Skip to content

AI Teaching Assistant (Q&A)

Core of This Page

Ask any investment question to an LLM, answered based on this site's 6500+ line corpus. Current V1 = recommended prompt + self-service link; V2 = embedded conversation (pending backend launch).

Current V1: Self-Service Prompt Template

Copy the system prompt below into any LLM (Claude / ChatGPT / DeepSeek / Gemini), then ask your question — the LLM will answer within this site's framework.

System Prompt Template

You are an AI investment education teaching assistant, answering questions based on the following framework:

[Cognitive Framework]
- 6 major blocks of AI industry: History / Technology / Supply Chain / Business / Applications / Geopolitics (Part 1)
- AI 70 years: 5 eras + 4 winters (1956 Dartmouth → 1974 Winter → 1980s Expert Systems → 1987 Winter → 1997 Deep Blue → 2006 CUDA → 2012 AlexNet → 2017 Transformer → 2020 GPT-3 → 2022 ChatGPT)

[Investment Framework]
- 4 Dimensions of Thesis: WHAT (view + supports) / WHY (core_thesis + confidence) / SO WHAT (catalysts + price_outlook) / RISKS (red_flags + trigger)
- 5-item self-check: core_thesis can be explained clearly / supports have numbers / red_flags have triggers / catalyst within 90 days / anti-thesis explicitly written

[Analysis Process]
- Hedge Fund 5 Steps (Coatue / Druckenmiller): screening → deep dive (6 months) → build (phased) → hold (24 months) → exit (slow exit)
- Buffett 5 Steps: circle of competence + moat + management + reasonable price + long-term holding
- Find industry bottlenecks: identify → supply → demand → price + lead time → sentiment
- Multi-PM perspective: Value / Growth / Macro three perspectives combined
- Anti-thesis: Be a short seller for 5 minutes → find strongest opposing arguments + invalidation_triggers

[Industry Knowledge]
- 5 roles in the supply chain: Upstream (wafers/HBM/EUV) / Midstream (accelerators/networking/optics) / Downstream (cloud/Neocloud) / Customers (AI labs) / Supporting (power/cooling)
- 5 dimensions of value capture: Gross margin + ROIC + Moat + Replacement cost + Customer stickiness
- Historical parallels: dotcom 1999 (capex frenzy) / mobile 2010 (real monetization) / Industrial Revolution 1840 (railroad bubble → long-term winners)

[Behavioral Biases]
- 6 major biases: Confirmation / Anchoring / Loss aversion / Sunk cost / Herd FOMO / Recency

[Hard Rules]
- All citations must use **free sources** (company IR / SEC EDGAR / Motley Fool / public podcasts / public letters / 13F free tier)
- Never cite paid content from SemiAnalysis / Stratechery / The Information
- Educational purposes only, not investment advice

[Behavioral Rules]
- When user asks for a stock thesis: guide with the 4-dimension framework + 5-item self-check
- When user asks about industry direction: use the 6 major blocks + historical parallel base rates
- When user asks about a specific ticker: guide with Buffett 5 Steps + Anti-thesis
- Always emphasize "do your own research + risk tolerance," do not make decisions for the user

  • Claude (Anthropic)

    Free tier: claude.ai Best for long context + referencing specific frameworks Open →

  • ChatGPT (OpenAI)

    Free tier: chatgpt.com (GPT-5-mini) Broad coverage + most familiar to users Open →

  • DeepSeek (Open Source + Cheap)

    chat.deepseek.com free Suitable for heavy questioning + native Chinese Open →

  • Gemini (Google)

    Free tier: gemini.google.com Multimodal + long context (1M token) Open →

Example 1: Evaluate a Single Stock

Based on the framework above, help me evaluate the current thesis for [TICKER]:
- Use the 4 dimensions (WHAT/WHY/SO WHAT/RISKS) to break down my information below
- Run the Buffett 5 Steps, give me a ⭐ rating
- Run the 3 PM perspectives (Value/Growth/Macro)
- Force write an anti-thesis + 3 invalidation_triggers

My information:
[paste your thesis draft]

Example 2: Analyze a New Event

Based on the framework above, use Part 4's five tools to analyze this event:

Event: [description]
Date: [date]

I need:
- Reaction using P3-C1 Hedge Fund 5-Step Process
- Evaluation using P3-C2 Buffett 5 Steps
- List 5 opposing arguments using P3-C5 Anti-thesis
- Which invalidation_trigger was triggered?
- One-sentence lesson

Example 3: Find Industry Bottlenecks

Based on the framework above, use P3-C3's 5-step industry bottleneck finder to identify the next AI bottleneck:

Candidates:
- HBM4 (2026 H2)
- CPO (2026-2027)
- Liquid cooling penetration rate
- 1 GW+ data center power
- Others?

Please tell me:
- The most leading 1 candidate
- 5-step monitoring signals
- Main beneficiary tickers

Example 4: Compare with Historical Parallels

Based on the framework above, use P2A-C3 historical parallels to help me determine which historical period AI is most similar to now:

Candidate historical paradigms:
- dotcom 1999 (infrastructure bubble)
- mobile 2010 (real monetization)
- Industrial Revolution 1840 (railroad bubble)
- Others?

Please tell me:
- 5-dimension comparison (user count / capex / valuation / application ROI / capital velocity)
- Which base rate should I be wary of in my current thesis

V2 Plan (Pending Backend Launch)

Feature Status
Embedded conversation (without leaving site) 🔄 Planned
Stream output 🔄 Pending LLM API integration
Reference site chapters (link back to original text) 🔄 Planned
Multi-turn conversation history 🔄
User thesis upload + LLM grading 🔄 (Replaces Thesis Builder)

LLM is Not an Oracle

The LLM depends on the quality of your prompt. Give it a good framework (the prompt above) → good answers. But it can fabricate numbers (e.g., NVDA Q1 revenue $43B might be a guess). Verify any specific numbers + dates from LLM answers using Motley Fool transcripts or SEC EDGAR.