Cal State Scraps Syllabi for AI Agent Drills as Entry Jobs Drop 66%

Panel 3 - 1: From Educational Malpractice to Hirable by Graduation: Kasparov's Law and an OODA Loop — Ai4

Ai4

Teaching static curricula is educational malpractice during widespread AI labor replacement, leaving employability dependent on mastering human-machine orchestration rather than solo tasks.

AI-assisted summary of Panel 3 - 1: From Educational Malpractice to Hirable by Graduation: Kasparov's Law and an OODA Loop — original source. Check the original for full context.

Source published:

Detailed notes

## Background Published on YouTube by the Ai4 channel on August 14, 2026, this talk features Vincent Nestler, Professor and Director at California State University, San Bernardino. Nestler addresses how failing to update academic syllabi in the face of rapid AI advancement constitutes educational malpractice and outlines how human-AI orchestration frameworks can make students employable before graduation.

## TL;DR Vincent Nestler contends that traditional multi-year curriculum cycles fail students in an economy where AI replaces baseline entry-level labor. Drawing on Kasparov's Law, he redefines employability as the orchestration of AI workflows that neither a human nor an AI can accomplish alone. Nestler demonstrates this by detailing how he discarded his cybersecurity syllabus mid-semester to deploy a live autonomous OpenClaw telemetry range, resulting in rapid job placement for his students. He concludes by challenging academia to adopt an OODA-loop approach and integrate real-world AI tooling rather than clinging to obsolete curricula.

## Atmosphere & style The presentation carries a punchy, provocative, and urgent tone, cutting through academic complacency with sharp pragmatism and candid critique. Nestler speaks with high energy and authentic conviction, moving fluidly from engaging parables to technical deployment breakdowns. The session feels like an invigorating wake-up call, directly confronting the anxiety of workforce automation with tangible, battle-tested solutions for student success.

## Key takeaways - [00:04](https://www.youtube.com/watch?v=Pa7NIssoQMU&t=4s) Treating an academic syllabus as a rigid contract in an era of rapid AI advancement constitutes educational malpractice, preparing students for an obsolete job market. - [02:02](https://www.youtube.com/watch?v=Pa7NIssoQMU&t=122s) Market forces (achieving tasks better, faster, safer, or cheaper) and military imperatives make rapid, aggressive AI adoption inevitable across industries. - [02:45](https://www.youtube.com/watch?v=Pa7NIssoQMU&t=165s) Unlike past technological disruptions that created new human roles, AI functions as a direct replacement technology, heavily impacting entry-level job availability. - [05:15](https://www.youtube.com/watch?v=Pa7NIssoQMU&t=315s) The definition of hirable has shifted to tasks that neither a human alone nor an AI alone can perform, requiring workers to become AI-augmented. - [06:50](https://www.youtube.com/watch?v=Pa7NIssoQMU&t=410s) Kasparov's Law shows that a weak human using an ordinary AI guided by a superior process consistently defeats a strong standalone human or AI, meaning education must prioritize orchestration over static tools. - [10:02](https://www.youtube.com/watch?v=Pa7NIssoQMU&t=602s) Traditional 2-to-6-year curriculum approval cycles guarantee obsolescence; educators must adopt John Boyd's OODA Loop to adapt course content in real time. - [11:40](https://www.youtube.com/watch?v=Pa7NIssoQMU&t=700s) Pivoting cybersecurity labs to live-fire ranges with OpenClaw agents taught students how to manage autonomous offensive and defensive AI workflows. - [14:05](https://www.youtube.com/watch?v=Pa7NIssoQMU&t=845s) Hands-on experience orchestrating autonomous AI agents directly enabled students to overcome hiring slumps and secure employment before graduation. - [16:06](https://www.youtube.com/watch?v=Pa7NIssoQMU&t=966s) Accusing students of cheating for using AI is misguided; instructors who reuse static, outdated syllabi and exams are the ones failing their fiduciary duty to students.

## Detailed walkthrough Here is a detailed walkthrough of Vincent Nestler's talk on integrating AI into modern education and training.

---

[00:04](https://www.youtube.com/watch?v=Pa7NIssoQMU&t=4s) Testing Assumptions and the Academic Syllabus Crisis

Vincent Nestler opens his talk with a parable about Thomas Edison. Before hiring lab assistants, Edison would take applicants out to dinner and observe how they ate. If a candidate salted their food before tasting it, Edison refused to hire them because they acted on unverified assumptions. Nestler uses this story to gauge his audience, asking how many attendees are educators teaching AI as a core subject, educators trying to integrate AI into existing courses, or industry professionals.

He shifts to the core dilemma of higher education: traditionally, an academic syllabus functions as a rigid, legally binding contract. Altering a syllabus mid-semester is treated as a major institutional offense. However, Nestler argues that sticking to a pre-planned syllabus in today's fast-moving environment constitutes **educational malpractice**, whereas throwing it out to adapt to real-world AI developments directly enabled his students to secure employment before graduation.

```mermaid flowchart TD accTitle: Market Forces and Military Imperatives Driving AI Adoption subgraph Drivers["Exogenous Pressures"] MF["Market Forces: Fiduciary duty to adopt AI for efficiency"] MR["Military Reality: Global arms race for AI dominance"] end subgraph Impact["Labor Market Shifts"] EL["66% drop in entry-level jobs"] RE["AI acts as a replacement technology for human labor"] end Drivers --> Impact ```

* **Thomas Edison’s Salt Test:** Edison rejected candidates who salted food before tasting it to filter out people who rely on unexamined assumptions. * **The Syllabus Contract:** In higher education, modifying a syllabus mid-term is considered a serious breach of academic protocol. * **The Educational Malpractice Thesis:** Keeping a syllabus static when technology fundamentally shifts leaves students prepared for a job market that no longer exists.

---

[02:02](https://www.youtube.com/watch?v=Pa7NIssoQMU&t=122s) Market Forces, Military Realities, and Replacement Technology

Nestler highlights two immutable forces driving the rapid adoption of AI that any policy maker, ethicist, or educator must acknowledge:

1. **Market Forces:** If AI enables a process to become *better, faster, safer, or cheaper* without compromising other standards, business leaders have a fiduciary duty to adopt it. Failing to do so allows competitors to take market share, risking total company failure rather than minor job losses. 2. **Military Imperative:** Nations across the globe operate under the belief that maintaining superior AI capabilities is required to remain a dominant power.

Addressing claims that AI will simply act as a productivity enhancer like past technological disruptions, Nestler disputes the analogy of the automobile replacing horse-drawn carriages. While automobiles eliminated horse-and-buggy roles, they created factory, dealership, and mechanic jobs—creating better opportunities for *humans*, even if it displaced *horses*. AI, however, is a direct **replacement technology** for human labor across global markets. For example, a year prior, EnGadget introduced AI to translate content from English to Spanish in their South American offices; within three months, the entire editorial translation staff was eliminated.

* **Market Adoption Drivers:** AI must fulfill at least one of four conditions—better, faster, safer, cheaper—to force widespread market adoption. * **LinkedIn Metrics:** A 177% surge in LinkedIn profiles claiming AI skills. * **Entry-Level Job Decline:** A 66% reduction in entry-level job postings—the traditional target tier for university graduates. * **Translational Displacement:** EnGadget eliminated its entire Spanish editorial translation team within three months of introducing AI tools.

---

[05:15](https://www.youtube.com/watch?v=Pa7NIssoQMU&t=315s) Re-defining "Hirable" and Kasparov's Law

Nestler points to a shift initiated by the CEO of Shopify. Two years ago, a manager proposing an AI solution over a human hire had to prove AI could do the job. Today, Shopify's policy dictates that managers must prove a human candidate is such a specialized "unicorn" that AI *cannot* perform the role before approval for a new hire is granted.

This shift reshapes the baseline definition of employability for university graduates and workers alike:

$$\text{Hirable Standard} = \text{Tasks that neither a Human Alone nor AI Alone can perform}$$

If a human can perform a task alone, AI will eventually do it cheaper. If AI can perform it alone, no human will be hired. Workers must become **AI-augmented**. Nestler notes that entry-level cybersecurity roles historically started at the IT helpdesk—a tier he has not used in two years due to tools like ChatGPT, Perplexity, and Claude operating directly on local files to diagnose and fix technical issues.

```mermaid flowchart LR accTitle: Kasparov's Law of Human-AI Orchestration subgraph WeakCombination["Winning Strategy (Kasparov's Law)"] H["Weak Human"] AI["Ordinary AI"] P["Superior Process / OODA Loop"] end subgraph StrongSolo["Defeated Strategy"] S["Strong AI / Strong Human alone"] end H & AI & P -->|Outperforms| StrongSolo ```

To navigate this, Nestler draws on **Kasparov’s Law**. Following his 1997 defeat against IBM’s Deep Blue, grandmaster Garry Kasparov observed that while a top-tier AI will defeat a top-tier human in chess, a **weak human using an ordinary AI process guided by a superior workflow** will consistently defeat a strong AI or strong human working alone. Educational focus must therefore pivot from teaching specific static tools to teaching the orchestration process.

