Emergence of Hidden Capabilities: Exploring Learning Dynamics in Concept Space

TL;DR

This paper introduces a framework called concept space to analyze how generative models learn and develop hidden capabilities, revealing that latent abilities emerge suddenly during training and are controlled by data properties.

Contribution

It proposes a novel concept space framework to study learning dynamics and hidden capability emergence in generative models, linking these phenomena to data properties.

Findings

Learning speed and order are influenced by concept signal properties.

Sudden turns in learning dynamics correspond to the emergence of hidden capabilities.

Hidden capabilities can exist without being elicitable through naive prompts.

Abstract

Modern generative models demonstrate impressive capabilities, likely stemming from an ability to identify and manipulate abstract concepts underlying their training data. However, fundamental questions remain: what determines the concepts a model learns, the order in which it learns them, and its ability to manipulate those concepts? To address these questions, we propose analyzing a model's learning dynamics via a framework we call the concept space, where each axis represents an independent concept underlying the data generating process. By characterizing learning dynamics in this space, we identify how the speed at which a concept is learned, and hence the order of concept learning, is controlled by properties of the data we term concept signal. Further, we observe moments of sudden turns in the direction of a model's learning dynamics in concept space. Surprisingly, these points precisely correspond to the emergence of hidden capabilities, i.e., where latent interventions show the model possesses the capability to manipulate a concept, but these capabilities cannot yet be elicited via naive input prompting. While our results focus on synthetically defined toy datasets, we hypothesize a general claim on emergence of hidden capabilities may hold: generative models possess latent capabilities that emerge suddenly and consistently during training, though a model might not exhibit these capabilities under naive input prompting.