{
 "about": "VC / 投资人和教育研究机构自己写的 AI K12 教育文章（投资备忘、赛道论点、市场地图、年度预测），每篇打开原文核对过标题、日期、作者。gist 是核心观点的中文转述，不是原文；seg 对应 layers.json 的 code，all = 赛道总论。",
 "articles": [
  {
   "t": "Reinventing School for the AI Era | Joe Liemandt, Gagan Biyani & Erik Torenberg",
   "org": "a16z",
   "au": "Erik Torenberg, Gagan Biyani（嘉宾 Joe Liemandt）",
   "date": "2026-10-27",
   "url": "https://a16z.simplecast.com/episodes/reinventing-school-for-the-ai-era-joe-liemandt-gagan-biyani-erik-torenberg-qpcIrpfu",
   "kind": "赛道论点",
   "seg": [
    "m10"
   ],
   "co": [
    "Alpha School"
   ],
   "gist": "Alpha School 创始人 Liemandt 的观点是，AI 若让学生学得快得多，多出的时间不该塞更多学业，而该围绕主动性、团队、创业、沟通这些传统课堂难教的能力重做学校：学生每天两小时学核心课，其余做项目。a16z 把这种思路延伸到他们办的大学年龄项目 Horowitz and Andreessen Academy。",
   "cav": "节目简介页，只读到简介，未读全文；Liemandt 是被访的创办者，a16z 自己没有提出批评，与 Contrary 的质疑形成对照。",
   "star": true
  },
  {
   "t": "Building AI for human learning: why we invested in Aristotle",
   "org": "True Ventures",
   "au": "Mike Montano",
   "date": "2026-09-16",
   "url": "https://www.trueventures.com/writings/why-we-invested-in-aristotle",
   "kind": "投资备忘",
   "seg": [
    "s5"
   ],
   "co": [],
   "gist": "True 领投 Aristotle 种子轮，核心判断是一对一辅导有效但从没覆盖到多数学生，AI 有机会补上。他们批评通用大模型是糟糕的老师，因为直接给答案、抹掉了思考挣扎这一学习发生的地方；Aristotle 的语音家教让学生边想边说、基于数千小时带评分的多轮辅导对话训练，目前面向 13 至 18 岁。",
   "cav": "文中未披露金额（另有新闻稿称 500 万美元）；Aristotle 不在图谱公司名单里，故 co 留空。"
  },
  {
   "t": "Our August update: imagi announces $4.5M raise",
   "org": "Brighteye Ventures",
   "au": "Team Brighteye",
   "date": "2026-08-18",
   "url": "https://brighteye.substack.com/p/our-august-update-imagi-announces",
   "kind": "投资备忘",
   "seg": [
    "t3",
    "g11"
   ],
   "co": [
    "imagi"
   ],
   "gist": "Brighteye 与 Day One Capital 领投 imagi 450 万美元种子轮。它把 imagi 定位为「AI 工具和学校之间的安全教育层」，合伙人 Benoit Wirz 说 imagi 站在两个趋势交汇处：AI 扩散，以及对有结构、安全的数字素养的需求；文中称已有 70 万学生、140 国使用。",
   "cav": "月度通讯里的投资快讯，论证简短；使用量为公司口径。"
  },
  {
   "t": "EdTech Isn't One Industry. It's Nine.",
   "org": "Edtech Insiders",
   "au": "Ian McCullough",
   "date": "2026-08-13",
   "url": "https://edtechinsiders.substack.com/p/edtech-isnt-one-industry-its-nine",
   "kind": "赛道论点",
   "seg": [
    "all"
   ],
   "co": [],
   "gist": "作者认为 edtech 的老经济模型已不成立，AI 还在侵蚀收入，行业说不清自己卖什么；提出「客户类型（B2C/B2B/B2G）×方案类型（教学/基础设施/托管服务）」九宫格，主张把九个市场分开看、按各自逻辑投资。文中引 HolonIQ 称 2025 年 edtech 风投比 2018 年峰值低 68%。",
   "cav": "作者是 edtech 营销从业者，不是投资人，发表在 Edtech Insiders；全文几乎不谈 K12 AI，只有侧栏提到相关项目。"
  },
  {
   "t": "What Edtech Got Right, What It Missed, and What AI Changes Now (Jennifer Carolan, Reach Capital)",
   "org": "Reach Capital",
   "au": "Allison Dulin Salisbury（采访 Jennifer Carolan）",
