CONCEPT
Emergent Capabilities
The discovery — which nobody predicted and no one fully explains — that large language models acquire qualitatively new abilities at particular scale thresholds. Reasoning, translation, code generation, in-context learning: none were trained for explicitly; all emerged.
Emergent capabilities are the abilities a large model displays that were not present in smaller models of the same architecture trained in the same way. The term entered the field through Wei et al.'s Emergent Abilities of Large Language Models (2022), which documented that specific tasks showed step-function improvements as parameter count crossed certain thresholds, rather than the gradual improvement the pre-2022 literature expected. The concept has been contested (Schaeffer et al., 2023, argued the emergence was an artifact of discontinuous metrics) and at least partially vindicated (subsequent capability-elicitation work showed genuine qualitative shifts). The practical consequence for the AI-revolution moment is that scaling produced more than the scaling laws alone predicted.
In The You On AI Field Guide
Clarke's epilogue in You On AI notes that he predicted AI would arrive but missed the channel: "it would emerge from text prediction rather than logical programming." This is exactly the emergence phenomenon in its most concise