Research Network), powered by a monetary token mechanism to incentivize file sharing and exchange. A user could upload useful assets (books, slides, lecture notes, exercises, reports) to earn virtual money, and then use the these tokens to buy other people’s assets. This was the first stage of abundance in learning materials. Scarce was human time. With written knowledge available with a quick internet search, the difficulty of self-learning remained.

Before the pandemic, MOOC has been around on video platforms such as Youtube and Bilibili. College students found that some courses could be learned by watching videos. These videos were mostly prepared by famous teachers from top universities, with a dedication in teaching, which served as a selection mechanism. Poor teachers would not embarrass themselves by sharing online their videos of in-class teaching. However, watching videos took time, and people just had so much time. When the pandemic hit, all over a sudden universities became virtual, powered by Zoom or Tencent Conference. Sitting all day in front of a screen, university students did not like online teaching. It is a format with little interaction, and their remote teachers mostly were no better than the online courses from other universities.

Here comes AI. AI makes self-learning interactive, which was beyond imagination five years ago. The bottleneck is still the same—a person’s time. The time to understand a proof written by AI is no shorter than that written in a book. The interactions will AI can even take more time, with the questions and responses back and forth. Abundance is knowledge available via cloud services of large language models, powered by infrastructures of data centers, electricity, and alchemies of training and refinement.

At the end of the day, learning is about digesting and ruminating existing knowledge to fill a person’s brain, to be retrieved one day when needed. In the old time, the knowledge was printed in papers or in PDF, hibernating in the library bookshelves or my computer’s hard disk; today the knowledge is stored in the trillion parameters of a large language model, waiting to be waked by users’ prompts. Learning is an ultimate personal experience, while machine’s learning is simply a metaphor. If humans do not want to be enslaved by machines, we must keep learning, as we have

“nothing to offer but blood, toil, tears, and sweat.”

AI has not obtained these things yet.

Postscript

After I posted the previous message, I asked Kimi 3 to translate it into Chinese, and embed Chinese Wuxia genre stories in it. I am amazed by its capacity. I share it with you, with my revision.

一、孤本时代

我念本科时，找一本英文原版教材并不容易。为了读Hamilton那本砖头厚的《时间序列分析》，得先在图书馆系统里预约，到前台取书，被别的读者催还时乖乖归还，然后再提交一次预约，把书夺回来。那个年代，书本身就是稀缺品。

《射雕》的读者对此绝不陌生：《九阴真经》一现世，天下英雄抢得头破血流——王重阳要办一场华山论剑来定归属，欧阳锋机关算尽，梅超风夫妇干脆直接下手偷。可偷到半部经书又如何？没有名师指点，正宗功夫被她练成了九阴白骨爪。可见稀缺的不只是书，更是领路的人。

二、刊印时代

互联网一来，盗版电子书俯拾皆是，运气好的话，主流教材都搜得到。CENET上甚至兴起过一个PDF书市，用虚拟币结算：上传讲义赚币，再花币买别人的藏书——活脱脱一个武林人士私下交换武功图谱的黑市。藏经阁的大门被撬开了。

学习资料由此进入富足。但稀缺并未消失，只是搬了家：这回稀缺的是人的时间。王语嫣坐拥琅嬛玉洞，天下武学倒背如流，却连一个庄汉都打不过；萧远山、慕容博在藏经阁偷学数十年，没有相应的佛法打底，练出一身内伤。秘籍印得满大街都是，绝世神功依然稀有。

三、视频时代

疫情之前，MOOC早已在YouTube和B站扎根。把课挂到网上的，多是名校名师——讲得不好，谁好意思示众？这就像东邪西毒把毕生绝学录成影像公之于众：敢开坛的，本来就得是五绝级别。可惜看视频照样花时间，而人的时间就那么多。

疫情一来，大学一夜之间搬进了Zoom和腾讯会议。学生整天对着屏幕，味同嚼蜡：互动几近于无，自家老师隔着屏幕讲，未必比得过别校名师的录像。好比看录像学降龙十八掌——掌法看得真切，却没人喂招，比划出来的终究是三脚猫功夫。

四、AI时代

AI来了。自学第一次变成了互动：随时提问，请它换个角度讲定理，挑你证明里的毛病，按你的节奏陪练。这相当于黄药师住进了你家，洪七公一招一式给你喂降龙十八掌；放在五年前，这是想都不敢想的待遇。

可瓶颈还是那个瓶颈：一个人的时间。AI写的证明，读懂它并不比读书省工夫；一来一回地问答，有时反而更费时间。如今的知识栖身于万亿参数之中，靠数据中心和电力供养，等着被提示词唤醒。然而张无忌拿着全本《九阳真经》，也仍在昆仑山谷里独自练了五年；郭靖天资平平，靠的是”人家练一朝，我就练十朝”。AI能把师父请到你手边，却不能替你扎马步。

学习这回事，无非是把现成的知识消化反刍，装进自己大脑以备他日取用。归根结底，是个人体验。人类若不想被机器奴役，唯有继续学习。诚如丘吉尔所言——

I have nothing to offer but blood, toil, tears, and sweat.

这几样东西，AI还未曾拥有。