Truth is not a direction: a Tarski attack on LLM probes | Hacker News

A Hacker News discussion argues that phrases like “this sentence” do not literally refer to themselves, because a sentence is not a speaking subject.

The post says apparent self-reference is a category error between the speaker and the written or spoken sentence.

It invokes Tarski-style semantics to question whether self-referential probes for LLMs are conceptually sound.

The broader point is that if reference is ill-defined here, some language-model evaluation prompts may be built on shaky ground.