LessWrong AI
2026-11-11 14:50 UTC
By Ashe Vazquez Nuñez
USR-0152-20261111-community-fo-1582e026
Vibe-Explaining Diagonalisation
Written as a MATS extension fellow, mentored by Richard Ngo. Diagonalisation is a bundle of techniques used to generate paradoxes of self-reference in math. These techniques show up under different names across set theory, logic, and computer science. This post takes a stab at abstracting all the diagonalisation arguments into an informal common denominator. [1] [2] The argument Consider a language L , which we discuss using a meta-language ML. You can think of ML as an exhaustive collection of pointers that refer to sentences of the language L . Diagonalisation negates the question: can we have L refer to ML exhaustively, the way that ML refers to L? i.e can we have L work as a meta-language for its own meta-language? Assume that you could have L refer to ML in this way. Pick a sentence of ML that refers to exactly the sentences in L that aren't referred to by any element of ML that they refer to. Since L can also refer to ML, pick a sentence of L that refers to . Does refer to ? If refers to , then since refers to and considering the definition of we must have that doesn't refer to . Suppose doesn't refer to , then since refers to and considering the definition of we must have that does refer to . Intuition Your language breaks on statements that require it to simultaneously serve the role of the base language and the role of the meta-language of its meta-language. ^ Diagonalisation does have a category-theoretic formalisation (of course). I don't understand it well enough…
Written as a MATS extension fellow, mentored by Richard Ngo. Diagonalisation is a bundle of techniques used to generate paradoxes of self-reference in math. These techniques show up under different names across set theory, logic, and computer science. This post takes a stab at abstracting all the diagonalisation arguments into an informal common denominator. [1] [2] The argument Consider a language L , which we discuss using a meta-language ML. You can think of ML as an exhaustive collection of pointers that refer to sentences of the language L . Diagonalisation negates the question: can we have L refer to ML exhaustively, the way that ML refers to L? i.e can we have L work as a meta-language for its own meta-language? Assume that you could have L refer to ML in this way. Pick a sentence of ML that refers to exactly the sentences in L that aren't referred to by any element of ML that they refer to. Since L can also refer to ML, pick a sentence of L that refers to . Does refer to ? If refers to , then since refers to and considering the definition of we must have that doesn't refer to . Suppose doesn't refer to , then since refers to and considering the definition of we must have that does refer to . Intuition Your language breaks on statements that require it to simultaneously serve the role of the base language and the role of the meta-language of its meta-language. ^ Diagonalisation does have a category-theoretic formalisation (of course). I don't understand it well enough…
Full article content could not be extracted automatically. Read the original below.
Source:
LessWrong AI
· lesswrong.com