   "date": "2026-06-22",
   "url": "https://www.thehumanist.xyz/p/what-edtech-got-right-what-it-missed",
   "kind": "赛道论点",
   "seg": [
    "all",
    "s5"
   ],
   "co": [],
   "gist": "Carolan 说 AI 家教进步比她预想的快，可能成为正规教育变化的主要推手（文中说斯坦福学生小组几周就做出能用的家教）；但她仍认为老师是核心，托举老师而非替代老师的产品才会带来结果。她认为过去的 edtech 漏掉了学习的社群与人情一面，也低估了家庭对个性化的需求；并说 edtech 只拿到全部风险投资的 1.5%。",
   "cav": "是第三方通讯 The Humanist 的访谈，不是 Reach 官网文章；数字为受访者所说。",
   "star": true
  },
  {
   "t": "AI-Native K-12 EdTech Market Map (Spring 2026)",
   "org": "Stepmark Partners",
   "au": "Tejasveer Chugh",
   "date": "2026-05-29",
   "url": "https://stepmark.ai/2026/05/29/ai-native-k-12-edtech-market-map-spring-2026/",
   "kind": "市场地图",
   "seg": [
    "all"
   ],
   "co": [],
   "gist": "这家专做 AI 的投行认为 K12 edtech 依旧很有分量，AI 正在加速向个性化支持和减轻行政负担的转移。新版地图比上一版多收 11 家公司，按「老师主导、学生主导、全班级流程」三类整理，并提到采用正延伸到家庭，文中引用 Common Sense Media 称七成青少年用过生成式 AI。",
   "cav": "Stepmark 是投资银行不是 VC；文中没有指出哪些赛道拥挤，也无各类公司数量。"
  },
  {
   "t": "Why We Invested in Subject",
   "org": "Vistara Growth",
   "au": "Jordon Boerchers",
   "date": "2026-04-30",
   "url": "https://vistaragrowth.com/case-studies/why-we-invested-in-subject/",
   "kind": "投资备忘",
   "seg": [
    "i1",
    "m10"
   ],
   "co": [
    "Subject"
   ],
   "gist": "Vistara 以灵活的成长资本方式向 AI 原生 K12 平台 Subject 投了 2800 万美元。看好点是可衡量结果：文中称 24 个州 14.3 万个在读注册、课程完成率从 2022 年 68% 升到 2025 年 85%、补学分通过率两年从 55% 升到 73%；美国 K12 年支出约 1 万亿美元，Subject 才进入约 2 万个学区中的几百个，有机会从工具预算走向更大的预算池。",
   "cav": "页面只写「30 April」未标年份，年份依上下文与搜索结果推断为 2026；成效数据为公司与投资方口径。"
  },
  {
   "t": "Ian Chiu (Owl Ventures) LinkedIn post: keeping AI human-centered in education",
   "org": "Owl Ventures",
   "au": "Ian Chiu",
   "date": "2026-03",
   "url": "https://www.linkedin.com/posts/ianchiu_aiineducation-edtech-futureoflearning-activity-7435415772450979840-g3Rk",
   "kind": "赛道论点",
   "seg": [
    "all",
    "a2",
    "s6",
    "e13"
   ],
   "co": [
    "Newsela",
    "Amira Learning"
   ],
   "gist": "Owl 合伙人 Ian Chiu 说 AI 在教育里才刚开始，关键是让 AI 以人为中心：放大老师的差异化教学、反馈和洞察，把人牢牢放在驾驶位。他举了 Owl 被投公司的例子：Kiddom 用 AI 从日常作业里找出错误观念、Newsela 用 AI 做分级阅读、Amira 提供一对一口语朗读练习。",
   "cav": "LinkedIn 短帖，页面只显示相对时间，日期取自搜索结果；标题是按内容拟的（原帖无标题）；Kiddom 不在图谱名单。"
  },
  {
   "t": "The Future of Education?",
   "org": "Contrary Research",
   "au": "Claire Schultz",
   "date": "2026-02-21",
   "url": "https://research.contrary.com/report/the-future-of-education",
   "kind": "赛道论点",
   "seg": [
    "m10"
   ],
   "co": [
    "Alpha School"
   ],
   "gist": "这篇对 Alpha School 持怀疑态度：每年 4 万到 7.5 万美元、学生每天两小时跟 AI 家教学习，学校自称成绩在全国前 1%；但文中列出员工称 AI 生成过无法回答的题目、涉嫌抓取别人的教材，还有用零食和现金奖励逼达标、默认录制学生在家学习的网络摄像头画面。作者问这究竟是未来教育，还是为了考试分数牺牲更广的教育目标。",
