What do Language Models Learn and When? The Implicit Curriculum Hypothesis

Fuente: arXiv
Guardado en:
Detalles Bibliográficos
Autores principales: Liu, Emmy, Sun, Kaiser, Li, Millicent, Lee, Isabelle, Tjuatja, Lindia, Huang, Jen-tse, Neubig, Graham
Formato: Preprint
Publicado: 2026
Materias:
Acceso en línea:
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866918437003984896
author Liu, Emmy
Sun, Kaiser
Li, Millicent
Lee, Isabelle
Tjuatja, Lindia
Huang, Jen-tse
Neubig, Graham
author_facet Liu, Emmy
Sun, Kaiser
Li, Millicent
Lee, Isabelle
Tjuatja, Lindia
Huang, Jen-tse
Neubig, Graham
contents Large language models (LLMs) can perform remarkably complex tasks, yet the fine-grained details of how these capabilities emerge during pretraining remain poorly understood. Scaling laws on validation loss tell us how much a model improves with additional compute, but not what skills it acquires in which order. To remedy this, we propose the Implicit Curriculum Hypothesis: pretraining follows a compositional and predictable curriculum across models and data mixtures. We test this by designing a suite of simple, composable tasks spanning retrieval, morphological transformations, coreference, logical reasoning, and mathematics. Using these tasks, we track emergence points across four model families spanning sizes from 410M-13B parameters. We find that emergence orderings of when models reach fixed accuracy thresholds are strikingly consistent ($ρ= .81$ across 45 model pairs), and that composite tasks most often emerge after their component tasks. Furthermore, we find that this structure is encoded in model representations: tasks with similar function vector representations also tend to follow similar trajectories in training. By using the space of representations derived from our task set, we can effectively predict the training trajectories of simple held-out compositional tasks throughout the course of pretraining ($R^2 = .68$-$.84$ across models) without previously evaluating them. Together, these results suggest that pretraining is more structured than loss curves reveal: skills emerge in a compositional order that is consistent across models and readable from their internals.
format Preprint
id arxiv_https___arxiv_org_abs_2604_08510
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle What do Language Models Learn and When? The Implicit Curriculum Hypothesis
Liu, Emmy
Sun, Kaiser
Li, Millicent
Lee, Isabelle
Tjuatja, Lindia
Huang, Jen-tse
Neubig, Graham
Computation and Language
Large language models (LLMs) can perform remarkably complex tasks, yet the fine-grained details of how these capabilities emerge during pretraining remain poorly understood. Scaling laws on validation loss tell us how much a model improves with additional compute, but not what skills it acquires in which order. To remedy this, we propose the Implicit Curriculum Hypothesis: pretraining follows a compositional and predictable curriculum across models and data mixtures. We test this by designing a suite of simple, composable tasks spanning retrieval, morphological transformations, coreference, logical reasoning, and mathematics. Using these tasks, we track emergence points across four model families spanning sizes from 410M-13B parameters. We find that emergence orderings of when models reach fixed accuracy thresholds are strikingly consistent ($ρ= .81$ across 45 model pairs), and that composite tasks most often emerge after their component tasks. Furthermore, we find that this structure is encoded in model representations: tasks with similar function vector representations also tend to follow similar trajectories in training. By using the space of representations derived from our task set, we can effectively predict the training trajectories of simple held-out compositional tasks throughout the course of pretraining ($R^2 = .68$-$.84$ across models) without previously evaluating them. Together, these results suggest that pretraining is more structured than loss curves reveal: skills emerge in a compositional order that is consistent across models and readable from their internals.
title What do Language Models Learn and When? The Implicit Curriculum Hypothesis
topic Computation and Language
url https://arxiv.org/abs/2604.08510