Materiales del curso · Academia de IA SiliconRoshi

Members join every weekly live session free, replays included · Become a member →

T1S1AI 来了，干就完了 — AI 的本质LLM 是建立在人类语言上的超级概率数学模型 — 它什么都知道，但你说的话决定了它给你什么 · LLM 和 Agent 是两个东西：LLM = 悟空（原始能力），Agent = 龟仙人写的秘籍 + 悟空 · 同一个模型，不同提示词，结果天差地别 — 提示词就是你控制悟空的能力 · 你的独门绝技（行业经验、客户理解）是护城河，AI 能力越来越便宜，你的世界才值钱 · 六颗龙珠路线：勇气 → 得力 → 自由 → 眼光 → 精炼 → 出师★ 勇气

🎬▶ YouTube ▶ Bilibili

T1S2召唤超级变身英雄出场变身咒语四层心法：Role（角色）→ Rules（规则）→ Constraints（约束）→ Examples（示例） — 一层都不能缺 · 写 Role 不是写简历，是注入哲学 — 用三句话定义此人看世界的方式，让 AI "先思再答" · 用「现象 → 本质 → 哲学」三层穿梭把英雄写成可运行的 Role，再用 XML Tags 把四层组起来 · 召唤宗师：Paul Graham、Peter Thiel、Linus、或你领域里那个真正的巨人 — 他们不在对面，而在你身后、在你身体里 · 今天的第二颗龙珠：得力·召巨人 — 同一个你，用对了 prompt，AI 给你的是你自己想不到的东西★★ 得力

🎬▶ YouTube ▶ Bilibili

T1S3悟空悟了全世界，但只有空的记忆中心论点：LLM 是 stateless 一指禅；Agent 是剧本打印机；做 agent = 写剧本 · Token 经济学：input 便宜、output 贵 4×；context window = 纸槽；不当编剧就是别人剧本里的角色 · 小悟空 = 悟空 (LLM) + 你 (专有 10 年的活)；赢 ChatGPT 靠三轴：① 窄 ② 深 ③ 真 · 幻觉不是大模型的问题，是剧本的问题；铁律必须写明，否则答对了也会被推翻★★★ 自由

🎬▶ YouTube ▶ Bilibili

T1S3.5AI 时代极速前进 · 成为 AI 时代的悟空两跌一悟：自己都不用的东西不要做 · 不能放大自己能力的事少做 · 卡在点上抓眼前的大好机会 · 三层穿梭三个 So：So why?（穿下去）→ So what?（回头验证）→ So now?（穿回现象）— 不是问三次，是验三次 · 红海蓝海 · 找痛点先确认你站的是哪一边：你能看见、你能解 = 蓝海；拼价格拼速度 = 红海 · 起手式 · 先扫描自己：你 10+ 年最懂的、谁经常找你、你帮哪种人不累 — 各写 3 个，小悟空就长在上面 · 老师领进门，修行靠 AI · 客户从你教会的人里来 — 教育不是营销，教育是信任的兑现凭证☆ 觉醒Paid

🎬▶ Full ▶ YouTube ▶ Bilibili

T1S4悟空破石而出中心论点：模型我选 + 数据我控 — 两条宪法 · 今晚兑现 · 写在悟空道场的第一条规矩里 · 真假 agent · 分水岭：智能在哪一步介入 — 写死的 if/else 是假 agent，悟空在每个节点决策的才是真 agent · 代理循环通用心法：接收 → 回忆 → 推理 → 动作 → 记下 — 任何 agent 都在跑这一圈，你的小悟空跟 ChatGPT 同一个 tier · 12 个 model 白名单：从 $0.03 地板到 $5 旗舰 1000× 价差 — 你按 task 自己选，默认地板，升级一键 · 现场 Lab：自己在悟空道场上建两个不同剧本的 agent — 同一题问两遍，看剧本差异怎样改变输出 · 成为悟空 · 你可以 — 大悟空醒了 · 小悟空生了 · 就在今天★★★★ 眼光

