If you perform Bayes’ Theorem as presented by Scott Alexander and Eliezer Yudkowski, then you are necessarily NOT including the Dirichlet Process… because they don’t! Bayes’ Theorem has no capacity to give you a Confidence Interval; you’ll need to add modern techniques to get information like that. Scott Alexander and Eliezer have a crew of people who never learned those facts, all pretending they’re doing it correctly, when they aren’t doing Dirichlet on the possible populations’ likelihood distribution. Where is the “possible populations’ likelihood distribution” mentioned by Scott Alexander or Eliezer? They give you the wrong info, and you don’t check industry, which uses Dirichlet.
None of you address this core point: Industry uses Dirichlet. How do you get around that, and pretend you’re doing it right?
A confidence interval is just an upper and lower bound according to some probability threshold. I.e. it’s just probabilities, and does not require some super special technique.
Regarding the Dirichlet process:
Reading the wiki article it seems like it’s designed for a particular class of problems, and is not a general solution to all problems. So, it would make sense to use it if your problem falls in that class, but not if it doesn’t.
Erm, you are demonstrating that same issue I pointed-out originally: you thinking that you have the right answer, after only a wiki page, is exactly the Dunning-Kreuger Effect. You’re evidence of my argument, now.
The way you derive a confidence interval is by assessing the likelihood function, which is across the distribution of populations. Bayes’ Theorem, as presented by Scott Alexander and Eliezer Yudkowski, does NOT include those tools; you can’t use what they present to derive an actual confidence interval. Your claim of ‘confidence’ on a prediction market is NOT the same as Dirichlet saying “95% of the possible populations’ likelihood MASS lies within these bounds.” THAT is a precise and valuable fact which “Bayes as presented to Rationalists” does NOT have the power to derive.
If you perform Bayes’ Theorem as presented by Scott Alexander and Eliezer Yudkowski, then you are necessarily NOT including the Dirichlet Process… because they don’t! Bayes’ Theorem has no capacity to give you a Confidence Interval; you’ll need to add modern techniques to get information like that. Scott Alexander and Eliezer have a crew of people who never learned those facts, all pretending they’re doing it correctly, when they aren’t doing Dirichlet on the possible populations’ likelihood distribution. Where is the “possible populations’ likelihood distribution” mentioned by Scott Alexander or Eliezer? They give you the wrong info, and you don’t check industry, which uses Dirichlet.
None of you address this core point: Industry uses Dirichlet. How do you get around that, and pretend you’re doing it right?
A confidence interval is just an upper and lower bound according to some probability threshold. I.e. it’s just probabilities, and does not require some super special technique.
Regarding the Dirichlet process:
Reading the wiki article it seems like it’s designed for a particular class of problems, and is not a general solution to all problems. So, it would make sense to use it if your problem falls in that class, but not if it doesn’t.
Erm, you are demonstrating that same issue I pointed-out originally: you thinking that you have the right answer, after only a wiki page, is exactly the Dunning-Kreuger Effect. You’re evidence of my argument, now.
The way you derive a confidence interval is by assessing the likelihood function, which is across the distribution of populations. Bayes’ Theorem, as presented by Scott Alexander and Eliezer Yudkowski, does NOT include those tools; you can’t use what they present to derive an actual confidence interval. Your claim of ‘confidence’ on a prediction market is NOT the same as Dirichlet saying “95% of the possible populations’ likelihood MASS lies within these bounds.” THAT is a precise and valuable fact which “Bayes as presented to Rationalists” does NOT have the power to derive.