   "cav": "Contrary 是风投公司 Contrary 旗下研究部门，不是直接的基金备忘；成绩数据只来自学校自己，批评来自员工和报道。",
   "star": true
  },
  {
   "t": "Out Now: The Updated Edtech Insiders K-12 Generative AI Market Maps",
   "org": "Edtech Insiders",
   "au": "Sarah, Ben Kornell, Alex Sarlin, Jen Lapaz",
   "date": "2026-02-18",
   "url": "https://edtechinsiders.substack.com/p/out-now-the-updated-edtech-insiders",
   "kind": "市场地图",
   "seg": [
    "all",
    "b12",
    "e13"
   ],
   "co": [],
   "gist": "这一版把 K12 生成式 AI 地图拆成两张：AI 原生创业公司一张，老牌 edtech、大科技公司和教育服务商一张。理由是把初创和巨头放一张图上会夸大巨头的分量、盖住真正的创新；他们发现老牌玩家最常从学生支持和教师实践支持切入，最敢试验的仍是 AI 原生创业公司。",
   "cav": "Edtech Insiders 是行业媒体与社区，合伙人 Ben Kornell 也是投资人，不是基金；页面未给出公司数量。",
   "star": true
  },
  {
   "t": "The Next Managerial Revolution and 2026 Edtech Predictions",
   "org": "Brighteye Ventures",
   "au": "Brighteye Ventures",
   "date": "2026-01-20",
   "url": "https://brighteyeweekly.substack.com/p/the-next-managerial-revolution-and",
   "kind": "年度预测",
   "seg": [
    "all",
    "a2",
    "s5"
   ],
   "co": [],
   "gist": "Brighteye 通讯汇总了 Edtech Insiders 里 20 位业内人士的 2026 预测：2025 年 AI 从试点走向真实使用；论文、考试作为学习证据在贬值，学校会转向口试、作品集和表现评价；AI 家教可能触达大得多的学生群体，但前提是可信、可靠并且与老师目标一致；人的辅导和关系依旧重要。",
   "cav": "是对他人预测的转述，没有逐人署名；摘录中没有 K12 专属内容。"
  },
  {
   "t": "AI, an enabler of universal access, increased equity, and global influence for India",
   "org": "Khosla Ventures",
   "au": "Vinod Khosla",
   "date": "2026-01-20",
   "url": "https://www.khoslaventures.com/posts/ai-an-enabler-of-universal-access-increased-equity-and-global-influence-for-india",
   "kind": "赛道论点",
   "seg": [
    "s5"
   ],
   "co": [],
   "gist": "Khosla 认为 2026 年每个印度人就能用上各地方语言的基础 AI 家教，可用性和表现都胜过大多数家庭请不起的真人家教，从幼儿园到博士都接近零成本、按兴趣个性化。他说建教育基础设施和培训老师要几十年，AI 家教是改善学习成果更快的路。",
   "cav": "讲的是印度，不专门针对 K12，是 Khosla 一贯的乐观判断，没有成效证据。"
  },
  {
   "t": "Ben Kornell and Tom Vander Ark on Predictions for Education in 2026",
   "org": "Edtech Insiders",
   "au": "Ben Kornell",
   "date": "2025-12-19",
   "url": "https://www.gettingsmart.com/podcast/ben-kornell-on/",
   "kind": "年度预测",
   "seg": [
    "all",
    "g11"
   ],
   "co": [],
   "gist": "Kornell 预测 2026 年是 K12 的「反弹之年」：受政治压力和对屏幕时间的担忧推动，至少 5 个美国大学区将禁止学生直接使用 AI。他同时认为教师培训和上手培训是 edtech 一块很大的预算机会。",
   "cav": "节目由 Getting Smart 发布，是访谈而非 Edtech Insiders 自有页面；预测为个人判断。"
  },
  {
   "t": "2026 Global Education Outlook",
   "org": "HolonIQ",
   "au": "HolonIQ Education Intelligence Unit",
   "date": "2025-12-19",
   "url": "https://www.holoniq.com/notes/2026-global-education-outlook",
   "kind": "年度预测",
   "seg": [
