返回资讯中心
外部精选
软件工程
#开发者实时流

Colleges need to teach last-mile knowledge

Back when a college degree signaled competence and employers met with applicants in person (before apps and algorithms), a bright college graduate who showed initiative, judgment, and the ability to look up facts was wo…

Hacker News 最新paulpauper2026/10/18 · 2026/10/1816 分钟阅读

以下正文同步自 Hacker News 最新,版权归原站所有,已转换为易读排版。

Hollis Robbins

Oct 08, 2026

101

30

29

Share

Back when a college degree signaled competence and employers met with applicants in person (before apps and algorithms), a bright college graduate who showed initiative, judgment, and the ability to look up facts was worth the investment. The diploma was a “signal” as Bryan Caplan put it. An employer could trust that the graduate had persisted through four years of coursework and would likely do a pretty good job. Today, a diploma may only signal that a student can prompt an LLM.

In the AI era, universities need to ensure graduates have specific knowledge and hands-on experience. This would seem obvious, yes? Yet the new fad on campus is teaching concepts, not facts. The argument is that “durable skills” like “critical thinking” and “emotional competence” are needed instead. This is wrong-headed. LLMs can supply nearly all the critical reasoning a firm needs. What they can’t supply is local, on-the-ground, long-tail knowledge, tacit knowledge, continual learning a person picks up on the job, and what I’ve called last-mile knowledge.

The problem is the thinking class’s habit of organizing ideas by height. You know what I mean: facts are low; symbols and theories are high. Think of phrases like “facts on the ground” and “grand theory.” In Language in Thought and Action (1949), S. I. Hayakawa illustrated this with a ladder, with a specific cow at the bottom and the word “wealth” at the top. The high words are conceptual and say little about particular facts. You can know lots of things about one particular cow, “Bessie.” You can know a lot about “livestock” (cows, sheep, pigs) but know nothing about Bessie. You can have lots to say about “wealth” and have no facts at your fingertips about anything.

Ladder of abstraction

In 1970, the political scientist Giovanni Sartori described the tradeoff as concepts gaining coverage by giving up defining attributes. Broader categories can include more cases but require less understanding of each case. Benjamin Bloom has caused immeasurable damage with that 1956 pyramid that relegated remembering and understanding facts at the bottom.

Sartori, Bloom

Facts on the ground are the new gold in the AI era. I argued in “AI and the Last Mile,” rightly praised by Forbes as a top ten tech article of 2024, that the distance between what AI knows and can do and the particular facts on the ground would become more and more important. AI has become increasingly powerful in the last two years, but everyone knows a person is still needed for granular detail. The economists Luis Garicano, Jin Li, and Yanhui Wu describe this problem in “Messy Jobs: The Work That AI Cannot Reach,” reviewed by Joshua Rothman in The New Yorker. They call it a “90/10 production function”: AI can do good and very good work but not always excellent work. When AI takes over routine problems and “the human handles the exceptions,” experienced workers can become more valuable. “If every team can generate better AI-based analysis to support their arguments, then the demand for conflict resolution and authority-based decisions will increase dramatically,” the authors say.

In short, whoever has the best “last mile” knowledge, at the bottom of the ladder and the pyramid, wins. Students need to be told this. Universities need to toss “durable skills” out the window.

The only part of a college education worth paying for is the last mile of specific knowledge, the local facts and particular wisdom that AI does not have access to. Higher education needs to focus on the unknown quadrants to show students that there are facts that AI does not yet have. Students who go deep into a subject, who get hands-on experience with facts on the ground, are more likely to notice the particulars of a situation. They realize when something is missing. They see that a key fact is wrong.

I ask students what they know that their LLM does not know. If they look blank, I say: AI does not know what brand of sneakers you wear. AI does not know if your grandfather was left-handed. Half of the crime shows on TV hinge on things like allergies and old nicknames that AI does not know.

Some fields still teach and test specific knowledge and competency. Language graduates can speak and write the language. Engineers still have to measure and calculate things. But the most popular undergraduate degrees today are almost wholly fact-free and abstract. UC Irvine’s business administration program promises only “mastery of specific business discipline knowledge” though without specifying any knowledge (seriously click on this website). Missouri State’s psychology program promises only familiarity with “major concepts, theoretical perspectives, empirical findings, and historical trends.” Rhode Island’s human development program asks students to apply an “Ecological Framework” to individual and family dynamics. Assessment is ambiguous, as Joshua Travis Brown puts it. Students today are tested from the top of the ladder.

I am reminded of Vittorio Mussolini’s famous observation that from the air, bombs look like budding roses.

You can’t do “critical thinking” or “emotional intelligence” from the great heights. I look at lists of “durable skills” like “collaboration” and “creativity” and wonder why universities are chasing cognitive abstraction at a time when employees want on-the-ground knowledge.

Faculty are told to downplay facts. Miami University’s guide tells professors to avoid “know,” “learn,” and “understand” in learning outcomes because what occurs in a student’s mind cannot be measured. RIT’s guide tells faculty to avoid words like “Understand, Know, Realize.” Chaffey College tells faculty to avoid “Memorize, Perceive, Realize, Recognize, See, Think, Understand, Value.”

Knowing actual things matters in the AI era. Consider why it is that kids who do well on spelling bees do well in life.

I understand I am arguing against an entire movement in higher ed. But I’m right. Specific, last-mile knowledge matters more than ever. Faculty who know more than AI are more valuable than ever. Studying with an expert may turn out to be more valuable than a degree from a college, as I said recently. Higher education needs to move down the ladder, from abstractions to expertise and facts.

正文由 探How 从来源站点 RSS 同步,内容未经改写;遇到排版缺失或需要图片、视频时请以原文为准。

打开原文
去社区讨论 →