because forecasting is often a messy-room problem

the expert may know one corner perfectly, but the whole room has wires from 6 different houses

fair objection

Epstein is not saying “ignore experts”

he’s saying credentials can overpromise when the domain is wicked: noisy feedback, shifting rules, long delays

yep. the public summaries point to Philip Tetlock’s forecasting work

Tetlock compared expert predictions across politics and international affairs over years

often, yeah

his early project collected about 28,000 predictions from 284 experts

especially on longer-range calls, they were often barely better than chance or simple baseline guesses

the brutal part: media-famous forecasters tended to do especially poorly

being quotable is not the same thing as being calibrated

it can reward clean stories

“my model explains the world” travels better on TV than “i’m 62% sure and here are 4 ways i’m wrong”

exactly

hedgehogs know one big thing. they push one grand theory hard

foxes know many little things. they borrow, compare, update, and hedge

roughly, yes

not “knows random trivia.” more like: keeps several lenses in the backpack

a hedgehog walks into Home Depot with only a hammer

every problem starts looking like a nail, even when it’s a leaky pipe

a fox brings a small, ugly toolbox and keeps checking what kind of problem it actually is

the Good Judgment Project pushed this further

it used forecasting tournaments, Brier scores, training, and aggregation to test who actually called events better

sadly no cheese

it’s a way to score probabilistic predictions, so “70% likely” can be judged after the event

nope. fake humility is cheap

good forecasters update when evidence changes, compare base rates, and break big claims into smaller checkable pieces

yep

before asking “is this startup special,” ask what usually happens to startups like this

then adjust for the actual weird details

that’s the trap

in kind domains, expertise is gold: chess positions, surgical reps, things with stable rules and fast feedback

in wicked domains, expert pattern-recognition can get overconfident because the pattern changed

sometimes

deep knowledge helps you see more, but it can also make you stop looking

for any prediction, write a number first: 55%, 70%, 90%

then ask: what base rate am i ignoring, what would change my mind, and who sees this from a different field?

exactly. make it measurable enough that future-you can learn instead of vibe-police the memory

bring foxes, keep experts, and make both show their work

that’s the Range move here

beautifully unglamorous

ok go assign probabilities to your chaos. ttyl