Spooky training quirk: emergence always shows up at step 13, no matter the setup
cephaloform · x · 2026-09-27
A developer reports a counterintuitive observation: no matter how they tweak parameters, optimizers, or algorithms, the "emergence" moment — sparks, nucleation, the jump from random to intelligence — always happens at step 13 of training. The post invites discussion on whether such fixed-step emergence is reproducible and what mechanism could explain it.
More from Fun
- Music Video for "I Am Actually Scared of Linear Algebra" Animated by Claude Opus 5.5 — anderssandberg · 2026-09-27
- User reports three OpenAI site bugs, gets asked to do free advanced diagnostics — StewartalsopIII · 2026-09-27
- Hot or Not founder credits 1999 Jester dataset as the pioneer behind collaborative filtering — jfiance · 2026-09-27
- Dev turns Annecy into a playable 3D boat tour using Claude Opus 5.5 and Three.js — techartist_ · 2026-09-27
- One-shot with Opus: a PHP 4.1.1 simulator with Zend Engine compiled to WebAssembly — generativist · 2026-09-27
- DHH on AI coding: thinking "architecture" is your fortress? Models are good at that too — tawnniee · 2026-09-27