Predictions
What can be learned from the past and present reality to predict the future?
Every nervous system exists to predict — to model what comes next so the organism can survive and thrive.
What questions should you be asking to help make the invisible visible and build conviction in how the future needs to play out?
Why Prediction Matters
Prediction isn't a technique. It's what minds do.
The most effective way to reason about the future is to combine the outside view (base rates from history), the inside view (current signals dated and sourced), and falsifying conditions (what would change your mind). Then stake a probability and a date — and expose your reasoning to domain experts who can see what your model can't.
| Species | Prediction Capability | Survival Function |
|---|---|---|
| Bacteria | Chemical gradients | Move toward food, away from poison |
| Fish | Water pressure patterns | Avoid predators, find prey |
| Mammals | Complex world models | Navigate social hierarchies, plan hunts |
| Humans | Abstract future simulation | Build civilizations, create technology |
Your brain is a prediction machine. Every perception is a prediction about what's causing your sensory input. Every decision is a prediction about which action leads to better outcomes.
The quality of your predictions determines the quality of your life.
PREDICTION → ACTION → CONSEQUENCE → FEEDBACK → BETTER PREDICTION
This is the Validated Virtuous Feedback Loop applied to beliefs — validated (tested against evidence), virtuous (serves beyond confirmation bias), compounding (each cycle starts from a higher baseline of calibration). This is agency. This is survival.
Prediction as Commitment to Truth
A prediction is a commitment to truth. You can't claim conviction without defining what would change it.
The Truth-Seeking Protocol applies:
| Question | Investment Context | Life Context |
|---|---|---|
| What am I predicting? | Specific thesis | What outcome do I expect? |
| How confident am I? | Conviction 0-5 | How much would I bet? |
| How will I know? | Success criteria | What would prove me right/wrong? |
| What would change my mind? | Update triggers | What evidence would shift me? |
The feedback loop (monitor → evaluate → update) is what separates prediction from wishful thinking.
AI and Robotics represents a Supersonic Tsunami - Elon Musk
Better to ride the wave than get caught in it's impact zone.
Principles
Sequence: Nowcasting captures real-time signals (what is happening now). Forecasting builds scenarios from those signals (what might happen next). The Database records conviction scores. Evidence evaluation scores accuracy. This is the loop, not a menu.
| Capability | Key Prompt | When to Use |
|---|---|---|
| Evaluation | "Is this prediction worth tracking?" | Scoring accuracy of past predictions |
| Probability | "What's my base rate before new evidence?" | Estimating base rates for new predictions |
| Forecasting | "What does the future require to be true?" | Building future scenarios from current signals |
| Markets | "What do I know that the market doesn't?" | Reading collective belief about probability |
| Process | "What's my conviction level (0-5)?" | Recording and updating conviction scores |
| Superforecaster | "How do the top 2% forecast?" | The decomposition process that produces calibrated predictions |
Our live forecast — 13 dated, falsifiable predictions across six domains, scored against base rates and named falsifiers — lives on the frontend: /vision/predictions. That is the canonical record. The methodology pages above teach how each prediction was built.
The Problem
The problem isn't data — it's conviction.
- We consume endless forecasts but rarely act on them
- We see trends but can't size the bets
- We know change is coming but don't know when
- We understand probability intellectually but feel certainty emotionally
Predictions are not guesses. They're bets on which micro-moments will compound. The prediction "AI replaces 50% of knowledge work by 2027" is a bet that thousands of small collisions — each team adopting AI for one more task — will add up. The scoreboard tracks whether those collisions are happening. The prediction is what tells you which collisions to watch.
Character
What traits do superforecasters share?
Calibration — Know how confident you should be. A 70% prediction should be right 70% of the time—not more, not less.
Updating — Strong opinions, loosely held. New evidence changes beliefs. Ego doesn't protect priors.
Decomposition — Break big questions into smaller, answerable parts. Fermi estimation over gut feeling.
Intellectual Humility — "I might be wrong" is the prerequisite to "What am I missing?"
Curiosity — Genuine interest in how systems work and what drives change.
Practice
Use the tracking-predictions skill to maintain the prediction database systematically.
Daily
Before making decisions:
- What's my base rate for this outcome?
- What evidence would change my mind?
- Am I updating enough—or too much?
Weekly (15 min)
Scan for prediction-relevant signals:
- What news affects existing predictions?
- Flag predictions needing conviction updates
- Note: "Evidence strengthened" or "Evidence weakened"
Monthly (30 min)
Review and update:
- Update conviction scores with reasoning
- Add new predictions from month's observations
- Check: Any predictions ready to resolve?
