奇创岛

From «Antifragile: Things That Gain from Disorder»

Build an Antifragile Decision AI Toolkit

You'll take the barbell strategy, convexity test, skin-in-the-game check, Lindy effect filter, and black-swan warning from *Antifragile* and transform them into a set of AI prompts you can call directly in ChatGPT or Claude. Each prompt includes a complete system-role definition, a structured input template, and an output format. You'll test every prompt against real decisions of your own, and the result is a ready-to-grab 'Antifragile Decision AI Toolkit.'

Final work

One *Antifragile Decision AI Toolkit* (containing 5 complete prompts)

Estimated time

1.5–2 hr

Submitted

…

Your final work

Purpose:Turn Taleb's decision frameworks into operational AI tools so that when you face uncertain decisions, trial-and-error evaluations, or extreme scenarios, you have a reusable AI interrogation system instead of starting from scratch every time.

Parts:

  • Barbell Decision Evaluator: identifies whether a given option sits in the fragile middle ground
  • Convexity Identification Prompt: checks whether a decision's upside/downside structure is worth the risk
  • Black-Swan Risk Alert: performs a structured scan of extreme-world terrain and tail risks for a given situation
  • Skin-in-the-Game Verification Prompt: quickly assesses the credibility and interest-alignment of a piece of advice
  • Lindy Effect Filter: judges whether a tool, idea, or method can withstand the test of time
  • Each prompt comes with: usage-scenario notes + input template + output-format requirements
  • Each prompt comes with: a real test case (validated against one of your own decisions)

Use cases:

  • · Run a structured self-interrogation before major investment, career, or startup decisions
  • · Quickly assess the credibility of advice you receive from others
  • · Identify tail risks and convexity opportunities when planning new projects or experiments
  • · Keep the toolkit as a long-term 'antifragile defense layer' in your personal decision system

Pick a topic

Pick the topic closest to you, or write a custom one when you submit.

Personal Life

Learning & Growth

Work & Projects

Communication & Relationships

Tools you'll use from the book

Three-State Judgment System Prompt

A system prompt that assigns the AI the role of 'antifragility auditor,' requiring it to classify any decision input into one of three structural states — fragile / robust / antifragile — rather than offering ordinary advice.

How to use it here:

Acts as the base role layer for all other prompts; can be used standalone or stacked before any other prompt. To use: fix the system section as the antifragility-auditor role, then enter the specific decision context in the user section.

Boundaries:

The AI's three-state judgment is an interrogation aid, not a final verdict. The result depends on the quality of the information you provide — distorted inputs produce incorrect classifications.

Barbell Combo Generator

A structured prompt that has the AI output a barbell-configuration recommendation for a given decision scenario — 'extreme-conservative end + extreme-aggressive end' — while explicitly marking where the 'middle ground' is and why it's fragile.

How to use it here:

Input: your current resource allocation (time / money / energy) + the option you're considering. AI output: 2–3 specific options for each end of the barbell + middle-ground identification + why your current allocation may be drifting toward the middle.

Boundaries:

The barbell generator doesn't make choices for you; its value is 'exposing the hidden dangers of the middle ground.' The two-end configurations in the AI output are directional suggestions — execution still requires fitting your own constraints.

Convexity Check Prompt

Forces the AI to separately analyze the upper and lower bounds of an option's 'loss structure' and 'gain structure': does the worst case have a floor? Does the best case have no ceiling?

How to use it here:

Input: one specific option (e.g., joining a paid community, attending a training, committing 3 months to a side project). AI output: downside analysis (worst case + whether losses are linear or non-linear) + upside analysis (best case + whether gains are scalable) + convex/concave conclusion + recommendation on whether to continue evaluating.

Boundaries:

The convexity check is based on the information you provide; AI cannot predict real black swans. Its value is revealing whether 'the structure you're currently considering' makes sense. Don't treat 'AI says it's convex' as authorization to act.

Black-Swan Trigger Alert Prompt

Has the AI systematically scan a situation for 'extreme events that could happen but are being underestimated,' then assess whether your current configuration would be harmed (fragile) or would benefit (antifragile) if each event occurred.

How to use it here:

Input: your current situation (a segment of your business / career / life structure) + 2–3 scenarios you think are unlikely. AI output: identifies 3–5 additional extreme events you hadn't considered + exposure analysis for each + where your current setup gains vs. suffers + the 1–2 tail risks most worth watching.

Boundaries:

By definition, black swans are hard to predict; the AI can only scan within the 'imaginable range' of extreme events — true black swans lie outside the scan results. The value of this tool is widening your 'imaginable extreme-event' range, not giving you the illusion that black swans can be predicted.

Skin-in-the-Game Verification Prompt

Has the AI quickly analyze the 'consequence-bearing level' of an advisor, judging whether the person giving the advice has a real stake in the outcome, in order to calibrate how much weight to give the advice.

How to use it here:

Input: the advice itself + the source (who gave it, their role and interests) + what the advisor would experience if you followed the advice. AI output: skin-in-the-game score (high / medium / low) + analysis of the advisor's potential motivations + how much weight you should give this advice + 2–3 ways to further verify its credibility.