    "all"
   ],
   "co": [],
   "gist": "HolonIQ 称 2025 年全球 edtech 风投约 24 亿美元，增长主要来自中小额交易；投资人偏爱「AI 赋能、嵌入工作流、对接就业」的模式，对基本面和价值证明的审查更严。K12 和高教的投资集中在数字课程、学生成功和 AI 辅助学习。",
   "cav": "页面未给出 K12 单独的融资数字（图表为图片）；这是公开摘要页，完整 160 多页报告未读。"
  },
  {
   "t": "Reimagining School's Purpose in the Age of AI",
   "org": "Reach Capital",
   "au": "Jennifer Carolan",
   "date": "2025-09-18",
   "url": "https://thewire.reachcapital.com/p/writing-with-soul-schools-with-purpose-medicine-with-sense",
   "kind": "赛道论点",
   "seg": [
    "all"
   ],
   "co": [],
   "gist": "学生本来就不投入、作业脱离现实，AI 只是把这种脱节放大；与其不停写 AI 使用规定，不如借机重想学校为了什么。她认为学校是培养公民价值、学会跟不同的人合作的社区，在两极分化、孤立的社会里更重要。",
   "cav": "文章收在 Reach 月刊 The Wire 的一期里，页面无明确日期，日期取自文章链接。"
  },
  {
   "t": "GenAI and K-12: Hype, Headwinds, and Durable Investment Opportunities",
   "org": "Tyton Partners",
   "au": "Adam Newman, Mats Penberthy",
   "date": "2025-09-18",
   "url": "https://tytonpartners.com/genai-and-k-12-hype-headwinds-and-durable-investment-opportunities/",
   "kind": "赛道论点",
   "seg": [
    "all",
    "t3",
    "s7"
   ],
   "co": [],
   "gist": "Tyton 认为 K12 的生成式 AI 接近炒作顶峰，「带 AI」已成卖给学校的入场券；价值最扎实的是行政和老师工作流，学生端工具效果参差、需要成人参与。逆风包括通用平台抢走单点工具、老牌厂商抄新产品、ESSER 之后预算收紧；长期机会在帮老师、解决学区真实问题、培养沟通与毅力这类人的能力。文中称早期研究显示老师每周最多省 6 小时。",
   "cav": "Tyton 是教育战略咨询与投行，不是基金；数字为文中所引。",
   "star": true
  },
  {
   "t": "A Personal Tutor for Everybody: Our investment in Arivihan",
   "org": "Accel",
   "au": "Anand Daniel, Anagh Prasad",
   "date": "2025-07-31",
   "url": "https://www.accel.com/news/a-personal-tutor-for-everybody-our-investment-in-arivihan",
   "kind": "投资备忘",
   "seg": [
    "s5",
    "x9"
   ],
   "co": [],
   "gist": "Accel 用 Bloom 的「两个标准差」研究讲故事：一对一辅导最有效但无法规模化，生成式 AI 也许第一次让它可行。印度有 2.5 亿以上在校生、师生比低于 1:200，现有 edtech 多是静态内容，满足不了个性化备考需求；Accel 与 Prosus Ventures 共同领投 Arivihan 450 万美元，它是中央邦高中生下载最多的教育 App。",
   "cav": "印度高中备考，文中没出现「K-12」一词；Arivihan 不在图谱名单，co 留空。"
  },
  {
   "t": "The State of AI & Education",
   "org": "a16z",
   "au": "Justine Moore, Olivia Moore, Zach Cohen",
   "date": "2025-06-20",
   "url": "https://podscripts.co/podcasts/the-a16z-show/the-state-of-ai-education",
   "kind": "赛道论点",
   "seg": [
    "all",
    "t3",
    "s5",
    "m10"
   ],
   "co": [
    "Alpha School"
   ],
   "gist": "a16z 消费者团队认为现在推动 AI 进学校的主要是老师，因为他们愿意为日常工作流付费，用来减轻批改和备课；学生则多是用聊天机器人写作业。他们觉得 AI 在 K12 仍处边缘，AI 老师还很远，更可能让老师更高效而非取代；并提醒学习成效未经证实、缺基准评测、冲击教材出版商、加剧不平等，也拿不准会出现一家大公司还是一批应用层公司。",
   "cav": "播客稿，来自转录站，原节目在 a16z 官网；几位合伙人属 a16z 消费者团队，不是教育专职。"
  },
  {
   "t": "Announcing Our Investment in SchoolAI: Personalizing Education for Every Student and Teacher",