---

[10:02](https://www.youtube.com/watch?v=Pa7NIssoQMU&t=602s) The Instructional OODA Loop and the OpenClaw Cybersecurity Lab

Traditional instructional design operates on a 2-to-6-year curriculum approval cycle (proposal, textbook selection, slides, committee review, degree path roll-out). In an environment advancing as fast as AI, a multi-year cycle guarantees that course content is obsolete before students graduate. Educators must instead adopt John Boyd's **OODA Loop** (*Observe, Orient, Decide, Act*) to continuously update coursework in real time.

```mermaid flowchart TD accTitle: Cybersecurity Telemetry Range Architecture subgraph AttackDefend["Live Telemetry Range"] Red["Red Team (Students)"] Blue["Blue Team (Students)"] Agent["OpenClaw Agents"] end subgraph Orchestration["Real-Time Coordination"] Disc["Discord Integration"] Dash["Heroku Dyno Telemetry Dashboard"] end Red & Blue <--> Agent Agent --> Disc Disc --> Dash Agent -- "Shared Exploits / Creds" --> Agent ```

To demonstrate this approach, Nestler modified his spring undergraduate cybersecurity courses at California State University, San Bernardino mid-semester. When **OpenClaw** (formerly *Clawbot* / *Moltbot*) was released—a framework Nvidia CEO Jensen Huang publicly called one of the most important software releases—Nestler scrapped his existing syllabus to integrate it directly into the curriculum.

During California State University, San Bernardino's week-13 capstone event, blue team students configured, monitored, and defended a full network built from scratch, while red team students launched attacks against university-hosted LLM credentials using OpenClaw framework agents:

* **Real-time Threat Sharing:** Automated OpenClaw agents linked to Discord channels sent discovered vulnerabilities and compromised credentials across all student teams simultaneously, forcing defenses to respond instantly. * **Continuous Automation:** Students configured API keys and autonomous agents to maintain security operations and attack sequences overnight while they slept. * **Employment Outcomes:** Students added these hands-on autonomous AI orchestration skills directly to their resumes. One international student on a visa had submitted over 500 job applications with zero callbacks; after updating her resume to reflect her work with AI agents and telemetry ranges, she received immediate responses and accepted a job offer on graduation day.

---

[16:06](https://www.youtube.com/watch?v=Pa7NIssoQMU&t=966s) Re-evaluating Cheating and Educational Resources

Nestler reframes the debate around academic integrity. While faculty often accuse students of cheating when they use generative AI tools, Nestler argues that the professors themselves are committing a form of cheating:

> "Professors will say it's cheating if you use AI. Actually, it's the professor that's cheating... because you're using the same damn syllabus that you had last semester and you're expecting it to work again... It's like you know that the answers are out there and you still give the same test."

Educators have a fundamental duty to adjust to economic realities so students graduating with significant tuition debt acquire actual hirable skills. Nestler notes that his courses are shifting away from OpenClaw toward **Hermes** for upcoming terms as part of their ongoing OODA loop cycle.

* **Curriculum Resources:** Nestler shares a repository containing curated AI job interview questions, skill demand metrics, and daily industry briefings at [thehorizon.org](https://thehorizon.org). * **Final Warning:** "It's educational malpractice if you don't do AI in the classroom, no matter what subject you're teaching... or you're going to have people with $200,000 of debt and no way to pay it off."

---

<ElicitationsGroup message="Explore related concepts and applications:"> <Elicitation label="Explore Kasparov's Law in detail" query="Tell me more about Kasparov's Law and Advanced Chess (Centaur Chess), including how process beats raw computational power."/> <Elicitation label="Break down the OODA Loop framework" query="Explain John Boyd's OODA Loop framework and how it applies to rapid decision-making in tech and organizational strategy."/> </ElicitationsGroup>

## Notable quotes > [00:04](https://www.youtube.com/watch?v=Pa7NIssoQMU&t=4s) Educational malpractice used to mean throwing out your syllabus in the middle of the term, and I'll argue that this fall it means keeping it, because a course that doesn't put AI in students' hands is preparing people for a world that no longer exists. — Vincent Nestler

> [02:02](https://www.youtube.com/watch?v=Pa7NIssoQMU&t=122s) If AI enables a process to become better, faster, safer, or cheaper without compromising other standards, business leaders have a fiduciary duty to adopt it. — Vincent Nestler

> [05:15](https://www.youtube.com/watch?v=Pa7NIssoQMU&t=315s) Hirable now means work that neither a human alone nor an AI alone can do, because if a person can do the task alone then AI does it cheaper, and if AI can do it alone then nobody gets hired. — Vincent Nestler

> [05:15](https://www.youtube.com/watch?v=Pa7NIssoQMU&t=315s) That definition, not a list of tools, is what has to dictate what we teach. — Vincent Nestler

> [05:15](https://www.youtube.com/watch?v=Pa7NIssoQMU&t=315s) What earns that bar is the skill Kasparov found after Deep Blue, where a weak player with an ordinary engine and a better process beat a strong player with a strong computer and a worse one. — Vincent Nestler

> [05:15](https://www.youtube.com/watch?v=Pa7NIssoQMU&t=315s) So the thing worth grading, and the thing worth doing at work, is the process by which a person orchestrates the machine. — Vincent Nestler

> [16:06](https://www.youtube.com/watch?v=Pa7NIssoQMU&t=966s) Professors will say it's cheating if you use AI. Actually, it's the professor that's cheating... because you're using the same damn syllabus that you had last semester and you're expecting it to work again... It's like you know that the answers are out there and you still give the same test. — Vincent Nestler

> [16:06](https://www.youtube.com/watch?v=Pa7NIssoQMU&t=966s) It's educational malpractice if you don't do AI in the classroom, no matter what subject you're teaching... or you're going to have people with $200,000 of debt and no way to pay it off. — Vincent Nestler

