Are Emergent Abilities of Large Language Models a Mirage?

TL;DR

This paper argues that what appears as emergent abilities in large language models may actually be artifacts of the metrics used to evaluate them, rather than true new capabilities arising from scale.

Contribution

The authors propose an alternative explanation for emergent abilities, showing they can result from metric choices rather than fundamental model changes, supported by mathematical modeling and empirical tests.

Findings

Emergent abilities depend on the choice of evaluation metrics.

Better metrics can eliminate the appearance of emergent abilities.

Emergent abilities may not be intrinsic properties of scaled models.

Abstract

Recent work claims that large language models display emergent abilities, abilities not present in smaller-scale models that are present in larger-scale models. What makes emergent abilities intriguing is two-fold: their sharpness, transitioning seemingly instantaneously from not present to present, and their unpredictability, appearing at seemingly unforeseeable model scales. Here, we present an alternative explanation for emergent abilities: that for a particular task and model family, when analyzing fixed model outputs, emergent abilities appear due to the researcher's choice of metric rather than due to fundamental changes in model behavior with scale. Specifically, nonlinear or discontinuous metrics produce apparent emergent abilities, whereas linear or continuous metrics produce smooth, continuous predictable changes in model performance. We present our alternative explanation in a simple mathematical model, then test it in three complementary ways: we (1) make, test and confirm three predictions on the effect of metric choice using the InstructGPT/GPT-3 family on tasks with claimed emergent abilities; (2) make, test and confirm two predictions about metric choices in a meta-analysis of emergent abilities on BIG-Bench; and (3) show to choose metrics to produce never-before-seen seemingly emergent abilities in multiple vision tasks across diverse deep networks. Via all three analyses, we provide evidence that alleged emergent abilities evaporate with different metrics or with better statistics, and may not be a fundamental property of scaling AI models.