How to bet on yourself without lying about the odds (ft. Bayes' theorem)

Apr 18, 2026 · Life

There's one idea that's quietly shaped how I make decisions more than almost anything else, and it's Bayesian thinking. I want to explain what it actually is, in plain language, and then get to the part I really care about, which is how I use it to think about the big bet I've made with my own life.

Let me start with a puzzle. Imagine you're at a hackathon and you get talking to someone, call them Sam. Within two minutes Sam is pitching you their idea, talking a mile a minute about how it's going to be huge, totally all in, eyes lit up. Here's my question. If you had to guess, is Sam more likely to be a founder who has actually raised money for a startup, or a student who just showed up to build something for the weekend? Assume it's one or the other.

Most people guess the funded founder, because that all-in, pitch-everyone energy just feels like a founder thing. And that's fair, it probably is more common among people who've actually gone out and raised money. So the instinct isn't wrong.

But there's a second fact almost everyone skips over, which is how many of each kind of person is actually in the room. At a hackathon it's overwhelmingly students and weekend hobbyists. Funded founders are rare. Call it ten weekend builders for every one founder who's raised money. So now you've got two facts pulling against each other. The pitch energy is more common among founders, but there are way more students in the room. The whole trick is how you put those two together.

Here's the way I picture it. Imagine a rectangle split to show how many of each there are, say one founder for every ten students. Now look at just the founder slice and ask how many of them have that loud all-in energy. Say roughly seventy five percent. Then look at just the student slice and ask the same thing, maybe fifteen percent. Sam has the energy, so he's in one of the two energetic groups. To figure out whether he's more likely a founder or a student, you just compare the sizes of those two groups. And when you do, the energetic-student group comes out about twice as big, even though the energy is rarer among students, just because there are so many more of them. The small group times a high rate still loses to the huge group times a lower rate. So Sam is actually about twice as likely to be a student. That's Bayes' rule. That's the whole mechanic.

Now here's the honest part. In real life I don't plug numbers into a formula and I almost never draw the rectangle. What I actually carry around are three habits that fall out of the math, and those are the useful thing. I first picked up this way of boiling Bayes down to three habits from Julia Galef's talk on it, which is still the clearest explanation I've come across.

One: remember your priors

In that hackathon example, the mistake is locking onto the vivid evidence (the pitch energy) and forgetting the boring background fact (the room is mostly students). That mistake has a name, base rate neglect, and it's everywhere once you start looking for it.

Here's a version that shows up constantly when you're building something. Say you check your analytics one morning and signups are down forty percent overnight. Your gut screams that the product is dying. And sure, a forty percent drop is real evidence, it should bump your worry up. But stop and remember the prior. How often is a sudden overnight cliff actually your product dying, versus a tracking bug, a broken script, a holiday, a logging change, or one big referrer going quiet? Almost always it's one of the boring ones. So even a scary number should only move "the product is dying" from maybe one percent to maybe eight percent. Go check the tracking before you rewrite the whole roadmap. The evidence was real and the prior still won.

Two: ask what you'd expect to see if you were wrong

Say you ship a new feature and within a day a few users tweet that they love it. You think, see, this proves it, people want this, we've got something.

But flip it around. Suppose you're wrong and the thing doesn't actually have legs. In that world, would a handful of nice tweets still show up? Of course they would. Early fans, friends, and plain polite people say nice things about mediocre products all day long. So a few kind tweets are almost as likely whether or not you've really got something, which means they barely tell you anything. And that's the trap most of us live in, and the reason this is the principle I lean on hardest. We collect the little moments that fit what we already hope is true and feel more sure every time, without ever asking the one question that matters: if I were wrong, would this look any different? The signals that actually count are the ones that would be unlikely if you were wrong, people paying, coming back on their own without a nudge, dragging their coworkers in. A compliment is only evidence if that person would have said no to a worse version.

Three: update a little at a time

This is sort of the flip side of the second one. Say you're convinced some new tool is overhyped, that the AI coding stuff is a toy and real engineers don't need it. One person raving won't flip you, and it shouldn't. But then a developer you actually respect ships something fast with it. Then a friend quietly stops complaining about it and starts depending on it. Then you try it on something annoying and it just works. Any single one of those is easy to wave off. But each one is a little more likely in the world where the tool is genuinely good than in the world where it isn't, so each one nudges you a bit. You don't flip your whole opinion overnight. You just keep nudging, until one day the honest position has quietly moved. These little flakes of evidence weigh almost nothing on their own, but if you actually pay attention to them they stack, and enough of them will break the branch. Most people either ignore the flakes completely or let a single one avalanche their whole opinion. The move is to keep updating in small steps.

That's Bayes, really. Start from what's normally true, notice what would actually be surprising, and shift your beliefs in proportion to the evidence instead of swinging between total certainty and total doubt. I don't think it's the answer to everything, the way some people treat it. But for the question of what to believe and how strongly, it's the best tool I've got. And the place it's mattered most for me isn't any of these little puzzles. It's my own life.