Boundaries:

Skin-in-the-game verification does not mean 'advice from anyone with nothing to lose is worthless.' Taleb's principle is about *weighting*, not *blocking*. Advice given out of genuine care without direct financial stake — such as parental guidance — can still be valuable. The AI's analysis helps you calibrate the weight, not dismiss the advice.

Work rules

Your work MUST include

  • At least 5 complete prompts (each containing: usage scenario + system prompt text + user input template + expected output format)
  • Each prompt must have one real test case (enter a genuine decision scenario into the AI and record its output)
  • Each prompt must include a 'boundary statement' (what this tool cannot replace)
  • The barbell combo generator must explicitly identify the 'middle ground'
  • The skin-in-the-game check must analyze a real advisor — no fictional scenarios

Your work CANNOT just be

  • Don't just excerpt Taleb's concepts without converting them into usable AI prompts
  • Don't substitute a vague 'What do you think of this decision?' for a structured prompt
  • Don't skip real testing — an untested prompt set is not a toolkit
  • Don't treat AI output as the final decision authority; the toolkit must emphasize that AI is a supporting interrogation tool, not an answer machine
  • Don't design all 5 prompts in the 'tell me what to do' mode — design them in the 'help me find my blind spots' mode

AI can help you here

Round 1: Help me design my first antifragile prompt

When to use: You're not sure which tool to start with, or you want AI to help you convert a Taleb concept into a prompt framework.

I'm completing the '{{route name}}' project using *{{book title}}*, with the goal of turning the book's decision tools into AI prompts I can call in ChatGPT or Claude.

My situation:
[Describe the decision challenges you face most often, and the Taleb concept you most want to turn into a tool — e.g., barbell strategy / convexity check / skin in the game / Lindy effect / black-swan alert]

Please design a complete AI prompt for the concept I most want to operationalize, including:
1. Role definition (what specific role should the system prompt assign to the AI?)
2. User input template (what specific information should the user fill in? list it in a structured format)
3. Output format requirements (what structure should the AI use to respond, and what does each section cover?)
4. Boundary statement (what can't this prompt do? — to prevent misuse)

Extra requirement: the prompt design should embody 'help users find blind spots' rather than 'give users answers.'

Yellow placeholders need you to fill in before using the AI.

AI can help you organize ideas, but cannot make final judgments for you. Don't let AI fabricate experiences, cases, or misleading content.

Round 2: Help me refine an existing prompt

When to use: You've drafted a prompt and want AI to check whether it truly reflects the book's concept, and identify prompt-engineering improvements.

My project is the '{{route name}}' route in *{{book title}}*.

My topic is:
{{topic}}

I've written the following prompt draft:
[Paste your draft prompt, including the system section and the user input template]

Please review and improve this prompt from two angles:

Angle 1 — Conceptual accuracy
- Does this prompt truly reflect the core logic of '{{topic}}' from *Antifragile*?
- Where is the concept misunderstood and in need of correction?
- Are there critical interrogation dimensions of this tool that are missing?

Angle 2 — Prompt engineering quality
- Is the system prompt's role definition specific and constraining enough?
- Does the user input template make it clear what to fill in? Vague fields lead to generic AI output.
- Is the output format sufficiently structured to prevent the AI from producing empty encouragement?

Please output:
1. List of conceptual-accuracy issues (if any)
2. Prompt-engineering improvement suggestions (point by point)
3. A complete revised version of the prompt

Yellow placeholders need you to fill in before using the AI.

AI can help you organize ideas, but cannot make final judgments for you. Don't let AI fabricate experiences, cases, or misleading content.

Round 3: Help me review the full toolkit draft

When to use: You've finished all 5 prompts and their test records and are ready to submit; you want AI to do a comprehensive review.

I'm submitting my work for a Shufang Island project.

Book: *{{book title}}*
Route: {{route name}}
My topic: {{topic}}

My draft work:
{{draft work}}

Please conduct a comprehensive review of this 'Antifragile Decision AI Toolkit,' checking the following dimensions:

1. Completeness check
- Does the toolkit cover at least 5 distinct prompts with non-overlapping functions?
- Does each prompt include system + input template + output format + boundary statement?
- Are there real test records for at least 2 prompts?

2. Conceptual fidelity
- Does each prompt genuinely reflect the corresponding tool from *Antifragile*, or is it just generic decision Q&A?
- Does the toolkit's design embody Taleb's core stance that 'simple beats complex'?
- Has any prompt confused 'giving answers' with 'finding blind spots'?

3. Practical usability
- Can this toolkit actually be deployed, or is it too theoretical?
- Is the user input template specific enough for a user to know what to fill in?
- Do the test cases show that the prompts actually surface blind spots?

4. Differentiation check (vs. other routes for this book)
- Is this toolkit clearly distinct from routes like 'Barbell Decision Matrix' and 'Fragility Portrait'? The value of an AI toolkit lies in being reusable and conversational, not a static analysis framework.

Please output:
- Overall evaluation
- The 1–2 most valuable prompts (with reasons)
- Issues that must be fixed
- Areas that could be further strengthened
- Whether it is ready to submit

Yellow placeholders need you to fill in before using the AI.

AI can help you organize ideas, but cannot make final judgments for you. Don't let AI fabricate experiences, cases, or misleading content.

Submit my work