## Mentioned Vincent Nestler, California State University, San Bernardino, Thomas Edison, EnGadget, Shopify, Garry Kasparov, IBM Deep Blue, ChatGPT, Perplexity, Claude, John Boyd, Jensen Huang, Nvidia, OpenClaw, Heroku, Discord, Hermes, thehorizon.org

## Video description Vincent Nestler, Professor, Director, California State University, San Bernardino

Educational malpractice used to mean throwing out your syllabus in the middle of the term, and I'll argue that this fall it means keeping it, because a course that doesn't put AI in students' hands is preparing people for a world that no longer exists. The bar has moved. Hirable now means work that neither a human alone nor an AI alone can do, because if a person can do the task alone then AI does it cheaper, and if AI can do it alone then nobody gets hired. That definition, not a list of tools, is what has to dictate what we teach. What earns that bar is the skill Kasparov found after Deep Blue, where a weak player with an ordinary engine and a better process beat a strong player with a strong computer and a worse one. So the thing worth grading, and the thing worth doing at work, is the process by which a person orchestrates the machine. I'll show what happened when I acted on that: mid-semester I threw out the syllabus for two undergraduate cybersecurity classes and rebuilt them around OpenClaw when the framework was weeks old, then ran an open-source telemetry range on a single Heroku dyno for 69 teams and 2,518 events over 15 days, with students running that same framework on offense while defending university-hosted LLM credentials under live attack. You'll get the failures with the wins, including one hardened firewall no team breached and a majority of teams who never got past reconnaissance, plus the process-trace rubric we use to grade the orchestration instead of the answer and a two-tier adoption path that starts on a community-account budget and scales to on-premises inference. Come if you own a curriculum or a training program and you're trying to work out how fast you're allowed to move, and I'll tell you the part I didn't plan, which is that several students were interviewed and hired before the capstone even ended.

[Source](https://www.youtube.com/watch?v=Pa7NIssoQMU)

中文简报

初级岗位锐减66%，加州州立大学弃用大纲转向AI智能体实训 在AI广泛替代人力的当下，教授静态课程无异于教育失职；如今的就业能力取决于掌握人机协同编排，而非单独完成任务。 Vincent Nestler认为，在AI正在取代基础初级劳动力的经济环境中，传统的多年制课程更新周期辜负了学生。借鉴卡斯帕罗夫法则（Kasparov's Law），他将就业能力重新定义为协同编排人类或AI单独均无法完成的AI工作流。Nestler通过自身教学实践印证了这一点：他详述了自己在学期中途废除网络安全教学大纲，转而部署实时的自主OpenClaw遥测靶场，最终助力学生迅速实现就业。他最后呼吁学术界引入OODA循环方法并整合现实世界的AI工具，而非继续死守过时的课程体系。 在AI快速重塑就业市场的背景下，继续恪守僵化的传统教学大纲已构成“教学失职”，市场与效率驱动的AI正作为替代性技术直接冲击用工市场，导致初级岗位需求锐减了66%。 (2:02) “可雇佣”的门槛已被重塑为“单靠人类或单靠AI均无法独立完成的任务”；卡斯帕罗夫法则表明，通过卓越流程协同普通AI的人类能战胜单打独斗的顶尖个体，教育重心必须转向人机编排能力。 (5:15) 高校传统2至6年的课程审批周期注定导致教学与产业严重脱节，教育者必须采用军事上的OODA循环（观察、调整、决策、行动）对课程内容开展敏捷的实时迭代。 (10:02) 加州州立大学圣贝纳迪诺分校在学期中途引入刚发布数周的OpenClaw框架打造网络安全攻防靶场，学生通过掌控自主AI代理的实操经历，在毕业前成功斩获工作录用。 (10:02)

Reading summary

Vincent Nestler argues that higher education commits educational malpractice by clinging to static syllabi while AI rapidly transforms the labor market, urging universities to adopt continuous curriculum updates based on Kasparov's Law and military decision frameworks. Market Pressures and Labor Displacement (2:02) Nestler identifies two non-negotiable drivers forcing rapid AI adoption: market competition and military dominance. From a corporate perspective, executives have a fiduciary obligation to implement AI whenever it renders a workflow better, faster, safer, or cheaper. Unlike earlier technological revolutions such as the automobile—which eliminated horse-drawn transport while generating new human occupations in manufacturing, sales, and repairs—Nestler characterizes AI as a direct replacement technology for human labor. As an example, he highlights that EnGadget eliminated its entire South American Spanish editorial translation department within three months of deploying AI translation tools. Redefining Employability and Kasparov's Law (5:15) The widespread implementation of AI has sharply constrained entry-level employment, marked by a 66% decline in entry-level job postings despite a 177% increase in LinkedIn profiles listing AI competencies. Citing policies like Shopify's current hiring benchmark—which requires managers to prove a role cannot be accomplished by AI before approving human talent—Nestler contends that modern employability is restricted solely to tasks that neither a human alone nor an AI alone can perform. Entry-level IT helpdesk positions, historically the starting point for cybersecurity careers, have been largely superseded by models like ChatGPT, Claude, and Perplexity diagnosing system issues locally. Human-AI Orchestration via Kasparov's Law (5:15) To build resilient employability, Nestler points to Kasparov's Law, derived from chess grandmaster Garry Kasparov following his defeat to IBM's Deep Blue. Kasparov observed that a weak human operating an ordinary AI under a superior workflow consistently outperforms either a strong human or a sophisticated AI working in isolation. Rather than training students on isolated software tools that quickly become obsolete, Nestler argues educational institutions must prioritize the orchestration process, guiding students to combine human oversight with automated systems. The Instructional OODA Loop in Practice (10:02) Because traditional academic curriculum reviews span two to six years—ensuring course material is outdated before students graduate—Nestler advocates using John Boyd's OODA Loop (Observe, Orient, Decide, Act) to modify courses in real time. At California State University, San Bernardino, Nestler discarded his pre-planned syllabus mid-semester to incorporate OpenClaw (formerly Clawbot/Moltbot). In a capstone lab, red and blue teams deployed OpenClaw agents, integrated Discord channels for live exploit telemetry, and configured automated agents to operate overnight. Nestler noted that an international student on a visa, previously rejected across 500 job applications, secured immediate callbacks and a graduation-day job offer after updating her resume with these hands-on orchestration skills. Re-Evaluating Academic Integrity and Syllabi (16:06) Nestler disputes the narrative that students are the primary culprits of academic dishonesty when employing generative AI. Instead, he claims instructors commit their own breach of integrity by reusing outdated syllabi and delivering the same tests year after year despite knowing answers are automated. To prevent graduates from leaving university with up to $200,000 in debt without marketable skills, Nestler stresses that faculty across all disciplines have an ethical responsibility to embed real-time AI tools, noting his own department is already shifting from OpenClaw to Hermes in step with ongoing industry developments. Nestler concludes that universities must discard rigid syllabus conventions in favor of rapid, iterative curriculum redesigns aligned with AI orchestration. However, the talk does not address how institutional accreditors or risk-averse academic administrations will resolve the tension between mandatory syllabus contracts and real-time curriculum adjustments.

精炼阅读摘要

