Think in probabilities, not words

Tetlock noticed that the ordinary guessers and the really good guessers spoke almost different languages. The ordinary ones used words - might, could, likely, poised, well-placed. The best ones used numbers - "35% chance," "closer to two in three." That difference in words turned out to be a difference in thinking, and it's one of the cheapest upgrades an everyday investor can make.

Here's the simplest way to feel it. Say you tell your family, "I might do well in Friday's test." Can that sentence ever be marked right or wrong? No - whatever score you get, you can still say "well, I might have." It's slippery on purpose. Now say instead, "I'll score above 80% on Friday's test." That one sticks its neck out. Friday comes, the score arrives, and you find out: right or wrong. And only the second sentence teaches you anything - because only the second one can ever be checked.

That's the whole idea. A view with no number and no date can never be checked, and anything that can never be checked can never teach you. Words like "likely" are stretchy: tell ten people something is "likely" and each one secretly pictures anything from 55% to 90%, so everyone just hears what they already believed. The best guessers do the slightly uncomfortable thing instead - they put a number on it. Not because the number is perfect (it isn't), but because a number can be proved wrong, and being proved wrong is where all the learning hides.

Let's slow down on why a number beats a word, because it feels backwards at first. Surely, when you're unsure, a soft word like "probably" is the honest choice, and a hard number is fake precision? It's the other way around, and here's the everyday way to feel it. Two friends are deciding whether to carry an umbrella. Aarvi says "it might rain." Aarohi says "I'd say 70% chance of rain." Aarvi's sentence is safe - whatever the sky does, she was right, because "might" covers everything. But it also helped nobody decide anything, and tomorrow she'll have learned nothing about her weather sense. Aarohi stuck her neck out. If it rains most of the time she says "70%," her instinct is well-tuned; if it stays dry, she'll notice she runs too gloomy and adjust. The number didn't make Aarohi more sure than Aarvi - it made her accountable, and accountable is the only state a person can improve from.

And notice the sneaky thing "likely" does in a group. Imagine a WhatsApp group where someone writes "this stock will likely do well." One reader pictures a near-certainty and puts in half his savings. Another pictures "a bit better than a coin flip" and buys a token amount. They think they agree - they used the same word - but they've made wildly different bets, and only one of them can be right about what "likely" meant. A number ends the confusion instantly. "65%" means the same thing to all ten readers. Words feel friendlier, but they let everyone quietly hear what they already wanted to; a number forces one shared, checkable meaning.

A guess you can learn from has three parts, and a fuzzy view is usually missing all three. First a number - how sure you are, as an actual number, not a word. Second a target - the exact thing that has to happen, spelled out clearly enough that two people would agree on whether it did. Third a date - by when, so the clock can eventually run out and settle it.

Think of the three parts as the three things a referee needs before a match can even be scored. The target is the goal-line - everyone must agree exactly what counts as a goal, or you'll argue forever. The date is the final whistle - without it the game never ends, so no result is ever declared. And the number is your called shot - how sure you were you'd score - which is what lets you learn whether you're a sharp judge of your own chances or a wishful one. Drop any one of the three and there's no match to score: a target with no date plays on forever, a date with no clear target ends in a shouting match, and either one without a number teaches you nothing about your own aim. All three, together, turn a private hope into something reality can referee.

Tetlock also found that most people have only three settings in their head - yes, no, and maybe. The trouble is that "maybe" swallows everything that's uncertain, which is nearly everything worth guessing about, so it does no real work. The best guessers use lots of settings: they'll tell a 60% apart from a 70% and bet differently on each. You don't need fancy decimal points. You just need to break the giant "maybe" into smaller steps - because a "barely more likely than a coin flip" call and a "pretty darn sure" call should get very different bet sizes.

Picture the difference as a switch versus a dial. A three-setting mind is a switch with a broken middle: hard "yes," hard "no," and one enormous "maybe" that everything unclear gets dumped into. The problem is that almost every interesting question about a company lives in that middle. "Will profits grow next year?" isn't a clean yes or no - it's a 55% here, a 75% there - and a switch simply can't tell those apart. So the person with only three settings treats a barely-better-than-a-coin-flip guess and a near-sure thing as the same "maybe," and, fatally, bets the same amount on both.

You don't need to become a maths whiz to use the dial. A simple ladder of rough rungs is plenty: "very unlikely" is around 10%, "unlikely" around 30%, "even" is 50%, "likely" around 70%, "very likely" around 90%. The trick is just to place your view on a rung and say the number out loud, because the number is what carries into the next decision - how much to bet. Two calls you'd both wave off as "maybe" might really be a 55% and an 80%; on the dial they're far apart, and the 80% honestly deserves a bigger bet than the 55%. The switch hides that difference; the dial pays you for noticing it.

