zlacker

>>tromp+(OP)
The article discusses different encodings of numbers to give non-dense representations of numbers exceeding (2^64)-1. (These representations are inherently non-dense by the pigeonhole principle.) But I feel like this is missing a key point. A 64-bit string that we agree represents only numbers can represent 18446744073709551616 different numbers. The choice of what numbers are represented is completely up to us. If we want certain properties (dense integers) we end up with the highest being 18446744073709551615. If we want other properties (nearly logarithmic distribution, signedness, and good mapping to hardware for arithmetic) we might end up with FP64 with a maximum value around 10^308. And if we want no interesting property constraints except being dual to a program on a Turing machine, we end up with a busy beaver number. But... remember, we can choose any 18446744073709551616 values we want to be representable. There's no restriction on the interpretation of these strings; or, equivalently, the amount of external information required to explain the interpretation of these strings is unbounded. As a result, we can choose any computable number to be encoded by the string 64'b1, or by any other string, and exceed any desired bounds.

>>addaon+ab6
Came here to say that.

People sometimes mistakenly think that numbers or data in computers exist in some meaningful way.

Everything in a computer is a model or record or representation that serves a purpose. All bugs and limits are features.

>>teknop+uh6
> People sometimes mistakenly think that numbers or data in computers exist in some meaningful way.

That's basically Platonism. I think it's a reasonable position for some things, e.g. Booleans (two-valued logic), natural/integer/rational numbers, tuples, lists, binary trees, etc. I think it's meaningful to talk about, say, the number 2, separately from the way it may be encoded in RAM as a model of e.g. the number of items in a user's shopping cart.

This position gets less reasonable/interesting/useful as we consider data whose properties are more arbitrary and less "natural"; e.g. there's not much point separating the "essence" of an IEEE754 double-precision float from its representation in RAM; or pontificating about the fundamental nature of a InternalFrameInternalFrameTitlePaneInternalFrameTitlePaneMaximizeButtonWindowNotFocusedState[0]

The question in the article is whether lambda calculus is "natural" enough to be usefully Platonic. It's certainly a better candidate than, say, Javascript; although I have a soft spot for combinatory logic (which the author has also created a binary encoding for; although its self-interpreter is slightly larger), and alternatives like concatenative languages, linear combinators (which seem closer to physics), etc.

[0] https://web.archive.org/web/20160818035145/http://www.javafi...