In this commentary, the author critiques the historical assumption that intelligence requires explicit self-referentiality or "strange loops," a concept popularized by Douglas Hofstadter’s *Gödel, Escher, Bach*. While theorists like Roger Penrose argued that Gödel's Theorem implied AI would fail without non-computable, self-referential mechanisms, the emergence of large language models (LLMs) like GPT 5.6 Pro and Fable suggests otherwise. The author notes that these models demonstrate advanced conversational intelligence and the ability to discuss complex topics—including self-reference and mathematical logic—without any intentional engineering of self-referentiality into the transformer architecture, GPU hardware, or training protocols. Instead, these capabilities emerged as a byproduct of large-scale pretraining. The text posits that LLM success validates a different paradigm: that intelligence is fundamentally rooted in prediction, compression, and finding upper bounds on Kolmogorov complexity. Much like how universality in computer science allows for self-executing code without explicit design, the ability to discuss the "self" in LLMs is presented as a free byproduct of achieving general linguistic competence within a massive discourse universe. Consequently, the author argues that the necessity of explicit self-referentiality for achieving high-level reasoning is a defunct concept in the face of modern scaling laws.