   "org": "NextView Ventures",
   "au": "Stephanie Palmeri",
   "date": "2025-04-02",
   "url": "https://nextview.vc/blog/announcing-our-investment-in-schoolai-personalizing-education-for-every-student-and-teacher/",
   "kind": "投资备忘",
   "seg": [
    "t3",
    "s5"
   ],
   "co": [
    "SchoolAI"
   ],
   "gist": "NextView 在 2023 年秋领投种子轮，这次继续跟投由 Insight Partners 领投的 2500 万美元 A 轮。看好的理由：创始人 Caleb Hicks 是老师出身，产品从教师真实痛点出发；老师负担重、疫情后学习缺口大，是 AI 进课堂的好时机；并强调符合 FERPA、COPPA、SOC 2。她也承认 K12 edtech 向来难做，对课堂新技术的门槛必须极高。",
   "cav": "自家被投公司的宣传性文章，使用量（超 100 万个课堂、400+ 学区）为公司口径，无独立成效数据。"
  },
  {
   "t": "AI and the Role of the Teacher",
   "org": "Reach Capital",
   "au": "Jennifer Carolan",
   "date": "2025-03-28",
   "url": "https://www.reachcapital.com/resources/thought-leadership/ai-and-the-role-of-the-teacher/",
   "kind": "赛道论点",
   "seg": [
    "t3"
   ],
   "co": [],
   "gist": "Reach 创始合伙人 Carolan 说 AI 是她见过进校园最快的技术，文中称约一半美国老师已在用（主要批改、家校沟通、备课），两年半触达 200 万老师；但师生之间的人际连接仍是最强的学习杠杆，AI 应该帮老师把每周约 53 小时的工作压到 40 小时左右，把时间还给激励学生这类 AI 做不好的事。",
   "cav": "改编自演讲稿；数字为文中所述。"
  },
  {
   "t": "Giving teachers their time back: Our investment in Brisk Teaching",
   "org": "Bessemer",
   "au": "Kent Bennett, Maha Malik, Bhavik Nagda",
   "date": "2025-03-25",
   "url": "https://www.bvp.com/news/giving-teachers-their-time-back-our-investment-in-brisk-teaching",
   "kind": "投资备忘",
   "seg": [
    "t3",
    "a2"
   ],
   "co": [
    "Brisk Teaching"
   ],
   "gist": "Bessemer 领投 Brisk 1500 万美元 A 轮。它的判断是过去的教育软件失败在增加复杂度，而 Brisk 以浏览器插件形式嵌进老师已经在用的 Google Docs、PDF、YouTube，能每周为老师省出最多约 10 小时，并有机会长成完整的课堂智能体和 AI 原生教育栈。文中称一年出头已有超 100 万老师、覆盖 100 国和 2000 多所学校/学区。",
   "cav": "数字来自 Bessemer 文中转述的公司数据。",
   "star": true
  },
  {
   "t": "Empowering Educators: Generative AI Trends For K12 Teachers",
   "org": "Edtech Insiders",
   "au": "Alex Sarlin",
   "date": "2025-03-07",
   "url": "https://edtechinsiders.substack.com/p/empowering-educators-generative-ai",
   "kind": "市场地图",
   "seg": [
    "i1",
    "t3",
    "p4"
   ],
   "co": [
    "MagicSchool",
    "Brisk Teaching"
   ],
   "gist": "文中称老师周工作时长中位数 54 小时，行政负担是倦怠第二大原因；MagicSchool、Brisk 这类老师工具集正从几项核心功能扩成全套。他们判断 AI 若能压缩批改、备课、行政时间，老师就能把精力放回师生关系，但前提是公平落地；并指出老师一年平均用 49 个工具，能打通学生数据的 AI 平台有机会。",
   "cav": "属 Edtech Insiders 2025 年 3 月地图更新系列的一篇；数字为文中所述。"
  },
  {
   "t": "Rethink Education Newsletter April 2024",
   "org": "Rethink Education",
   "au": "",
   "date": "2024-04-10",
   "url": "https://rethink-capital.com/rethink-education-newsletter-q1-2024/",
   "kind": "赛道论点",
   "seg": [
    "t3",
    "s5"
   ],
   "co": [],
