alexknvl

My suggestion to refresh the feeling of wonder is to go to site directories instead of the search engine of choice, like: https://dmoz-odp.org

Perhaps you think there are certain feedback loops that need to be broken (your model is usually a static representation of the reality), or perhaps you prefer low rate of false negatives / false positives, or perhaps if a model is uncertain you would like to defer to a human.

If a model predicts high probability of re-offence, you might decide to delay an action as an example, or re-examine the case in more detail.

I get a feeling that machine learning practitioners are not properly trained to recognize such subtleties, everything is a binary softmax output these days :)

EDIT: This seems to stem from the obsession with purely discriminative models that either directly model a binary function y = f(x) or model a function that returns some kind of score, which is later compared to a threshold y = f(x) > theta. Neither of these lend themselves nicely to this conceptual separation between modeling and decision making.

98% accuracy, 100 seconds, crash twice. Take that, (data) scientists!

[1] - https://www.reuters.com/article/us-europe-migrants-germany-s...

[1] - http://www.telegraph.co.uk/news/2017/04/25/migrant-crime-ger...

[2] - https://papers.ssrn.com/sol3/papers.cfm?abstract_id=2903116

What you are talking about is completely orthogonal to the first amendment. Everyone has a right to speech and everyone is protected from vehicular assault & murder by law. I hope that everyone on Hacker News condemns that act of violence.

Could you please point to the source of your claim that he was a "card-carrying Nazi"?

Should we just throw them into water and see if they float? Or administer a "are you a Nazi" test on a polygraph?

EDIT: This is a problem of the current political climate where words have lost their meaning. A lot of people even remotely right of center have been called a Nazi or "alt-right" or "white-supremacist".

Now, I have no idea which one is the original, but AFAIU once you take a jpg and scan it into pdf, it's easy to move the text around (?).

EDIT: I hope my implication is clear. I think that the jpg file is more likely to be the original photo. It was later scanned as a pdf. Then OCR'd version was edited a bit by merging parts of the text (or only parts of it were OCR'd). Then a journalist opened it up in an editor and saw those layers.

Going in the opposite direction would require someone to produce an authentic looking jpg from an obviously OCR'd pdf. That seems like less likely to me.

Why does everything have to be about politics nowadays? I want an unbiased analysis. It is plausible, but not certain. There is a bunch of assumptions here that I'd rather keep as hypotheses.

There is a statement of the following fact: "It is hard to express heterogeneous array type in most statically typed programming languages and verify operations at compile time." Well, duh. AFAIK this would require dependent types* (not something that you get for granted in mainstream languages).

But can we check types in runtime? Yes. Does it mean that it can only be done in a dynamic language? No. It can be easily done in any language that has integers and arrays. Even x86 assembly is enough.

* Scala's type system should be enough, if we use shapeless.HList's for column and row types and then a ton of implicits to resolve the resulting type and check that everything is sound.

Having a type system that can catch some errors at compile time but not all is better than having a type system that doesn't catch any.

Having a static type system doesn't mean that you can't use dynamic typing as much as you want (e.g. in Scala you can make all variables of type `Any` and use type refinements whenever you want to call a method).

Define "the best", and it won't take much time to figure out who it is.

P.S. Personally, my choice would be Paul Graham (measured by the number of his essays I read and liked).