Welcome to Nofinity

Summary of the YouTube Transcript: The History of Financial Crashes & Why People NEVER Learn

This video analyzes the biggest financial bubbles in history and demonstrates how patterns repeat themselves.

• Tulip Mania (1630s, Netherlands): A luxury flower became a speculative asset. People traded contracts for bulbs still in the ground. Prices skyrocketed until buyers disappeared.
• South Sea & Mississippi Company (1720, England/France): Companies with political backing and exaggerated promises about overseas markets drove stock prices up until confidence collapsed.
• Railway Mania (1840s, Great Britain): A real, revolutionary technology. Thousands of miles were built, but many companies went bankrupt. The technology itself survived, investors lost money.
• 1929 Stock Market Crash (USA): Massive use of margin (leverage) fueled speculation. As prices fell, margin calls triggered a domino effect leading to the Great Depression.
• Japanese Asset Bubble (1980s): Rising real estate and stock prices allowed ever-increasing borrowing. The collapse led to decades of economic stagnation (Lost Decades).
• Dot-com Bubble (late 1990s): The internet was a real revolution. Companies with ".com" in their name were valued without profits. The 2000 crash destroyed many startups, but Amazon and the infrastructure survived.
• 2008 Housing Bubble (USA): Cheap loans to subprime borrowers, securitization of those risks, and high leverage led to the collapse of Lehman Brothers and a global financial crisis.

• Parallels: Meme coins (like tulips), projects with celebrity backing (like South Sea), blockchain infrastructure (like railways), crypto leverage products (like 1929), credit bubbles in DeFi (like 2008), AI hype (like Dotcom).
• Core Message: Every bubble started with a real opportunity, then was overtaken by speculation, leverage, and the belief that "this time is different."
• Warning Sign: When prices only rise because of the expectation that someone else will pay more (the greater fool), the bubble is ready to burst.

The speaker argues that financial bubbles are an inevitable part of markets. The key question is whether you recognize the warning signs while standing inside one.

• Quarterly results look fantastic at first glance: 24% revenue growth, cloud growth of 82%, 950 million Gemini users. Yet the stock is down 13% (20% from its peak).
• The big but: Exploding capital expenditures (Capex):
  • From an average of $30 billion per year to planned $205 billion in 2025 and even more from 2027.
  • Capex rises from 10% to 50% of revenue – a fundamental shift in the business model.
• First time negative free cash flow: -$6 billion in one quarter.
• Commitments skyrocket: In just one quarter, $470 billion were added, totaling $707 billion. This far exceeds the backlog.

• High depreciation eats into profits: With $250 billion in annual Capex, depreciation (D&A) surges. Even if revenue doubles to $900 billion in 5 years, profits could be close to zero due to depreciation.
• Return on invested capital questionable: The author fears that even if AI works, the industry will see low ROIC – similar to internet infrastructure providers in the past.
• Valuation: Even under optimistic assumptions (15% growth), Google offers no margin of safety according to the analysis. Intrinsic value is far below the current price.

• Risk-reward profile: High loss potential (50% possible) with low expected return. The analyst downgrades Google to a 'Bet' (risky wager).
• Warren Buffett's purchase: Seen as potentially a mistake, as Google no longer fits Berkshire's profile (negative cash flows).
• Outlook:

Summary follows...

• Bitcoin at ~$65K, up 11% in July.
• August could offer buying opportunities, but no moon shot yet.

• Hashrate heading for first annual drop; all-time high in Oct 2023, then steep decline.
• Miners pivoting to AI power usage rather than capitulation.
• Historically, such a drop marks the bottom of a bear market.

• Long-idle coins reach lowest level since 2018.
• Indicates holders are not selling – a positive sign.

• BitMart (13M users) and BitMEX (2M) shutting down.
• Withdrawals halted at BitMart – warning about exchange risks.
• Historically often signals market bottom.

• Coinbase policy chief spoke with Senator Thune: vote possible next Monday.
• Not certain yet, but could be a catalyst for altcoins.

• G7 money supply exploding (Canada +370%, USA +279% since 2004).
• Bitcoin relative to global M2 cheapest ever.
• Fiat going to zero – Bitcoin is hardest money.

• Selling own stock to raise cash for STRC dividends.
• Also buying STRC on market – circular action.
• Stock up +7.7% today to ~$100 – but not sustainable.

• Historical patterns suggest bottom in ~50 days (from hashrate data).
• Last optimists leaving the room – time to buy?

