Emergent Abilities

Emergent Abilities refer to sudden, unpredictable leaps in capability that Large Language Models exhibit as their parameter count and training compute scale up.

For example, a model trained purely to predict the next word in a sequence might suddenly demonstrate the ability to perform arithmetic, translate languages, or write Python code once it crosses a certain size threshold, despite never being explicitly programmed to do so.

Understanding and predicting emergent abilities is a major area of research in AI safety, as models frequently develop capabilities that their creators did not anticipate.