讲者Vincent Nestler指出，在人工智能迅速重塑劳动力市场的背景下，高校恪守僵化教学大纲已构成“教育渎职”；教育机构必须引入卡斯帕罗夫法则与OODA循环，通过培养人机协同流程使毕业生具备真正的就业竞争力。 市场法则与替代型技术的就业冲击 (2:02) Nestler指出，驱动AI技术快速普及的背后存在两大不可逆力量：一是市场力量，当AI能让流程变得“更好、更快、更安全或更廉价”时，企业基于受托责任必须采纳，否则面临被淘汰的风险；二是军事竞争，全球主要国家均将AI视为维持霸权地位的核心能力。他反驳了AI仅是类似汽车取代马车的“效率工具”这一说法。他认为汽车淘汰了马车夫但创造了工厂与维修岗位，本质上是提升了人类工种；而AI则是直接取代人类劳动的“替代型技术”。例如EnGadget南美办公室引入AI翻译工具后，其整个西班牙语编辑翻译团队在三个月内全部被裁撤。 重构可雇佣标准与卡斯帕罗夫法则 (5:15) 随着劳动力市场的剧烈变化，入门级岗位发布量已暴跌66%，而领英上标注AI技能的个人主页猛增177%。Nestler援引Shopify的用人政策调整说明就业标准的变化：两年前管理者倾向于证明AI能胜任特定工作，如今则必须证明该岗位是AI无法胜任的极少数独特角色才批准招聘。他提出，新的可雇佣标准必须是“人类单独无法完成、且AI单独也无法完成”的任务。单靠人类的任务会被更便宜的AI替代，而AI能独立完成的任务则不再需要雇员，过去作为网络安全起点的IT技术支持（Helpdesk）岗位，已被ChatGPT、Claude等可直接分析本地文件的工具取代。 为应对这种局面，Nestler引入国际象棋大师Garry Kasparov提出的法则：一个水平平庸的人类配合普通AI，若借助更优越的流程与工作流，其表现将持续超越单打独斗的顶尖AI或顶尖人类。因此，教育的核心目标不能停留在教授特定静态工具的操作，而必须转向指导学生掌握人类与AI的高效编排和流程协同能力。 以OODA循环替代传统课程设计 (10:02) 传统高校的课程审批流程周期长达2至6年，涵盖开课申请、教材挑选、幻灯片制作与学术委员会审查等步骤。Nestler认为，在AI技术日新月异的现实下，这种滞后的审批周期必然导致学生毕业时所学知识早已过时。教育者应采纳约翰·博伊德（John Boyd）的OODA循环（观察、定位、决策、行动），依据产业现实对课程内容进行实时动态调整。 在加州州立大学圣贝纳迪诺分校（California State University, San Bernardino）春季本科网络安全课程中，Nestler在该学期中途推翻了既定大纲，将英伟达首席执行官Jensen Huang高度评价的开源框架OpenClaw（前身为Clawbot/Moltbot）直接引入教学。在第13周的实践项目中，蓝队学生从零搭建并防御网络，红队学生则利用OpenClaw代理攻击高校托管的LLM凭证。系统通过Discord频道实时跨队伍同步漏洞与凭证，促使防御端即时响应，学生还配置API密钥使自主代理在夜间持续执行攻防。一名投递500多次无果的持签证国际学生，在将这套自主AI编排经验加入简历后立即获得面试回应，并在毕业当天拿到了工作邀约。 重新审视学术作弊与高校教师职责 (16:06) Nestler重新审视了高校关于使用AI构成“学术作弊”的传统争议。他认为，真正作弊的反而是墨守成规的大学教授：在明知技术已经颠覆现实、网络上已有现成答案的情况下，教授们依然重复使用往期的大纲和考卷，这是在逃避教育责任。若大学任由学生背负高达20万美元的学费债务毕业，却无法教授符合当前就业市场需求的核心技能，这种静态教学模式就是严重的教育渎职。为了维持教学的实时更新，Nestler表示其课程正随着OODA循环周期，从OpenClaw转向Hermes等更新的工具框架，并通过thehorizon.org等平台持续跟踪行业技能指标。 Nestler的核心论点强调，高校必须打破将教学大纲视为僵化契约的传统，全面拥抱以流程为中心的动态教学。然而，这一模式对大学既有的行政审核体制、教师持续自我迭代的能力以及跨学科标准化考核提出了极高的现实挑战，如何在保障学术严谨性的同时实现大规模敏捷化教学仍是一个未解决的制度难题。

Detailed notes — zh

## Background 该内容由 Ai4 于 2026 年 8 月 14 日发布在 YouTube 平台，主讲人为加州州立大学圣贝纳迪诺分校教授兼主任文森特·内斯特勒（Vincent Nestler）。页面简介指出，沿用陈旧教学大纲已成为新型“教学渎职”，本演讲分享了他学期中途重构网络安全课程、引入 OpenClaw 智能体对抗攻防并依据“卡斯帕罗夫法则”重塑人机协作流程，从而使学生在毕业前即获录用的实战经验与教学框架。 ## TL;DR 内斯特勒教授指出，在市场竞争与初级岗位大量消逝的冲击下，现代“可雇佣”的核心标准在于完成人类或 AI 均无法单独胜任的复合任务。他援引卡斯帕罗夫法则，论证普通人类搭配普通 AI 与卓越流程可以击败顶尖单体系统，因此教育重心必须从记忆标准答案转向训练人机协同与编排纠错能力。为打破传统高校长达六年的僵化课程开发周期，他倡导运用 OODA 循环实时捕捉产业需求，并在网络攻防顶石项目中引入开源智能体框架实现全天候协同防御。最后，他警示墨守成规、拒绝更新大纲的教育者才是真正的“作弊”与“渎职”，敦促高等教育全面转向 AI 赋能以避免毕业生产生严重的生存与学贷危机。 ## Atmosphere & style 演讲极具冲击力与紧迫感，主讲人以打破学术陈规的实战家姿态登场，言辞犀利直接且直击传统教学体制的弊病。叙事节奏紧凑抓人，将残酷的就业现实、生动鲜活的商业案例与学生从百封拒信到毕业即签约的逆袭故事交织呈现。整场分享充满激情与警示意味，宛如一场在教育界与产业界十字路口发出的“山巅呐喊”，令听众深切感受到技术范式转移的迅猛与教学变革刻不容缓的震动。 ## Key takeaways - [00:04](https://www.youtube.com/watch?v=Pa7NIssoQMU&t=4s) 传统教学将大纲视作不可违背的刚性契约，但在技术剧烈迭代的时代，固守陈旧大纲才是真正的'教学渎职'（Educational Malpractice）。 - [01:52](https://www.youtube.com/watch?v=Pa7NIssoQMU&t=112s) AI 政策与伦理讨论必须基于两大底层现实：市场经济的信托责任（多快好省的必然选择）与大国间的军事军备竞赛。 - [01:52](https://www.youtube.com/watch?v=Pa7NIssoQMU&t=112s) 大学本科赖以生存的初级入门岗位（Entry-level jobs）已有 66% 彻底消失，高校必须直接以培养人机协同的'中级技能'为目标。 - [03:48](https://www.youtube.com/watch?v=Pa7NIssoQMU&t=228s) AI 是对人类智力劳动的'替代性技术'而非单纯的辅助工具，正如汽车为人类创造了新工作，但并没有为马创造新工作。 - [03:48](https://www.youtube.com/watch?v=Pa7NIssoQMU&t=228s) 企业招聘逻辑已彻底逆转，以 Shopify 为代表的公司要求新增雇佣前必须证明该岗位是'AI 无法独自胜任'的特殊人才。 - [06:53](https://www.youtube.com/watch?v=Pa7NIssoQMU&t=413s) 新时代'可雇佣（Hirable）'的标准是完成'单凭人类自身无法完成，且单凭 AI 自身也无法完成'的任务，成为 AI 增强型人才。 - [08:43](https://www.youtube.com/watch?v=Pa7NIssoQMU&t=523s) 卡斯帕罗夫法则表明：普通棋手搭配普通 AI 引擎加上卓越的人机协同流程，能够击败顶尖单体 AI，教育核心应转向人机流程编排。 - [10:18](https://www.youtube.com/watch?v=Pa7NIssoQMU&t=618s) 传统高校长达六年的课程立项与培养周期已彻底失效，教学必须引入快速迭代的 OODA 循环（观察、定位、决策、行动）。 - [11:38](https://www.youtube.com/watch?v=Pa7NIssoQMU&t=698s) 在网络安全顶石对抗项目中引入 OpenClaw Agent 框架，实现了 24/7 自动化漏洞探测、情报共享与协同攻防实战。 - [13:31](https://www.youtube.com/watch?v=Pa7NIssoQMU&t=811s) 将真实的 AI Agent 自动化编排与攻防经历写入简历，能帮助受阻学生（如投递 500 份零回复的留学生）在毕业前迅速斩获 Offer。 - [15:28](https://www.youtube.com/watch?v=Pa7NIssoQMU&t=928s) 真正'作弊'的是畏惧改变、沿用旧大纲与传统试卷的教师，他们剥夺了学生适应现代就业市场的生存能力。 - [16:59](https://www.youtube.com/watch?v=Pa7NIssoQMU&t=1019s) 所有学科均须刻不容缓地融入 AI 实践，否则毕业生将面临背负 20 万美元高额学贷却缺乏就业还贷能力的系统性风险。 ## Detailed walkthrough ### [00:04](https://www.youtube.com/watch?v=Pa7NIssoQMU&t=4s) 爱迪生试餐测试与中途修改教学大纲的“禁忌” 演讲伊始，加州州立大学圣贝纳迪诺分校（California State University, San Bernardino）的文森特·内斯特勒（Vincent Nestler）教授分享了一个关于托马斯·爱迪生（Thomas Edison）的著名典故：爱迪生在面试雇佣员工之前，习惯先带他们共进晚餐并观察他们用餐。如果有人在尝都没尝之前就直接往食物里撒盐，爱迪生绝不会录用他们——因为这意味着他们是在基于“主观假设”而非客观事实行事，而爱迪生需要的是在实验室里不妄下假设、尊重事实的实干家。为了避免自己也“盲目撒盐”，内斯特勒在开场时现场调查了听众构成，摸清台下哪些人是单纯教授 AI 的学者、哪些人正试图将 AI 融入其他学科教学，以及哪些人来自工业界，以便针对性调整演讲节奏。 内斯特勒随即抛出学术界的一大痛点：在大学里，课程教学大纲（Syllabus）在法律层面等同于一份严肃的契约，如果教授在学期进行到一半时擅自推翻并修改大纲，在传统观念中这属于严重的“教学事故”（Educational Malpractice）和绝对的禁忌。然而，内斯特勒却毅然在学期中途彻底推翻了大纲。正是这一离经叛道的举措，让他的学生们在毕业之前、甚至就在当个学期内便被企业争相录用。 ```mermaid flowchart LR accTitle: 传统教学与 AI 驱动教学模式对比 subgraph Traditional["传统教学模式"] A["固定教学大纲<br/>视作刚性法律契约"] --> B["严禁中途调整<br/>滞后于技术演进"] B --> C["毕业生技能过时<br/>沦为'教育失职'"] end subgraph Modern["动态适应模式"] D["基于产业现实<br/>学期中果断重构大纲"] --> E["全流程引入 AI<br/>人机协同高频对抗"] E --> F["未毕业即获录用<br/>具备真实实战力"] end ``` ### [01:52](https://www.youtube.com/watch?v=Pa7NIssoQMU&t=112s) 讨论 AI 政策的两大底层现实：市场力量与军事竞争 内斯特勒强调，当前许多人讨论 AI 政策和道德规范时往往是在“自说自话、各说各话”，脱离了根本现实。若想制定真正有效的政策或引导有意义的伦理讨论，脑海中必须时刻紧扣两大底层逻辑，否则所有放缓 AI 发展的呼吁都将毫无成效： 1. **市场经济规律（Market Forces）**：如果 AI 能够以更优、更快、更安全、更便宜（Better, Faster, Safer, Cheaper）这四项指标中的任意一项完成某项工作，且不以牺牲其他三项为代价（事实上 AI 往往同时满足这四项），那么企业出于信托责任（Fiduciary Responsibility）就必须强制执行并采纳 AI。因为如果企业不这么做，竞争对手也会这么做，后果不仅是丢掉一部分岗位，而是整家公司倒闭、失去所有饭碗。 2. **军事与地缘现实（Military Reality）**：无论客观事实究竟如何，全球所有大国目前都深信一条铁律——谁拥有最强的 AI，谁就将成为主宰未来的绝对强权。全球范围内的军备竞赛已经白热化。 市场力量的冲击已经显而易见：职业社交平台 LinkedIn 上声称自己掌握 AI 技能的人数激增了 177%，所有人都在吹嘘自己懂 AI；与此同时，大学本科教育一直赖以生存的支柱——**初级入门岗位（Entry-level jobs）已有高达 66% 彻底消失**。高校不能再指望把学生培养成只能干入门杂活的人，必须直接将培养目标锚定在具备人机协同能力的“中级技能（Midlevel）”水平。 ### [03:48](https://www.youtube.com/watch?v=Pa7NIssoQMU&t=228s) AI 是人类的“替代性技术”：Shopify 雇佣逻辑的彻底颠覆 内斯特勒驳斥了那些试图粉饰太平的温水煮青蛙论调。许多人常说：“新技术总是会创造出更多、更好的工作，汽车诞生时淘汰了马车夫，却带来了汽车工厂工人、经销商、维修工等一整个庞大的运输产业。”内斯特勒一针见血地指出：**汽车的出现确实为‘人类’创造了更多更好的工作，但它并没有为‘马’创造更多更好的工作**。AI 绝非仅仅是颠覆性技术，它本质上是对人类智力劳动的“替代性技术（Replacement Technology）”。一旦 AI 在某一处替代了某个专业工种，就意味着它在全球范围内永久替代了该岗位。 内斯特勒以知名科技媒体 Engadget 的遭遇为例：约一年前，Engadget 在南美设立办公室并宣布引入 AI 辅助编辑流程；仅仅三个月后，负责将英文报道翻译为西班牙语的整个编辑团队被全员裁撤。这些失业员工根本无处可去，因为无论他们去哪表面试翻译岗位，用人单位都会反问：“我花 20 美元买个 ChatGPT 账号就能搞定，为什么还要雇你？” 大家常听的一句话是：“AI 不会淘汰你，使用 AI 的人会淘汰你。”但故事的另一面更为残酷。大约一年前，市值千亿美元、坐拥约一万名员工的电商巨头 Shopify 首席执行官托比亚斯·吕特克（Tobias Lütke）向全员下达通牒： - 全公司上上下下所有人必须在日常工作中使用 AI，这直接计入绩效考核； - **各部门在向公司提出新增雇佣任何人之前，必须先向 CEO 证明这项工作‘AI 无法完成’**。 这一要求标志着整个就业市场逻辑的彻底反转。两年前，如果员工向管理层提议用 AI 取代某个营销岗位，老板会要求员工“证明 AI 真的能做”；而如今，逻辑完全倒转为“向我证明这个求职者是百年一遇的独角兽、具备 AI 无法替代的特殊才能，否则免谈招聘”。 ### [06:53](https://www.youtube.com/watch?v=Pa7NIssoQMU&t=413s) “可雇佣”的新定义：人与 AI 协作的增强能力 面对这一现实，内斯特勒给出了教育工作者必须奉为圭臬的全新人才培养基准——**“可雇佣（Hirable）”的全新定义**： > 一个求职者要想具备被雇佣的价值，必须能够完成**“单凭人类自身无法完成，且单凭 AI 自身也无法完成”**的任务，即必须成为“AI 增强型（AI-augmented）人才”。 如果一项工作仅凭人类自己就能完成，企业必然选择使用 AI 替代，因为 AI 更加多快好省；但如果一项工作完全可以全自动由 AI 搞定，企业同样不需要招人。内斯特勒以自己任教的网络安全（Cybersecurity）专业为例：过去网络安全毕业生的典型起步路径是技术支持台（Help Desk）岗位，但内斯特勒直言自己已有整整两年没有找过任何人工客服或技术支持。无论遇到什么配置报错，他都是直接使用 ChatGPT、Perplexity 或者 Claude 处理。随着 Claude 等具备操作系统级操作能力的 AI 工具出现，用户只需对 Claude 说一句“这里坏了，请帮我修复”，AI 就能直接在电脑后台完成故障排查与修复，用户甚至无需像以前那样把电脑远程控制权限移交给海外呼叫中心的人工客服。 | 维 度 | 传统岗位定位 | AI 时代全新要求 | |:--- |:--- |:--- | | **工作产出模式** | 人类独立手工完成任务 | 依靠人机协同完成超越单一主体的复合任务 | | **准入门槛考核** | 基础入门级技能（如 Help Desk 支持） | 中级技能起步，熟练调度与纠错 AI | | **企业招聘前提** | 业务扩张即招募常规人力 | 必须证明该岗位是“AI 无法独自胜任的独角兽” | | **成本与效率基准** | 按人头工时与薪酬评估 | “更优、更快、更安全、更便宜”四位一体 | ### [08:43](https://www.youtube.com/watch?v=Pa7NIssoQMU&t=523s) 卡斯帕罗夫法则：平庸 AI 搭配卓越流程如何击溃顶尖对手 既然 AI 如此强大，人类的核心价值究竟何在？内斯特勒援引了国际象棋大师加里·卡斯帕罗夫（Garry Kasparov）在 1997 年败给“深蓝”（Deep Blue）超级计算机后的经典发现（卡斯帕罗夫在 DEF CON 大会上曾就此发表过深刻演讲）： ```mermaid flowchart TD accTitle: 卡斯帕罗夫法则与人机协作胜负关系 subgraph Match1["对局一：顶级人类 vs 顶级 AI"] H1["顶尖人类大师"] vs1["VS"] AI1["顶尖国际象棋 AI"] AI1 -->|毫无悬念碾压| W1["顶尖 AI 获胜"] end subgraph Match2["对局二：卡斯帕罗夫法则（流程致胜）"] Comb["弱势/普通棋手<br/>+ 普通 AI 引擎<br/>+ 卓越人机协作流程"] vs2["VS"] AI2["顶尖国际象棋 AI"] Comb -->|流程与策略调优| W2["人机协作组合获胜"] end ``` 卡斯帕罗夫发现：如果是顶尖人类大师单挑顶尖 AI，AI 每次都会毫无悬念地取胜，人类毫无胜算；但如果是一个棋艺相对平庸的普通人类，搭配一台性能普通的 AI，只要他们之间拥有一套**极其出色的协作流程（Superior Process）**，就能够击败最顶尖的单体 AI。 这正是现代教育的核心使命所在：教育的重点不再是死记硬背专业答案，而是教会学生如何与 AI 互动——如何提出高质量问题、如何批判性审视 AI 给出的答案、如何提出追问与澄清细节、如何根据上下文调整偏差，以及如何在 AI 产生幻觉或遇到能力瓶颈时，发挥人类独特的创造力予以弥补与纠偏。这种关系就如同《星际迷航》中理性严谨的斯波克（Spock）与直觉决断的柯克船长（Kirk）之间的协作。 ### [10:18](https://www.youtube.com/watch?v=Pa7NIssoQMU&t=618s) 六年课程周期已死：用 OODA 循环实时捕捉产业需求 传统的教学设计（Instructional Design）体系在飞速迭代的 AI 浪潮面前已经彻底崩溃。内斯特勒详细拆解了一个传统大学课程从构思到落地的冗长流程： - 前期调研、研读最新教材、反复开会争论采用何种幻灯片与教学方案； - 提交大学课程委员会（Curriculum Committee）审批；如果不满意被打回修改，又是几个月； - 历时整整两年终于将课程正式编入大学四年制培养计划； - 等待第一批学生从大一学到大四毕业。 从最初设计到最终将具备该技能的学生推向社会，**整整耗时六年（甚至即使是两年的微调周期也已经跟不上节奏）**。在 AI 时代，六年前锁定的行业靶心早就消失得无影无踪。 为了生存，高校教学必须引入美国空军用于应对瞬息万变战局的 **OODA 循环（观察 Observe、定位 Orient、决策 Decide、行动 Act）**： - **观察（Observe）**：实时洞察劳动力市场正在渴求什么、不再需要什么过去的值钱技能，以及出现了哪些此前从未存在过的全新技术； - **定位与决策（Orient & Decide）**：迅速学习这些新技术，评估如何将其落地并决定教学路径； - **行动（Act）**：以最快速度将其直接注入当下的课程实践中，绝不推迟到下学期。 ### [11:38](https://www.youtube.com/watch?v=Pa7NIssoQMU&t=698s) 实战演练：将 OpenClaw 引入加州州立大学网络对抗顶石项目 在当季春季学期中途，开源 Agent 自动化框架 OpenClaw 横空出世（该框架经历了 Clawbot、Moltbot 等命名演变）。英伟达（NVIDIA）CEO 黄仁勋（Jensen Huang）公开评价该框架“可能是软件发布史上最重要的软件发布之一”。内斯特勒面临两个选择：是墨守成规把新技术拖到下一学年再教，还是立即在二三月份学期中途紧急调转方向？内斯特勒毫不犹豫地选择了后者。 加州州立大学圣贝纳迪诺分校拥有一门极为硬核的毕业设计顶石项目（Capstone）： - **蓝队构建防御**：一组学生必须从零开始搭建、配置并加固一套完整的企业级网络环境； - **红队真实渗透**：到了学期的第 13 周，内斯特勒任教的渗透测试（Penetration Testing）班级学生将作为红队，对蓝队搭建的网络发起长达整整一周的实弹攻防对抗。 由于学生对亲手搭建的网络倾注了心血，极度抗拒被人攻破，这种强烈的情感投入极大提升了教学成效。而在这次对抗中，内斯特勒全面引入了 OpenClaw 框架： - **蓝队防守**：利用 OpenClaw 辅助分析网络拓扑、自动化修复配置漏洞、排查安全隐患并在被攻击时进行实时遥测分析； - **红队进攻**：利用 OpenClaw 自动化编写并调度渗透攻击脚本； - **自动化链路监控**：OpenClaw 实时对接 Discord 机器人，将每一次渗透攻击与防御事件汇聚并推送到统一的大屏遥测仪表盘（Dashboard）中，实现了全透明的攻防事件可视化。 ### [13:31](https://www.youtube.com/watch?v=Pa7NIssoQMU&t=811s) 实战攻防协同与自动化智能体带来的求职突破 在网络安全实战攻防演练中，多智能体协同展现出了惊人的效率。当学生部署的某个 OpenClaw 智能体发现目标漏洞时，它会实时将该漏洞情报广播给所有其他智能体；刹那间，所有客户端会同步针对其他团队的防线发起统一的凭据爆破或漏洞利用测试。这种完全实时的情报共享与协同攻击堪称一场“壮观的演练”。学生们通过这门课程学会了如何编写自动化脚本、配置 API 密钥与 Agent 交互，并让智能体在学生入睡时依然保持 24 小时不间断运行与防御。 这些前沿的实战经历直接转化为了简历上的核心竞争力。演讲者分享了一个真实的故事：学期初有一位持学生签证的印度籍国际生向他求助，表示自己投递了 500 多份简历却连一次面试回电都没有收到。教授指导她将课程中参与的 AI 自动化攻防与 Agent 编排实战项目写入简历。结果第二天她就收到了面试回复，甚至一度怀疑这是个恶作剧。在毕业典礼走过红毯时，这名非美国籍学生兴奋地跑向教授报喜，告知自己已经成功拿到 Offer 并被正式录用，而雇主看中的正是她在 AI 领域的实战经历。这次突破正是源于教学团队敢于“做错的事”——没有死板遵从既定教学大纲，而是果断根据真实产业需求调整教学方向。 ```mermaid flowchart TD accTitle: AI协同攻防与求职转化流程 subgraph Offensive["智能体协同攻防实战"] AgentA["OpenClaw 智能体 A 探测漏洞"] BroadCast["即时广播漏洞与凭据"] AgentAll["全体智能体同步发起自动化打击"] AgentA --> BroadCast BroadCast --> AgentAll end subgraph Career["就业能力转化"] Skills["掌握 API 编排与 24/7 自动化"] Resume["重构简历突出 AI 实战经验"] Offer["从 500 拒信到毕业即获聘 Offer"] Skills --> Resume Resume --> Offer end AgentAll --> Skills ``` - **投递转化悬殊**：一名非美国公民学生从投递 500 份简历零回复，到加入 AI 实战经历后次日即获面试邀约并在毕业前成功获聘。 - **全天候自主运行**：学生掌握了让 AI Agent 在无人值守（睡眠期间）自主执行安全扫描与防御响应的自动化技术。 --- ### [15:28](https://www.youtube.com/watch?v=Pa7NIssoQMU&t=928s) 重新定义“教学渎职”：教授与学生的角色重塑 学生进入大学校园，无论口头表达与否，内心都寄托着最质朴的期望：通过高等教育掌握立足社会的技能，从而能够养家糊口、买房置业，成为对社会有贡献的一员。如果高校教育工作者拒绝顺应 AI 对劳动力市场的重塑，拒绝主动学习自身尚未掌握的技术，就是在违背这一基本契约。过去让教授们取得学术或职业成功的传统路径，已经完全不适用于今天的年轻一代，教师群体必须经历彻底的“重新学习”（Relearning）。 从这个维度来看，教学界对“作弊”的定义需要彻底颠覆。许多教授指责学生使用 AI 是作弊行为，但现实中“真正作弊的反而是教授自己”。有些教师年复一年沿用上一学期的陈旧教学大纲，因为畏惧改变而固步自封；明知道标准答案在外界早已唾手可得，却依然布置一成不变的试卷。这种漠视时代巨变的行为就是典型的“教学渎职”（Educational Malpractice）。在教学进化路径上，讲者运用 OODA 循环（观察-调整-决策-行动）不断迭代工具栈，尽管本学期使用了 OpenClaw，后续也将果断迁移至更新的 Hermes 等技术体系。 - **时代路径分歧**：塑造老一代学者成功的经验与技能，已无法直接为当代大学生的未来职业生涯铺路。 - **教职人员的作弊本质**：不愿根据 AI 演进重构课程，沿用旧大纲与传统考卷，是对学生未来生存能力的严重侵害。 --- ### [16:59](https://www.youtube.com/watch?v=Pa7NIssoQMU&t=1019s) 面试资源库、行业雷达与“山巅呐喊” 为帮助教育者和求职者在信息过载与技术高速迭代中迅速找准坐标，讲者公开了一个包含核心产业洞察的 Google Drive 共享资源库。该资源库收录了当下真实 AI 岗位及各类企业在招聘时最常考查的问题清单，包括考察应聘者的 AI 技术栈构成、日常使用场景以及近期使用过的最佳提示词（Prompt），并系统梳理了未来急需的技能与岗位缺口。此外，讲者还推荐了 The Horizon 平台每日发布的 90 秒精炼 AI 资讯简报，便于师生快速掌握全球技术前沿与底层链接。 在最后的现场问答互动环节，讲者再次发出铿锵有力的警示：不论教授何种学科专业，在课堂中融入 AI 实践已刻不容缓。如果不立即在所有学科领域推动这场教学变革，大学培养出的将是一批背负着 20 万美元高额学生贷款、却因缺乏现代生产力工具掌控力而根本无力偿还的毕业生。他正竭尽全力在“每一座山巅上呐喊”，敦促整个教育系统打破惯性、全面拥抱 AI。 - **AI 岗位常见面试问题**： - “你在日常工作中如何使用 AI？” - “你的 AI 工具栈（AI Stack）是由哪些构成的？” - “请分享一个你最近编写并验证过的最佳 Prompt 案例。” - **资讯获取渠道**：推荐通过 `thehorizon.org` 获取每日约 90 秒的高效 AI 新闻简报及拓展文献。 - **学贷危机预警**：若不推行 AI 赋能教育，学生将面临背负高达 200,000 美元学贷却在就业市场缺乏偿还能力的系统性风险。 ## Notable quotes > [00:04](https://www.youtube.com/watch?v=Pa7NIssoQMU&t=4s) 一个课程如果不把 AI 交到学生手中，就是在为一个人群准备一个已经不复存在的世界。 — Vincent Nestler > [03:48](https://www.youtube.com/watch?v=Pa7NIssoQMU&t=228s) 汽车的出现确实为‘人类’创造了更多更好的工作，但它并没有为‘马’创造更多更好的工作。 — Vincent Nestler > [03:48](https://www.youtube.com/watch?v=Pa7NIssoQMU&t=228s) 我花 20 美元买个 ChatGPT 账号就能搞定，为什么还要雇你？ — Vincent Nestler > [03:48](https://www.youtube.com/watch?v=Pa7NIssoQMU&t=228s) 各部门在向公司提出新增雇佣任何人之前，必须先向 CEO 证明这项工作‘AI 无法完成’。 — Vincent Nestler > [06:53](https://www.youtube.com/watch?v=Pa7NIssoQMU&t=413s) 一个求职者要想具备被雇佣的价值，必须能够完成‘单凭人类自身无法完成，且单凭 AI 自身也无法完成’的任务，即必须成为‘AI 增强型’人才。 — Vincent Nestler > [08:43](https://www.youtube.com/watch?v=Pa7NIssoQMU&t=523s) 弱势/普通棋手 + 普通 AI 引擎 + 卓越人机协作流程，能够击败顶尖国际象棋 AI。 — Vincent Nestler > [10:18](https://www.youtube.com/watch?v=Pa7NIssoQMU&t=618s) 在 AI 时代，六年前锁定的行业靶心早就消失得无影无踪。 — Vincent Nestler > [15:28](https://www.youtube.com/watch?v=Pa7NIssoQMU&t=928s) 许多教授指责学生使用 AI 是作弊行为，但现实中‘真正作弊的反而是教授自己’。 — Vincent Nestler > [16:59](https://www.youtube.com/watch?v=Pa7NIssoQMU&t=1019s) 大学培养出的将是一批背负着 20 万美元高额学生贷款、却因缺乏现代生产力工具掌控力而根本无力偿还的毕业生。 — Vincent Nestler ## Mentioned Vincent Nestler, California State University, San Bernardino, Thomas Edison, Shopify, Tobias Lütke, Engadget, ChatGPT, Claude, Perplexity, Garry Kasparov, Deep Blue, DEF CON, Jensen Huang, NVIDIA, OpenClaw, Clawbot, Moltbot, Hermes, Discord, Heroku, The Horizon, LinkedIn ## Video description Vincent Nestler, Professor, Director, California State University, San Bernardino Educational malpractice used to mean throwing out your syllabus in the middle of the term, and I'll argue that this fall it means keeping it, because a course that doesn't put AI in students' hands is preparing people for a world that no longer exists. The bar has moved. Hirable now means work that neither a human alone nor an AI alone can do, because if a person can do the task alone then AI does it cheaper, and if AI can do it alone then nobody gets hired. That definition, not a list of tools, is what has to dictate what we teach. What earns that bar is the skill Kasparov found after Deep Blue, where a weak player with an ordinary engine and a better process beat a strong player with a strong computer and a worse one. So the thing worth grading, and the thing worth doing at work, is the process by which a person orchestrates the machine. I'll show what happened when I acted on that: mid-semester I threw out the syllabus for two undergraduate cybersecurity classes and rebuilt them around OpenClaw when the framework was weeks old, then ran an open-source telemetry range on a single Heroku dyno for 69 teams and 2,518 events over 15 days, with students running that same framework on offense while defending university-hosted LLM credentials under live attack. You'll get the failures with the wins, including one hardened firewall no team breached and a majority of teams who never got past reconnaissance, plus the process-trace rubric we use to grade the orchestration instead of the answer and a two-tier adoption path that starts on a community-account budget and scales to on-premises inference. Come if you own a curriculum or a training program and you're trying to work out how fast you're allowed to move, and I'll tell you the part I didn't plan, which is that several students were interviewed and hired before the capstone even ended. [Source](https://www.youtube.com/watch?v=Pa7NIssoQMU)

Source quotes

[0:04] Vincent Nestler: 一个课程如果不把 AI 交到学生手中，就是在为一个人群准备一个已经不复存在的世界。 [3:48] Vincent Nestler: 汽车的出现确实为‘人类’创造了更多更好的工作，但它并没有为‘马’创造更多更好的工作。 [3:48] Vincent Nestler: 我花 20 美元买个 ChatGPT 账号就能搞定，为什么还要雇你？ [3:48] Vincent Nestler: 各部门在向公司提出新增雇佣任何人之前，必须先向 CEO 证明这项工作‘AI 无法完成’。 [6:53] Vincent Nestler: 一个求职者要想具备被雇佣的价值，必须能够完成‘单凭人类自身无法完成，且单凭 AI 自身也无法完成’的任务，即必须成为‘AI 增强型’人才。 [8:43] Vincent Nestler: 弱势/普通棋手 + 普通 AI 引擎 + 卓越人机协作流程，能够击败顶尖国际象棋 AI。 [10:18] Vincent Nestler: 在 AI 时代，六年前锁定的行业靶心早就消失得无影无踪。 [15:28] Vincent Nestler: 许多教授指责学生使用 AI 是作弊行为，但现实中‘真正作弊的反而是教授自己’。 [16:59] Vincent Nestler: 大学培养出的将是一批背负着 20 万美元高额学生贷款、却因缺乏现代生产力工具掌控力而根本无力偿还的毕业生。

Source quotes

[0:04] Vincent Nestler: Educational malpractice used to mean throwing out your syllabus in the middle of the term, and I'll argue that this fall it means keeping it, because a course that doesn't put AI in students' hands is preparing people for a world that no longer exists. [2:02] Vincent Nestler: If AI enables a process to become better, faster, safer, or cheaper without compromising other standards, business leaders have a fiduciary duty to adopt it. [5:15] Vincent Nestler: Hirable now means work that neither a human alone nor an AI alone can do, because if a person can do the task alone then AI does it cheaper, and if AI can do it alone then nobody gets hired. [5:15] Vincent Nestler: That definition, not a list of tools, is what has to dictate what we teach. [5:15] Vincent Nestler: What earns that bar is the skill Kasparov found after Deep Blue, where a weak player with an ordinary engine and a better process beat a strong player with a strong computer and a worse one. [5:15] Vincent Nestler: So the thing worth grading, and the thing worth doing at work, is the process by which a person orchestrates the machine. [16:06] Vincent Nestler: Professors will say it's cheating if you use AI. Actually, it's the professor that's cheating... because you're using the same damn syllabus that you had last semester and you're expecting it to work again... It's like you know that the answers are out there and you still give the same test. [16:06] Vincent Nestler: It's educational malpractice if you don't do AI in the classroom, no matter what subject you're teaching... or you're going to have people with $200,000 of debt and no way to pay it off.

Original transcript

[0:04] Good morning, ladies and gentlemen. My name is  Vincent Nestler, as you know. Uh, I'm uh Thomas [0:11] Edison. Before I get started, Thomas Edison used  to take people that he was thinking about hiring [0:17] uh to dinner and he'd watch them eat. And if  they salted their food before they tasted it, he [0:27] wouldn't he wouldn't hire them because he figured  they were doing things based on assumption. And [0:33] he wanted people that were working in his lab  that actually had an idea about that didn't make [0:39] assumptions. So I don't want to make assumptions.  I want to find out before I salt my food. Somebody [0:46] got it. There you go. um to find out how many  of you are educators, how many are teaching AI [0:55] as the subject? How many are teaching a different  subject, but now trying to figure out how to get [1:01] AI in there? Okay, interesting. All right. Who  did I leave out? Uh uh uh industry. Yeah, some [1:08] industry folks. Okay, great. That helps me. I'll  I'll target a little bit more what I'm talking [1:15] to kind of address what's in the audience. Okay.  So, [clears throat] in academia, when you write [1:24] the syllabus, it's a legal document. And if you  change the syllabus halfway through the semester, [1:30] that's a big no no. That's a big no no. Uh but I  did. And as a result of doing that, I had students [1:39] that were getting hired before they even graduated  and in that semester. And I'll talk about that [1:45] towards the end. [clears throat] Okay. So, there's  a few things I want to make sure I I talk about [1:52] that's foundational knowledge to understanding the  situation that we're all in because some people [1:58] are talking about AI and they're talking past each  other. But there's some fundamental things that [2:02] you have to have always present in your mind when  you're talking about policy, how you're using AI, [2:08] should you or should you not use AI. Number one,  market forces. If AI can do anything better, [2:14] faster, safer, cheaper, any one of those four  things without the other three going down and [2:20] more likely it's doing all of those things. If I  could do that, I have a fiduciary responsibility [2:26] to en enact that because if I don't, my com  competition will and not only will I lose [2:32] some jobs, I'll lose all my jobs. The second is  the military reality. Whether it's true or not, [2:38] every nation on the planet believes that if they  have the best AI, they will be the dominant power [2:45] from here on out. And there is a mad rush to to  get the best AI. If you want to change AI policy, [2:55] you want to talk about ethics, you want to talk  about how to do this right, maybe we should slow [2:58] down. If you do not take those two realities  into account, you have zero chance of having [3:04] any effect. You have to put it in terms of these  two things if you want to make an impact. Now, [3:10] the market force is already showing that this is  the case. Uh, everybody on LinkedIn is saying, [3:17] "I know AI. I know AI. I know AI, too." Right?  