   "gist": "Rethink 认为 AI 应该提升学校和老师的效率而不是取代他们，行业还要处理 AI 偏见、重新思考课程。合伙人 Ebony Brown 预计，能减轻老师行政工作的后台工具会先增长，并希望生成式 AI 最终带来 K12 的个性化学习。",
   "cav": "2024 年的季度通讯，没有个人署名；较旧，反映早期判断。"
  },
  {
   "t": "Five Predictions for the Future of Learning in the Age of AI",
   "org": "a16z",
   "au": "Anne Lee Skates",
   "date": "2023-02-08",
   "url": "https://a16z.com/the-future-of-learning-education-knowledge-in-the-age-of-ai/",
   "kind": "年度预测",
   "seg": [
    "all",
    "s5",
    "t3"
   ],
   "co": [],
   "gist": "a16z 认为 AI 会让一对一辅导变得普及（引 Bloom 的两个标准差研究）、学习内容个性化，并成为学生和老师最早采用的工具。老师可把写教案这类耗时事交给 AI；但她警告，公立与私立学校获得 AI 的机会不均，可能拉大差距。",
   "cav": "早期（2023）文章，ChatGPT 刚出现时的判断，较旧；仅作背景。"
  }
 ],
 "debates": [
  {
   "q": "Alpha School 这种「每天 2 小时 AI 学完主科」的学校能不能复制？",
   "a": "a16z 的节目请来 Alpha 创始人，基调是看好、要重做学校；Contrary Research 的报告则列出员工反映的题目出错、奖励机制、在家摄像头录制等问题，并指出成绩数据只来自学校自己。",
   "refs": [
    "https://a16z.simplecast.com/episodes/reinventing-school-for-the-ai-era-joe-liemandt-gagan-biyani-erik-torenberg-qpcIrpfu",
    "https://research.contrary.com/report/the-future-of-education"
   ]
  },
  {
   "q": "AI 家教到底能不能提升成绩？",
   "a": "Khosla、Accel 押注「人人有个私人家教」；Reach 的 Jennifer Carolan 说自己改了看法，AI 家教进步比预想快；a16z 2025 年的节目和 Tyton 则提醒效果还没被证实、要有大人陪着；True Ventures 认为通用大模型直接给答案，本身是糟糕的老师。",
   "refs": [
    "https://www.khoslaventures.com/posts/ai-an-enabler-of-universal-access-increased-equity-and-global-influence-for-india",
    "https://www.accel.com/news/a-personal-tutor-for-everybody-our-investment-in-arivihan",
    "https://www.thehumanist.xyz/p/what-edtech-got-right-what-it-missed",
    "https://podscripts.co/podcasts/the-a16z-show/the-state-of-ai-education",
    "https://tytonpartners.com/genai-and-k-12-hype-headwinds-and-durable-investment-opportunities/",
    "https://www.trueventures.com/writings/why-we-invested-in-aristotle"
   ]
  },
  {
   "q": "给老师的 AI 工具还值得投吗？",
   "a": "Bessemer（Brisk）和 NextView（SchoolAI）认为帮老师省时间是最快落地的切口；Tyton 认为 K12 生成式 AI 接近炒作顶峰，通用大模型免费、老牌厂商自带 AI、疫情补贴（ESSER）到期后学区预算收紧，都是逆风。",
   "refs": [
    "https://www.bvp.com/news/giving-teachers-their-time-back-our-investment-in-brisk-teaching",
    "https://nextview.vc/blog/announcing-our-investment-in-schoolai-personalizing-education-for-every-student-and-teacher/",
    "https://tytonpartners.com/genai-and-k-12-hype-headwinds-and-durable-investment-opportunities/"
   ]
  },
  {
   "q": "学生该不该直接用 AI？",
   "a": "多数文章默认学生会越来越多直接用 AI；Edtech Insiders 的 Ben Kornell 却预测 2026 年至少有几个大学区会禁止学生直接用 AI，只让老师用。",
   "refs": [
    "https://www.gettingsmart.com/podcast/ben-kornell-on/",
    "https://www.reachcapital.com/resources/thought-leadership/ai-and-the-role-of-the-teacher/"
   ]
  }
 ]
}