Core

Emergent Behavior

Capabilities that appear in large models but are absent in smaller ones, arising from scale rather than explicit training — such as in-context learning or chain-of-thought reasoning.

Explained at five levels

Level 1

When an AI suddenly learns to do something cool that nobody taught it — like a surprise new trick!

Level 2

When AI models develop unexpected abilities as they get bigger — skills that weren't specifically trained for, like solving math or writing code.

Level 3

Capabilities that appear in large models but are absent in smaller ones, arising from scale rather than explicit training — such as in-context learning or chain-of-thought reasoning.

Level 4

Qualitative capability transitions observed at certain model scale thresholds, where abilities like arithmetic, translation, or analogical reasoning appear discontinuously as a function of parameters or compute.

Level 5

Phase-transition-like phenomena in scaling curves where task performance jumps non-linearly beyond a critical compute threshold — debated as to whether they reflect genuine emergence or are artifacts of evaluation metric choice and threshold effects.

Definitions are educational summaries. Terminology can vary by source and context.

Sources