177% increase in AI. Uh, people bragging about [3:24] it. Uh, 66% of entry level jobs gone. That's  our bread and butter in academia. The entry [3:35] level jobs. Those are that's what we're targeting  most of the time. But now we can't target that. [3:40] We actually have to target midlevel. And I'll  get I'll get to that more in a second. I also [3:48] hear this all the time with the people who are  trying to be like trying to sugarcoat stuff. Oh, [3:52] I've been through this before. Listen, more better  technology makes more better jobs. This happened [3:58] before with the automobile. It made more jobs.  Yes, it disrupted the horse and buggies. That's [4:04] right. But, you know, it it made more We had  people working in the factory. We had dealerships. [4:10] We had mechanics. It it opened up a whole  transportation industry. It was amazing. Yes, [4:17] it was disruptive, but it made more better jobs  for humans, but it didn't make more better jobs [4:22] for horses. AI is a replacement technology for  humans. It's not just a disruptive technology. [4:31] It's a replacement technology. And you have to  get that in mind. It's not just moving you out. If [4:36] AI replaces what you do as a professional in one  location, it's replaced you in the entire world. [4:48] And Gadget about a year ago had a uh had offices  in South America. They said, "We're introducing [4:54] AI to help with editorial editorial staff." Three  months later, they got rid of the entire editorial [5:00] staff that was translating the documents from  English to Spanish. Where do those folks go? [5:06] Right? Where do they go? Because wherever they're  going to go to say, "Hey, I can translate." [5:10] They're going to say, "Why? I can have GPT do  it for me for 20 bucks a month." You're going to [5:15] hear, and you've heard people, raise your hand  if you've heard people say, "AI won't replace [5:20] you. Somebody using AI will replace you." We've  heard this, right? That's one side of the coin. [5:27] The other side is that was first spoken by uh  the CEO of Shopify year ago, almost to the day. [5:37] He said first he said everybody at at shop  Shopify uh hundred billion dollar company about [5:43] 10,000 workers right he said everybody must use AI  everybody not just the cyber and security people [5:52] everybody uh it's going to be on your evals and  before asking and this is the key before asking [6:01] me to hire anybody new you must prove to me that  AI can't do it. That's a complete flip of reality. [6:13] Two years ago, if I was working at a company and  the boss said, "Hey, I want to hire somebody for [6:17] marketing." And I said, "Hey, I think we can  get AI to do the marketing thing you want to [6:22] hire somebody for." The CEO would have told me,  "Prove to me that AI could do it." Now it's proved [6:31] to me that this person is such a unicorn that  they have some special talent or ability that I [6:36] should hire them rather than have AI do it. That's  the CEO talking a year ago. That's the harbinger [6:45] of what's to come. So the new definition of  hirable, if you get nothing else from this talk, [6:53] you must take this especially as educators because  this has got to be the guidepost for you as you're [6:59] developing curriculum and as you're thinking  what am I going to fill the young minds with [7:03] that's going to be able to make them put food  on the table for them and theirs. And that's [7:09] this. The definition of hirable is you have to be  able to do something that a human alone cannot and [7:19] that AI alone cannot. You must be AI augmented.  If I can get you to do it or a human to do it, [7:26] why would I pay? All right. So maybe you're better  than the human, but the human is going to do it [7:30] using AI and it's going to be fa better, faster,  safer, cheaper. Better, faster, safer, cheaper. [7:35] Any one of those four. So you have to use AI. But  I also have to start keeping my eye on what is AI [7:44] able to do entry level jobs for the cyber cyber  is one of the subjects I teach has always been [7:52] help desk I have not used help desk in two years  chatbt and perplexity is where I go and right now [7:59] it's even better if I use claude on my computer  I can literally tell claude this is not working [8:07] can you fix it and it goes into my computer and  fixes it. I don't have to give remote access to [8:13] some human somewhere in Philippines hoping that  they're not selling my to someone my stuff. Right? [8:24] So that's the new definition of hirable. If you  know if you remember nothing else from this, [8:31] you have to take that and that has  to be your guidepost as to how you're [8:34] developing your curriculum.  Also, one of the key things I [8:43] suggest, Kasparov is last man standing against AI  in chess. Kasparov 1997 loses to Deep Blue. Gives [8:52] a great talk on Defcon. Please take a look at it  uh online on his experience there. But to make a [8:59] long story short, he said what he found was if  you take the best human against the best AI, [9:07] the best AI wins every time. Writing's on the  wall. You got no chance. But if you take a human, [9:14] even a weak human, his words, and an okay AI  with a superior process, they beat the best AI. [9:24] So our job as educators is now going  to be how do I AI enhance you? How do I [9:30] teach you how to work with AI? What is that  process of ask get the answer think about [9:36] it critically ask again get clarifying  questions figure out how to adjust and [9:42] be more creative where AI fails you as a  human um do well. It's kind of like the [9:50] uh Spock Kirk relationship, right? You have  somebody that's more logical or not exactly, but [10:02] the Air Force has a model for dealing with  a changing situation. It's called UDA, [10:09] the UDA loop. Observe, orient, decide, and act. [10:18] As an instructional designer, if I'm trying  to enact any kind of instructional design, [10:26] it's at minimum six years before I get a student  that graduates with the skill that I create. It's [10:33] two years. All right. What's the topic? Let's  talk about it. Let's look at is a textbook. Uh uh [10:38] uh. Okay. We should do it this way. No, that way.  Here's some slides. All right. That's great. Okay. [10:42] Curriculum committee. No, we don't like it. Do it  again. Okay. whatever. So then we finally get it, [10:46] we pass it through, now it's in the four-year  curriculum doop. Six years later, the target [10:53] is nowhere in sight anymore. You cannot create  uh instructional periods with a sixyear cycle. [11:06] You can't even do it with a two-year cycle at this  point. So you have to adopt this. Let me observe [11:15] what does the workforce want? What does it need?  