The paper investigates how transformer models learn latent structure by training a small decoder-only transformer on three variants of the Alchemy benchmark. It finds that the model acquires different components of latent structure in discrete stages, with a notable asymmetry: it robustly composes fundamental transitions but struggles to decompose complex examples into atomic transitions. Layer‑specific causal interventions reveal plasticity windows where freezing layers delays or prevents stage completion, offering a detailed view of capability evolution during training.
By Rohan Saha, Farzane Aminmansour, Alona Fyshe
The paper compares latent representations in Selective State Space Models (SSMs) like Mamba and Transformers such as Pythia using Sparse Autoencoders. Across a 10‑million token corpus, 99.98% of Mamba features align closely with Pythia’s, supporting the Universality Hypothesis that core semantic representations are similar across architectures. A tiny 0.02% of features diverge, with Mamba’s recurrent bottleneck causing it to compress syntactic anomalies into polysemantic neurons, whereas Pythia’s attention can isolate distinct formatting edge‑cases.
By Rithin Nagaraj, Rupa Laalasa Oruganti, Prerna Subhashchandra Kunder, Ashwini M Joshi
arXiv:2608. 03921v2 Announce Type: replace Abstract: This paper offers a new interpretation of the Transformer during inference.
By Marco Giunti, Fabrizia Giulia Garavaglia
arXiv:2604. 02029v2 Announce Type: replace Abstract: Latent space is rapidly emerging as a native substrate for language-based models.
By Xinlei Yu, Zhangquan Chen, Yongbo He, Tianyu Fu, Guanting Dong, Cheng Yang, Chengming Xu, Yue Ma, Xiaobin Hu, Zhe Cao, Jie Xu, Guibin Zhang, Jiale Tao, Jiayi Zhang, Siyuan Ma, Kaituo Feng, Haojie Huang, Youxing Li, Ronghao Chen, Huacan Wang, Chenglin Wu, Zikun Su, Xiaogang Xu, Kelu Yao, Kun Wang, Chen Gao, Yue Liao, Ruqi Huang, Tao Jin, Zhucun Xue, Cheng Tan, Jiangning Zhang, Wenqi Ren, Yanwei Fu, Yong Liu, Yu Wang, Xiangyu Yue, Yu-Gang Jiang, Shuicheng Yan
The study examined whether brain-language model alignment reflects shared computational mechanisms or merely stable lexical‑semantic correspondences. Using whole‑brain encoding across Mandarin, English, and French, transformer representations predicted activity in a distributed network that overlapped across languages and remained stable across layers. Contextual embeddings and measures of prediction or compression did not outperform static lexical embeddings, suggesting that alignment is robust but not informative about shared computational processes.
By Ni Yang, Rui He, Philipp Homan, Iris Sommer, Davide Staub, Wolfram Hinzen
arXiv:2608. 03921v1 Announce Type: new Abstract: This paper offers a new interpretation of the Transformer during inference.
By Marco Giunti, Fabrizia Giulia Garavaglia
arXiv:2608.29034v1 Announce Type: cross
Abstract: A wide range of methods have been proposed for interpreting language models, delivering important insights into their inner workings. However, differ...
By Zhang Enyan, R. Thomas McCoy
The paper investigates how transformers acquire deep semantic dependencies, proposing a mechanistic framework that frames learning as a competition between surface statistics and deep semantics. It identifies a "Gradient Starvation" effect that suppresses error signals for sparse semantic dependencies early in training, delaying structural reasoning until a sudden phase transition. The study also explains the success of Chain-of-Thought strategies and introduces a topology‑aligned contrastive objective that improves variable binding performance by more than twice the gain of standard fine‑tuning.
By Jiangrui Zhao, Xiaoting Du
arXiv:2606. 17522v1 Announce Type: cross Abstract: Deep neural networks are widely believed to derive their expressive power from their ability to form \textbf{hierarchical representations}, capturing progressively more abstract and compositional features across layers.
By Vinoth Nandakumar, Qiang Qu, Pramod Thebe, Sakshi Khachariya, Tongliang Liu
arXiv:2607. 04525v1 Announce Type: cross Abstract: How concepts are represented in neural networks is a fundamental question in machine learning.
By Zhimin Hu, Lanhao Niu, Sashank Varma
arXiv:2603.18908v5 Announce Type: replace
Abstract: Independently trained language models often learn compatible late-stage representations, despite differences in training objectives, architectures,...
By Matt Gorbett, Suman Jana
How concepts are represented in neural networks is a fundamental question in machine learning. The dominant view treats concept representations as stationary geometric objects.