Conjecture-Driven Learning (Collecting and Connecting Dots)
Don't consume content — generate a conjecture, then hunt across every field for proof of it.
- Difficulty
- Moderate
- Time to result
- ~ongoing to results
- Steps
- 5
- Confidence
- 90%
Shah has no reading list and no favourite sources. His learning method inverts consumption: he forms a conjecture from dots he has already connected, then aggressively hunts for evidence across unrelated domains — chemistry, physics, history, evolutionary biology, human behaviour — and uses the result as the launchpad for the next conjecture. The compounding output is information asymmetry, which he defines as the actual substance of wealth.
Origin
Kunal Shah's own practice. He credits Danny Meyer (via Lenny) for the related 'always be collecting dots' phrasing, and notes LLMs have massively accelerated the proof-hunting step.
Core principles
- 01Wealth is nothing but information asymmetry; all the best companies have an unfair one.
- 02Curiosity is the demonstration that you are not proud of your expertise.
- 03Curiosity requires security — insecure people perform expertise instead of asking the dumb question.
- 04Every book (and every learned thing) makes your brain more porous to the next one.
- 05The world will now be unfair to people who can ask great questions, not those who look up basic answers.
- 06Reject 'favourites' — a favourite limits the mind.
How to run it
- 1
Form a conjecture from your existing dots
Rather than asking 'what should I read?', state a claim you suspect is true. Shah's example: everyone historically successful in vice businesses did heavy philanthropy to convert their reputation from vice-profiteer to respected citizen.
Pro tip A good conjecture is falsifiable and cross-domain — it should be checkable against history, biology or physics, not just your industry.
- 2
Hunt for proof with no domain boundaries
Go 'all over the place' to find whether the conjecture holds. Shah asked GPT for the vice-philanthropy conjecture and got ~50 named people, their vices, and how philanthropy repaired their reputations. He will follow the thread into chemistry, physics, human behaviour or universal principles without hesitation.
Pro tip LLMs are the ideal instrument here because you no longer depend on someone having pre-written the answer for a search engine.
Watch out Verify: the model will happily generate a confirming list. Confirmation is not proof.
- 3
Look for the parallel in an unrelated industry
Ask: where else has this pattern already run to completion? Shah, wondering whether lab-grown diamonds kill or grow the diamond market, went to the pearl industry — cultured pearls destroyed the royal status of pearls once everyone could have them.
Pro tip Derive the second-order insight explicitly: 'if this is true, where is the similarity of this found somewhere else?'
- 4
Chain the finding into the next conjecture
Use the proven (or disproven) conjecture as the dot that enables the next connection. The loop is: collect dots, connect dots, collect dots, connect dots — building an edge nobody else has.
Watch out You will never connect all the dots in one lifetime. The goal is compounding, not completion.
- 5
Protect the ability to ask dumb questions
Deliberately ask the question you're 'supposed' to know the answer to, in front of your team. Shah, as founder-CEO, will ask a 60-person WhatsApp group what a word means. The willingness requires security; the alternative — quietly Googling to protect your image — slows your compounding.
Pro tip Model it publicly. If the CEO can ask the naive question, everyone below can too.
Watch out People stop growing precisely at the point where they start defending their expertise instead of exposing their ignorance.
In the wild
Shah got obsessed with what lab-grown diamonds do to the diamond market. He found the historical parallel in pearls: cultured pearls made pearls universally available and the royal status collapsed. He reasoned lab-grown diamonds will produce a temporary profit window and then become parasitic on diamond status, destroying the profit pool.
→ A tradeable, testable investment thesis derived entirely from a cross-industry historical parallel — no analyst report required.
Shah hypothesised that people who built wealth on vices reliably use philanthropy to launder reputation. He put it to GPT and got roughly 50 examples with their vices and philanthropic arcs, then followed the research in every direction.
→ A validated pattern about status conversion, generated from a self-authored question rather than any existing source.
Common mistakes
Asking 'what should I read?'
Shah refuses to give book recommendations and admits he can barely finish a book. Curated lists outsource the conjecture step — which is the step that produces information asymmetry.
Re-reading the same book for comfort
Shah: repetition makes learning religious. You are evolving; the same insight arriving from a different book is a better signal than the same book arriving again.
Performing expertise instead of demonstrating curiosity
People who need to look expert never ask the dumb question, never adapt, and stop compounding. Curiosity requires the security to be seen not knowing.
Is it for you?
Best for
Founders, investors and operators who want a durable analytical edge rather than the same industry newsletters everyone else reads.
Not ideal for
Anyone who needs comprehensive, verified domain grounding fast (certification, compliance, medicine) — conjecture-hunting is exploratory, not systematic.
From the transcript
“I come up with conjectures in my head and then I go all over the place to find out if there are”
“my method of learning is constantly come up with conjectures from the dots that you've connected and then like absolutely work hard to find proof…”
“To me wealth is nothing but information asymmetry. All the best companies in the world have”
“everything that you every book you read makes your brain porous to read the next book.”
From the episode
Kunal Shah on winning in India, second-order thinking, the philosophy of startups, and more