Quarterly (1 hour)
Calibrate:
- Were my 70% predictions right ~70% of the time?
- Where am I consistently overconfident?
- Where am I consistently underconfident?
- What domains/horizons have gaps?
Engage Domain Experts
Your model has blind spots. A domain expert lives inside the system you are forecasting — they see the signals you cannot see from the outside. The most calibrated forecasters expose their reasoning to experts before they bet, not after they lose.
Three roles to recruit per domain:
- Practitioner — currently operates inside the system. Knows what is changing this week and what the metrics actually mean.
- Historian — has seen prior cycles. Knows which patterns repeat and which are genuinely new.
- Heretic — disagrees with consensus. Forces you to face the strongest counter-argument before reality does.
Cadence: one external expert conversation per domain per quarter. Not a survey, not a podcast — a 30-minute exchange where they can disagree with you in real time.
Protocol:
- Brief in advance — send a one-page note with your prediction, your base rate, your falsifiers, and the two questions you want their view on
- Listen before defending — the goal is to hear what your model misses, not to win the conversation
- Update on the record — within 24 hours, write what changed in your conviction and why. If nothing changed, name what would have changed your mind and didn't show up
- Compound the relationship — return six months later with what you got right and wrong. The second conversation is always better than the first
See Players for the archetypes worth recruiting and the Truth-Seeking Protocol for the discipline that keeps the exchange honest.
Tight Five for 2027
What's your platform for the AI transition?
Evolution Path
2024-2026: Foundations
- Build agentic infrastructure through DePIN
- Establish tokenized reputation systems
- Deploy AI-driven personalized learning
- Rise of AI workflow designers as highest-paid professionals
2027-2030: Human-AI Synergy
- Scale autonomous feedback loops with crypto agents
- Premium pricing for genuine human experiences
- Transition from traditional education to AI-personalized
- Implement UBI through tokenized AI task systems
2031+: Cultural Integration
- Full unification of physical and digital worlds
- Cultural motivators replace economic incentives
- Inner space exploration becomes primary human frontier
- New societal structures based on human agency
Implications for positioning:
Where will be the best place to live when intelligence has no moat and money is meaningless?
| Trad Lens (Now) | Future Lens (2031+) |
|---|---|
| Where can I make money? | Where can I live well? |
| Business opportunity | Good company |
| Talent pool | Values alignment |
| Scale capacity | Beauty / Nature |
| Tax optimization | Health / Longevity |
See Countries Framework for how this reorders country rankings.
Applied Prediction
Prediction frameworks for specific domains:
| Domain | Framework | Core Question |
|---|---|---|
| DePIN | Investment Appraisal | Does this infrastructure solve a real problem at scale? |
| State of the World | 2026-2028 Forecast | Six domains, two-year horizon, calibrated probabilities |
| Markets | Prediction Markets | What do I know that the market doesn't? |
Dig Deeper
- 2026-2028 Forecast — Our live state-of-the-world forecast with base rates and falsifying conditions
- Superforecaster — The process that separates the top 2%
- Perspective — Build the perception that enables prediction
- Decisions — Turn predictions into action
- Investing — Bet on your convictions
- Truth-Seeking Protocol — Mechanical enforcement of honest prediction
- Prophet Arena — Engineer a superforecaster
Context
- Performance — The scoreboard that reflects whether predictions compound
- Positioning — Where to focus based on what you predict
- Standards — Thresholds that make prediction testable
- Pipeline Nowcast PRD — Platform implementation: signal collection, nowcast scoring, prediction evidence ledger
- Belief System — How predictions keep belief grounded rather than drifting into narrative
- Planning — Judgment under uncertainty
- Investing — Conviction deployed as capital
- Working Memory — Track record that compounds
- Good Judgment Project — Superforecaster methodology
- ai-2027.com — AI trajectory modeling
- Farnam Street: Superforecasters — Calibration principles
Questions
What turns predictions from guesses into convictions you can stake decisions on?
- If predictions are bets on micro-moments, how do you know you've found the right moments to watch?
- When your nowcast (current signals) contradicts your forecast (future scenario), which do you trust — and what does that tell you about your model?
- Which of your conviction scores has zero evidence entries after 6 months — is the prediction wrong or is your collection process broken?
- At what threshold does a data point justify updating a conviction score — and who decides?
- If belief is the most AI-exposed system, what role does your prediction practice play in keeping your belief sovereignty intact?