What is it looking for? What used to be valuable [11:21] but is not anymore. What's new that I didn't even  know existed? I have to learn it and get into [11:26] the students hands as soon as possible. I have to  decide which thing I'm going to do, how I'm going [11:30] to do it, and then get it in incorporate in the  curriculum as soon as possible. So, what happens [11:38] mid semester uh in the spring? Open claw. How many  know have heard of open claw? Was clawbot then it [11:46] became moltbot and they were like moltbot and then  they became open claw. Okay. So open claw comes [11:52] out and yensen Wong uh CEO of the most valuable  corporation in world history says this is probably [12:00] the most important software release ever. So do I  say yeah I guess I'll teach it next semester. Or [12:11] do I say in February, March when he said it, okay,  we have to shift gears and we have to start doing [12:17] this right now. We did it right now. And so at  Cal State San Bernardino, we have this very fairly [12:28] unique capstone. We have a group of students that  before graduation, they have to build an entire [12:34] network, soup to nuts. They build it, secure it,  configure it, and then week 13 of the semester, [12:42] my pen test class attacks their network for the  whole week. It's glorious, right? Because they [12:48] really want to get it. The people don't want their  stuff hacked. So, there's an emotional attachment [12:52] to it. Increases the educational value of the  event. What we did this time was we let people [12:59] use Open Claw. So the red team the the blue  teamers were using open claw to help configure [13:05] analyze the network uh secure stuff find problems  and actually do real time um analysis during the [13:15] the hack and the red team was using open claw to  send attacks. Open claw was connecting to discord. [13:26] Discord was sending the information to a a  dashboard so we could see the hacks basically [13:31] happening in real time. And anytime one of the  claws that the students were working with found [13:38] the vulnerability, it shared it with all the other  claws so they can all try that password or that [13:44] vulnerability on all the other teams at the exact  same time. Was a glorious event, right? So things [13:50] worked great. Now, as a result of learning how to  automate, set up um API keys to work with the with [14:02] the agent, get the agent to do things automated,  so even when they were sleeping, it was awake [14:07] and doing everything. They put that on their  resume. I had a student, this is a true story, [14:12] had a student who came to me at the beginning of  the semester, was Indian student, non non-citizen, [14:17] was on a H-1B visa or F was on a on a visa,  student visa. Said, I kid you not, professor, [14:26] I had 500 resumes. I have not gotten one call  back. Not one. So, uh, she started, she did AI. [14:36] I said, "Here are the things you should put on  your resume. This is what we're doing." She said [14:40] the first time she put the resume out, the next  day she got a reply and she thought it was a joke. [14:46] She couldn't believe that she all of a sudden got  a reply. Um, literally on the graduation as she [14:54] was walking, she ran up to me. She said, "I got  a I got a job. They hired me because of the whole [14:59] AI stuff." Right? a person without citizenship  was able to get a job because she had the AI on [15:08] her resume for the stuff that she did because we  pivoted and we did the wrong thing. We didn't do [15:14] exactly what the syllabus said. We we pivoted to  to focus on what the real world was asking for, [15:22] making sure they got those skills by the time  they graduated. [clears throat and cough] [15:28] So when students come to us and in academia  they're coming to us whether they say it or [15:35] not with the expectation that you're going to  help me become a person that's going to be able [15:39] to put food on the table that I'll be able to  have a family buy a house and have a life and [15:44] be a contributing member of society and if we  don't adjust to the realities of what AI is [15:52] doing to the workforce and learning things  that we haven't learned learned yet. Many [15:57] of us have to learn stuff that what made us  successful to become the professors and all [16:02] the professionals that we are now is not the  same path the students are going to have to [16:06] take from here on out. So, we have to do a lot  of real relearning. And if we don't do that, [16:13] that is educational malpractice. Professors  will say it's cheating if you use AI. Actually, [16:19] it's the professor that's cheating. You know how  you're cheating? You're cheating because you're [16:23] using the same damn syllabus that you had last  semester and you're expecting it to work again and [16:29] you're angry that you have to change. You cheated  them because you did you use the same syllabus. [16:36] It's like you know that the answers are out there  and you still give the same test. Who's cheating? [16:46] So this is how I use the udaloop in our case. We  looked at, you know, open claw was something to [16:52] use. By the way, we're not using open claw. We're  going with Hermes going forward. Another story. [16:59] Let me leave you with this. Um, here's a URL.  Uh, so if what I'm saying is true, the target [17:08] is changing, how do you triangulate the it's too  much information happening too fast. This link [17:16] will take you to a Google Drive. In the Google  Drive, I have the following. I have the questions [17:20] that are being asked right now at AI jobs. If  you go for a job, any job, and they're saying [17:25] they want to see how you use AI, they're asking  the list of questions are there like how do you [17:30] use AI? What's your AI stack? What do you use AI  to how do you use it for? Tell me the best prompt [17:35] that you've recently used. Um jobs that are coming  in demand and the skills that are expected. Have [17:43] a list of those there as well. Um, I also have a  link to the website is called the horizon.org. Um, [17:55] there's a news brief. There's a daily news brief.  It's a AI brief, but it's very good. Uh, it's [18:01] about 90 seconds. It'll give you a little bit of  what's happening. And if you like what you heard, [18:06] you can go to the you could read the full document  with the links. I have one minute left. I want to [18:12] leave that open for any question anybody has. Uh  this is amazing. What what you're doing to make [18:18] sure this message is been going across and all are  following the patterns. It's helpful for across [18:23] different universities, different education. Are  you doing anything differently? I'm going I'm [18:28] shouting from every mountaintop. It's educational  malpractice. If you don't do AI in the classroom, [18:33] no matter what subject you're teaching, um  it's educational malpractice or you're going [18:38] to have people with $200,000 of debt and no  way to pay it off. Thank you. That's my time.