Hacker News new | ask | show | jobs
by carabiner 1047 days ago
I too like to assign probabilities to my vibes. Much more legitimate.
4 comments

Reminds me of how Data in Star Trek would say things like "There is a 79.85 probability of success".

I mean, maybe it is possible to assign probabilities to future events that you have never seen before, but somehow I doubt it (with 12.88% likelihood).

>but somehow I doubt it

You doubt it? That means you're assigning some underlying probability to the event (not) occurring. Why does it seem strange to you if people are explicit about their priors?

It's not that it's strange, it's that it is transparently silly to claim to precisely quantify in the absence of anything resembling sufficient information for the problem to be treated probabilistically.
How about this? I keep a running probability (e.g. a 32bit float) for the successful outcome of each future event I have absolutely no idea about. At first I make it 50%, but I gradually adjust with each result, such that I aim to minimize my bias. It might sound silly, but at the minimum it would encode some information about the kind of events I'm typically asked to predict. And I do expect it could end up something like 0.5642635 for a particular time in my life.
"Not even wrong." An event about whose outcome you by definition have no information ("I have absolutely no idea") is an event to which you cannot assign any probability, including by assuming it's a coin flip and then treating the assumption as axiomatic. It encodes no information whatsoever because there is none to encode, and it biases all future results because you're averaging over a range that includes "data" you made up.
I'm having trouble following your reasoning. "Not even wrong" typically is used to refer to something unfalsifiable, but I specifically provided an empirical process and even a mechanism to explain why it might work - a person in a given position would likely be asked to make predictions with a different base likelihood of success than a person in another position.

In any case, it is precisely my argument (and Scott Alexander's) that there are no events to which it is axiomatically impossible to assign probabilities.

Ya what if you just don’t know
Any jackass off the street can just not know things! If you want to get invited to the in-crowd's house parties, you're going to need to do a lot better than that.
I think it's the excessive precision that's funny.
I think this culturally comes from lesswrong and it really makes me laugh every time I see it
Yeah, Scott Alexander wrote a whole article about how people should use made up fake statistics instead of simply saying "I don't know"[1].

[1] https://slatestarcodex.com/2013/05/02/if-its-worth-doing-its...

Not always. "I don't know" is identical to saying "I have a uniform prior". In the particular case of LK-99, my prior is perfectly uniform - I know absolutely nothing about the subject matter and almost everything I read on the topic feels way above my head. As I refuse to be influenced by experts (or "experts") until there is some sort of consensus, I have absolutely no good reason to say anything but "I don't know".
A perfectly uniform prior suggests that you think that it's easy to make a room temperature super conductor and that people have made them before with equal likelihood to it being hard to make a room temperature superconductor and people haven't made them before, doesnt it?

I don't know much about superconductors, but I do know that people have been trying to make them work at room temperature, and generally failing

It’s the Bayesian way.
Common in tech: “95% of our users wont care if we make the product worse by…”
That's falsifiable though, in a way the other examples are not. You could sample the population of users after the change to the product