🎬▶ Full ▶ YouTube ▶ Bilibili

T1S5千锤百炼 · 养好你的小悟空中心论点：剧本越丰满 · agent 越按你要的来 —— S5 的丰满不在嘴上，在你 KB 里真喂出来的那一份 .md · 88% AI agents 死在 production —— 三堵墙：prompt engineering 不够 / context engineering 还不够 / harness engineering 必须有人看 · 四个都要动：元悟空（LLM）+ 平台（道场）+ 剧本（你写的）+ 你（human in the loop） —— 缺一养不出来 · 分寸有四档：Open 70/30（找灵感）· Blend 40/60（日常）· Strict 10/90（合规医疗）· Cite 0/100（法律 KB）—— 该自由时自由 · 该收住时收住 · 范例三例：Workflow（流程闭环）· Constraint（强约束 4 闸门）· Knowledge（域内独家 + 域外干净拒绝） —— 三种长法都是 95/95 的作业 · 念好紧箍咒 · 养好小悟空 —— 这一周打磨一刀，S6 我们接着往上架走★★★★★ 精炼

🎬▶ Full ▶ YouTube ▶ Bilibili

T1S6出师 · 让你的 agent 跑给真人看张雪峰死后被 AI "复活" —— 与其等人偷你 · 不如主动 own 自己 · 你自己 distill 自己 · 主动 publish · 悟空平台支持你 · 回望 S1→S5 · 1,203 份作业 · 316 个 S3 agent · 跨 12 个真实领域 · 有人省了千多美金 · 有人被起诉后用 agent 自救 · 每个台阶都是真做到了 · 8 件工具全上线 · 一键 builder / 16 模型 / 对话存档 / 分享 3% / Publish / WukongGPT / 用量看板 / 计费钱包 —— 都在 v0 · 跟你一起迭代 · AI 学习社区四类悟空 · 同行 / AI / 赋能 / 领域 —— 最重要的是互相试用彼此的 agent · 让 community 跑起来 · S6 作业 · 上架 · 分享 · 让 1 个真人用 · 采访他 · 写一份小报告 —— agent 不是文档 · 是要有人用的产品★★★★★★ 出师

🎬▶ YouTube ▶ Bilibili

T1ENS1The Awakening — AI EssentialsAn LLM and an Agent are different things: LLM = Wukong unchained, Agent = Wukong with a master — you want the trained one · Your instructions are everything — same AI, different instructions, completely different results. That is the skill this course trains · Your world is your moat: AI knowledge is free and getting cheaper; YOUR experience is what makes your agent unique · We are in the earliest days — like the internet in 1995. The people who learn now will lead later · Lab: your first agent went live on wukongdojo.ai — a 3-line prompt, your scene, one real question · Where this goes: after 6 weeks your own agent runs in Discord / WeChat / Telegram — the messaging apps you already use★ Courage

🎬▶ Part 1 ▶ Part 2 ▶ Part 3 ▶ Part 4 ▶ Part 5 ▶ Part 6 ▶ Part 7 ▶ Lab ▶ Q&A ▶ Lucy Agent Demo

T1ENS2Summon Your Hero — Prompt EngineeringA prompt is not a request — a prompt is a summoning. Every hero exists as a pattern in AI's memory; your job is to summon the right one · The Four-Layer Incantation: Role · Rules · Constraints · Examples — miss any layer, get a generic answer · Three-Layer Dive to write a hero's soul: Phenomenon (what they say) → Essence (why they say it) → Philosophy (what they believe about the world) · Three sages on one table: Paul Graham (craft) · Peter Thiel (strategy) · Marc Andreessen (PMF) — same case, three lenses, 360° coverage · The Linus prompt: an industry-grade production prompt Bill runs daily — three iron laws (Good Taste · Pragmatism · Simplicity) · The Business Model: $47B customer-support AI market · four pricing models in the wild · lawyer AI vertical as the zoom-in · Homework · Golden Hoop 2 · Leverage: summon your master in a domain you care about, extract 3 real pain points, sketch one S4 agent★★ Leverage