Watch that turn into rupees. illustrative Say Aarvi has ₹1,00,000 she's willing to spread across a few ideas this year, and two of them both feel like "maybe." On the dial, one is a 55% ("barely better than a coin flip") and the other an 80% ("pretty sure"). A three-setting mind, seeing two "maybes," might casually put ₹50,000 on each - treating a near-coin-flip exactly like a strong conviction. The dial says: don't. Back the 80% with more and the 55% with less - perhaps ₹35,000 on the shaky one and ₹65,000 on the strong one - so your money leans where your evidence actually leans. Same two ideas, same ₹1,00,000; the only thing that changed is that she let the number, not the vague word, decide the size of each bet. Over a lifetime of decisions, that single discipline - bet sizes that track your real odds - quietly separates the people who survive from the people who go all-in on every "maybe."

Let's try it on a real judgement. illustrative

Two investors look at the same medium-sized factory business whose margins have been squeezed by expensive raw materials. (A margin just means: out of every ₹100 of sales, how much the company keeps as profit.) The first investor writes in his notebook: "Margins should improve as costs come back down - looks well-placed." The second writes: "70% chance the profit margin is above 18% by the March 2026 results, up from 15% now."

For a whole year the two notes feel identical - both are hopeful, both sound sensible. Then the March 2026 results land with the margin at 16.5%. Now watch what each investor can do. The first is stuck: margins did improve, just less than he hoped, so his "should improve" was technically right and he learns nothing - he can't even tell whether he was too hopeful, because he never said how much. The second gets a clean lesson: he'd said above 18%, reality gave 16.5%, so his guess missed. Next time he'll be slower to assume costs come down right on schedule - and he can see in one glance that his hopes ran about one-and-a-half points too warm.

That's the entire payoff. The number didn't make the second investor smarter on the day he wrote it - margins were just as hard to predict for both of them. It made him correctable. And being correctable, over lots of guesses, is what slowly builds into real skill. A roughly-numbered guess you can check beats a beautiful phrase you never can.

Now stretch it across many calls and the magic compounds. Suppose the second investor writes a numbered call like this every month for two years. At the end he can line up all his "70% sure" notes and simply count: did about seven in ten actually happen? If his 70%s land 70% of the time, his aim is true and he can trust his own numbers when he sizes a bet. If his 70%s only land half the time, he's running hot, and now he knows it - by exactly how much - and can dial his confidence down. The first investor, with his drawer full of "should improve" notes, can never do this arithmetic, because you cannot count how often "should improve" came true. He is stuck feeling like a good judge forever, with no way to check. The number isn't just braver on the day; it's the only thing that ever lets you audit yourself.

Listen to how stock views actually get shared here and you'll hear a language built to dodge scoring. "Long-term story is intact." "Structurally well-placed." "Bahut potential hai." "Should compound nicely." Every one of these sounds thoughtful and commits to absolutely nothing - no number, no target, no date - so nobody can ever be marked right or wrong. The person saying it gets to sound wise for years while never once being on the hook.

You don't have to argue with any of it. You just quietly translate it into a checkable form before it touches your money. Someone says a stock is "poised for strong growth"? In your own notebook, rewrite it: "there's a __% chance sales grow more than % by March 20." The moment you try to fill in those blanks, one of two useful things happens. Either you find you can fill them with a straight face - and now you own a real, scoreable view - or you realise you can't, which tells you the exciting phrase was hollow all along. This one habit, turning every soft tip into a number-target-date, is the cheapest upgrade an Indian investor can make, precisely because the surrounding chatter is so allergic to it. And it keeps you honest with yourself too: it's just as easy to comfort yourself with "long-term story intact" as it is to hear it from a channel.

The slip is mixing up fuzziness with being careful. Saying "I don't want to slap a fake number on something so uncertain" sounds modest, but it quietly buys you the comfort of never being marked. The truly modest move is the opposite: commit to a rough number because you're unsure, so reality can correct you.

The other slip is letting soft words do a hard word's job. Poised, well-placed, structural, over the long term - these feel like real thinking but commit to nothing. They're the vocabulary of someone hiding from a score.

A number is a wonderful tool, but a tool can be misread, so here are the honest limits.

First, a number is not the same as knowing. Writing "80%" does not make a thing 80% likely; it only records how sure you felt. A confident-sounding figure can be pulled out of thin air and still be dead wrong. The number's whole value comes later, when reality checks it and you see whether your 80%s really land 80% of the time. Until you've built up that track record, treat your own percentages as honest guesses to be tested, not facts to lean on. Precision on the page is not the same as accuracy about the world.

Second, don't let the decimal points bully you. Writing "72.5%" instead of "roughly 70%" adds no real knowledge - it just borrows the look of science to make a rough guess feel solid. On genuinely uncertain questions, round numbers are the honest ones: 30, 50, 70, 90. False precision is its own trap, because a very exact-looking number tempts you to bet as if you were very certain, when you're really just guessing neatly.

Third, some things shouldn't be forced onto the dial at all. A few questions are so open - a freak event, a far-off future, something genuinely unknowable - that slapping a confident percentage on them is play-acting, not forecasting. There's no shame in writing "honestly, too uncertain to number well - so I'll bet small." That is the probabilistic answer. And none of this machinery, however tidy, ever tells you what to buy: it sharpens how you weigh a view and size a bet, and stops right there.