The prior was against me

Here's the bet. The statistically sensible version of my life is easy to describe. Finish college on time, don't fail classes, get a stable job, pay down the loan that brought me from Mumbai to the US, build up some savings, and take the big swings later, once I'd earned the room to absorb a failure. That path exists for a reason. For someone without family money, it's genuinely the safer line.

I did close to the opposite. I'm from Mumbai, lower middle class. I took a loan to study abroad. I quit a research assistant job that was basically pocket money but was also chipping away at my tuition month by month. And instead of protecting my GPA and lining up the predictable engineering career, I spent my time founding companies and hoping something would hit. Every one of those choices traded a high probability of a fine outcome for a low probability of a great one with a much wider range of how it could go.

It's tempting to tell that as a story about courage. It's just as easy to tell it as a story about being reckless. Bayes gives me a third option that's less flattering than the first and more useful than the second, which is to look at it as a bet and ask honestly what the odds actually are.

Privilege versus optionality

I want to be precise about one thing, because I used to be sloppy about it. I sometimes talked about "making your own privilege," and I don't think that's quite right. Taking a loan is not the same as having privilege. A wealthy person's safety net absorbs failure for free. A loan does the opposite. It pulls resources from my future self into my present and makes failing more expensive, not less.

So the honest version isn't that I manufactured a safety net. It's that I manufactured access. I converted future income into present opportunity. I borrowed optionality before I'd built up any security. Moving abroad, leaving the steady income, spending years building instead of optimizing for the safe route, all of that put me in rooms with people, markets, and chances that had a very low probability of ever showing up in my old life. But the debt and the tuition and the academic risk stayed completely real the whole time. That's exactly what makes it a real decision and not a fairy tale. I didn't delete the downside. I took it on in exchange for a wider spread of possible outcomes.

And this is why the same move means totally different things for different people. A founder with family money can try three startups and still go home safe. For someone carrying debt with no fallback, one failed startup can turn into a financial, academic, and immigration problem all at once. Same action on paper, completely different bet underneath.

The uncomfortable question

The question I'm tempted to ask is, how many people from a background like mine took risks like this and made it? But that one's almost impossible to answer cleanly, because I mostly only hear about the ones who won. The people who tried the same thing and it didn't work just quietly fold back into normal life, with the debt and the unfinished plans and the job they took to recover, and you never read about them. That's survivorship bias, and if I lean on it I'll trick myself.

The more useful question is, given everything I actually know about myself right now, what evidence should make my path more or less likely to work? Because Bayesian thinking is personal, but it can't be allowed to turn into flattery. I'm not allowed to just say the average startup fails but I'm different. I have to say why, and point at something. Have I built something hard. Have actual strangers paid me for something I made. Can I get unusually good people to work with me. Do I come out of a failure with reusable skills or just damage. Are my opportunities getting better over time, or am I just getting more comfortable with chaos. Those things move the odds. Wanting it badly does not. Courage is not evidence.

And the ledger has two sides. Customers who say they love it but never pay, deadlines I keep missing, debt getting riskier, health or grades slipping, the same approach failing over and over without teaching me anything new, me quietly redefining success so I never have to admit the original idea was wrong. That's all evidence too, pointing the other way, and the honest move is to count it. I can't tally up everything that makes me look exceptional while ignoring everything that makes my situation fragile.

The one thing Bayes alone won't save you from

Here's where expected value isn't enough on its own. A bet can look great on paper and still be a terrible idea if losing it takes away your ability to ever bet again. Picture two gambles with the same upside, say a ten percent shot at something life changing. In the first, the other ninety percent just lands you back in a stable career. In the second, the other ninety percent is unpayable debt, wrecked health, a deportation, no degree. The upside is identical. These are not the same decision, and it isn't close.

So the goal was never to avoid risk. It's to shape the risk so that failure is survivable, so that the bad outcome still leaves me standing with something. Protecting yourself from total ruin isn't cowardice. It's the thing that lets you stay ambitious, because it keeps you in the game long enough to take the next swing.

Treating ambition like a process

What this actually looks like, for me, is writing the bet down before I'm in the middle of it. What do I believe right now and why. What would genuinely raise my confidence, and what would lower it. When am I going to stop and reassess instead of just grinding. And what's the line I won't cross no matter what, the downside that has to stay survivable.

For a company that might sound like: over the next six months I expect to close a handful of paying customers, and if people keep praising the product but won't actually pay, I'll take that as real evidence against this market or this idea and update. If they pay, use it more, and introduce me to other people, I update the other way. The point of writing it down ahead of time is that it stops stubbornness from quietly turning into denial. It's a lot harder to move the goalposts when you wrote down where they were while you were still being honest with yourself.

Where this leaves me

The two failure modes are pretty clear once you see them. Priors with no action is just fatalism, deciding the average is your destiny and never trying. Ambition with no updating is just delusion, wanting something so badly you stop checking whether it's working. I'm trying to live in the narrow space between those. I make the bet, I watch what reality actually does, I update, and I keep enough of myself intact to bet again.

If I had to put the whole thing in one line, it'd be this. I wasn't privileged enough to ignore probability. I was just ambitious enough to try to change my posterior.