Abstract:Language modeling has shown us that transformers can discover latent structure from context, but the dynamics of how they acquire different components of that structure remain poorly understood, leading to assertions that models just remix training data. In this work, we use the Alchemy benchmark in a controlled setting (Wang et al.,2021) to investigate latent structure learning. We train a small decoder-only transformer on three task variants: 1) inferring missing transitions from partial contextual information, 2) composing simple rules to solve multi-transition sequences, and 3) decomposing complex multi-step examples to infer intermediate transitions. By factorizing each task into interpretable components, we show that the model learns the different latent structure components in discrete stages. We also observe an asymmetry: the model composes fundamental transitions robustly, but struggles to decompose complex examples to discover the atomic transitions. Finally, using causal interventions, we identify layer-specific plasticity windows during which freezing substantially delays or prevents stage completion. These findings provide insight into how a transformer model acquires latent structure, offering a detailed view of how capabilities evolve during training.
| Comments: | Preprint |
| Subjects: | Machine Learning (cs.LG) |
| Cite as: | arXiv:2511.19328 [cs.LG] |
| (or arXiv:2511.19328v2 [cs.LG] for this version) | |
| https://doi.org/10.48550/arXiv.2511.19328 arXiv-issued DOI via DataCite |
From: Rohan Saha [view email]
[v1]
Mon, 24 Nov 2025 17:20:42 UTC (18,581 KB)
[v2]
Wed, 22 Apr 2026 01:15:08 UTC (1,407 KB)
此内容由惯性聚合(RSS阅读器)自动聚合整理，仅供阅读参考。 原文来自 — 版权归原作者所有。