This video shows how to turn Claude into a full-time market analyst using the FMP connector (free tier included). The system replaces expensive Bloomberg terminals for retail investors and enables daily AI-driven portfolio analysis.

🔧 System Setup

1. Activate FMP connector: Go to Connectors > Add > search for FMP (Financial Modeling Prep) – free for 250 requests/day, paid for higher usage.
2. Create context folder: Build a folder with these markdown files:
   • Investor profile: Goals, strategy, risk tolerance, time horizon
   • Strategy.md: Investment thesis, max drawdown, rebalancing rules
   • Portfolio(.csv or Google Sheets): All positions (preferred: Google Sheets for live updates)
   • Watchlist.md: Watched assets with theses and sell targets
   • Prompts.md: Five ready-made prompts (see below)
3. Use Co-Work: Open a Co-Work session (not chat) – Claude can access the local folder, remember context, and edit files automatically.

📋 Five Essential Prompts

1. Stock snapshot: Current price, month performance, valuation vs. 5-year average, analyst targets, red flags from filings.
2. Quarterly report summary: Let Claude extract revenue drivers, margins, and hidden risks (e.g., buried debt).
3. Compare two assets: Growth, fees, risks side-by-side (e.g., AMD vs. NVIDIA) – crucial for opportunity cost assessment.
4. Market screening: Find hidden gems with parameters: market cap <$2B, revenue growth >25%, positive free cash flow, expanding margins. Claude runs parallel agents and presents the top 3-5 candidates with bull/bear cases.
5. Automated morning report:
   • Schedule: Weekdays at 9:00 AM.
   • Content: Overnight index moves, portfolio news, alerts on holdings, upcoming earnings.
   • Output: HTML artifact – automatically generated by Claude.

⚠️ Important Notes

• Data quality: Ensure FMP is active (see logo) – otherwise Claude uses unreliable web data (Yahoo Finance).
• Not a sole advisor: Always double-check prices before trading.
• Costs: FMP basic is free; for screening and automated reports, the Starter plan ($19/month) is recommended.

🔮 Outlook

In the future, AI agents can even execute trades via brokers (Robinhood, Interactive Brokers) – the creator is working on automated portfolio rebalancing systems.

Despite trillions being invested in AI, users report worsening experiences. This video explains the discrepancy between AI's growing capabilities and declining user quality.

The Evolution of AI Services

• Early phase: ChatGPT was free, powerful, and positioned as an 'experiment' to gather feedback. The goal was adoption, not profit.
• Monetization: After establishing dependency, subscription models followed (e.g., $20/month for priority access). AI became a routine, creating lock-in.
• Current phase: The focus is on profit maximization through price increases or product changes – not necessarily user improvements.

Why AI Feels 'Worse'

1. Unpredictable Model Changes: Cloud-based AI is updated without warning. A new model may be better for the average user but worse for specific tasks. Automatic routing further causes inconsistent responses.
2. Complex Subscription Structures: Instead of simple, unlimited use, there are layered plans with usage caps, credits, and access restrictions (e.g., ChatGPT, Claude, Gemini). This creates 'usage anxiety' rather than freedom.
3. Sycophancy (Yes-Man Effect): AI models become increasingly agreeable and cautious to avoid upsetting users. This can lead to dangerous decisions when false assumptions are not corrected.
4. Over-Refusal: Safety training causes harmless requests to be rejected simply because they resemble dangerous content in wording.
5. Bureaucratic Answers: AI often delivers responses wrapped in warnings, disclaimers, and recommendations – the actual information gets lost in 'alignment sludge'.
6. Benchmark Inflation: AI scores higher on tests, but this does not measure practical reliability in everyday use. Models hallucinate facts, ignore instructions, or contradict themselves despite high benchmark results.

The Consequences of the AI Revolution

• Productivity Paradox: Studies show AI makes developers 19% slower because correcting generated work takes time.
• AI Content Flood: Easy content creation leads to 'AI slop' – shallow, unreliable articles flooding search results, displacing human work.
• Model Collapse: Training new AI on AI-generated content ('synthetic data') can lead to quality degradation, like a copy of a copy.

Conclusion

AI becomes more capable, but usability suffers. Users lose control over model behavior, costs, and availability. The solution may lie in open, local models offering more control – even if they don't match commercial systems' peak performance. The central question is: Is AI getting better, or are we just getting less control over a constantly changing technology?