🎬▶ Part 1 ▶ Part 2 ▶ Part 3 ▶ Q&A

T1ENS3Ground Your Wukong — Model + KnowledgeWhat happens when you hit Send: YOU → the Agent → Wukong (the LLM — a brain in a jar: brilliant, no hands, no memory) · The Agent gathers three things every turn — persona card + chat history + your new message — into ONE stack: the context. "Memory" lives in the stack, not in Wukong · Token economics: every sheet in the stack is re-sent, re-read, re-billed every turn; the context window is the paper tray · The model menu: Claude / GPT / Gemini in three sizes each, and the open wave (DeepSeek #1 by token share) up to 20× cheaper — WukongDojo lets you pick any good LLM, your Wukong stays yours · You + Wukong = Your Wukong: your domain + your script beats generic ChatGPT on three axes — narrow · deep · real · Lab · Ground Your Wukong: attach one knowledge .md and watch the same model stop guessing★★★ Freedom

🎬▶ Part 1 ▶ Part 2 ▶ Part 3 ▶ Lab and Demo ▶ Q&A

T1ENS4Wukong Breaks Out of the StoneCentral thesis: Model is Mine + Data is Mine — two constitutional laws delivered tonight, written into the first rule of WukongDojo · Real vs fake agent watershed: where does intelligence enter — hard-coded if/else is a fake agent, only when Wukong decides at each node is it a real agent · Universal agent loop: Receive → Recall → Reason → Act → Remember — every agent runs this same circle, your Little Wukong is the same tier as ChatGPT · 12-model whitelist: from $0.03 floor to $5 frontier, 1000× spread — you pick by task, default to floor, upgrade in one click · Live Lab: build two agents on WukongDojo with different scripts — same question, see how script design changes the output · Become Wukong · You Can — Big Wukong awakens · Little Wukong is born · today★★★★ Vision

T1ENS5Forge — Raise Your Wukong WellCentral thesis: a script is not paperwork — a well-forged one is both the most powerful and most beautiful thing you write (Crouzeix conjecture · ~20 lines · 22 years · solved) · 88% of AI agents die on the way to production — three killers: 85%×10 steps = 20% / nobody governs context (Context Debt) / it won't say no · Three walls → three paradigms: prompt engineering (not enough) · context engineering (still not enough) · harness engineering (you own the runtime constraint layer) · Four actors must all move: LLM (grows on its own) + Platform (memory · harness · Studio) + Script (yours, dense) + You (human in the loop) — miss one and it never grows · One loop rules them all: workflow / constraint / knowledge all fold into a single Agent Loop — plan → act → observe → repeat · Forge one cut this week · raise your Wukong well — S6 is Demo Day · we walk it toward launch★★★★★ Refinement

🎬▶ Opening ▶ Part 2 ▶ Lab ▶ Q&A

T1ENS6Demo Day — Open Your DojoYou made it — graduation night · four real graduates demo live · six weeks ago they were you · the only gap between audience and stage is six weeks and a room · Tonight's four builders: Jeffrey (Five-Elements Relationship Coach) · Jinxia (Holistic Financial Planning) · Minzi (CoolArch Designer) · Grace (Trading Platform Selector) — real students, real agents · The student capability ladder S1→S5 — every step a real student who did it: see what AI is · summon the hero · write your own script · the three-layer shuttle · ship a working agent · The agent market is a blue ocean — value isn't handed to you, you find it in your own world · $400 per booked date proves it · polish one agent that truly delivers and it monetizes · Academy meets platform: the Academy teaches you to build · the platform carries and monetizes it · the members ship real projects together — this graduation is one proof, yours is next · S6 homework · publish · share · let 1 real person use it · interview them · write a short report — an agent is not a document, it is a product meant to be used. AI is here. Just build it.★★★★★★ Mastery

🎬▶ Demo 1 ▶ Demo 2 ▶ Demo 3 ▶ Demo 4 ▶ Closing

Iniciar sesión

Correo enviado