The paper investigates how depth growth and looping in language models influence in-context learning (ICL). It shows that models with these architectural features rely more on contextual information, acquire ICL earlier, and retain it better than fully dense counterparts. Attention analysis reveals that depth growth propagates early induction heads across layers, and ablations confirm that these models depend more on induction-ranked heads for few‑shot performance and token prediction.
By Ferdinand Kapl, Kaitlin Maile, Stefan Bauer, Johannes von Oswald
arXiv:2605. 28854v2 Announce Type: replace-cross Abstract: Large language models (LLMs) exhibit remarkable flexibility in adapting to novel tasks from in-context examples without parameter updates, a capability known as in-context learning (ICL).
By Hua-Dong Xiong, Li Ji-An, Robert C. Wilson, Kwonjoon Lee, Xue-Xin Wei
arXiv:2601. 21996v2 Announce Type: replace-cross Abstract: While Mechanistic Interpretability has identified interpretable circuits in LLMs, their causal origins in training data remain elusive.
By Jianhui Chen, Yuzhang Luo, Liangming Pan
The study investigates how language models acquire in-context learning (ICL) by distinguishing between copying a nearby demonstration and integrating information across a prompt. Across 17 models, roughly one-third default to copied answers, a few are undecided, and the remainder favor rule-consistent answers that yield higher accuracy and are guided by the requested value rather than the nearest conflicting example. Analyses show that models prioritize task-relevant prompt tokens before accuracy surpasses chance, and that synthetic or fine-tuned training can shift models toward compositional responses or improve accuracy when prompts reassign label meanings.
By Minglai Yang, Xinyu Guo, Renxiong Wang, Sunjie Hou, Daniel Yue Zhang, Yunzhong He
The paper investigates when in‑context learning (ICL) can succeed even if the instruction or target task is not covered by a model’s pretraining data. It introduces a theoretical framework that distinguishes two types of distribution shift—unsupported instructions and unsupported target tasks—and shows how demonstrations can mitigate these shifts. The authors further extend the analysis to long‑horizon tasks that can be decomposed into pretraining‑supported subtasks, providing error bounds that explain how ICL can handle tasks beyond the model’s original support.
By Xuhan Tong, Yuchen Zeng, Jiawei Zhang
arXiv:2507. 01414v2 Announce Type: replace Abstract: We introduce a new family of toy problems that combine features of linear-regression-style continuous in-context learning (ICL) with discrete associative recall.
By Sultan Daniels, Dylan Davis, Dhruv Gautam, Wentinn Liao, Gireeja Ranade, Anant Sahai
arXiv:2609.36448v1 Announce Type: new
Abstract: Transformers have demonstrated remarkable in-context learning (ICL) capabilities, enabling them to perform new tasks without additional fine-tuning. Ho...
By Junze Deng, Daouda Sow, Sen Lin, Yingbin Liang
The paper investigates whether few-shot in-context learning (ICL) emerges similarly across different data modalities. Using a controlled cross-modality framework, the authors test the Convergent Emergence Hypothesis, which posits that tasks benefiting from ICL in one modality will also benefit in others. They find that paired-mapping ICL appears in six modalities—language, genome, integer sequences, time series, images, and proteins—outperforming baselines and showing correlated task effects in five of them, supporting the hypothesis in some but not all cases.
By Nathan Breslow, Seungwook Han, Daniel Hyunsoo Lee, Aayush Mishra, Anqi Liu, Daniel Khashabi
arXiv:2610.01712v1 Announce Type: cross
Abstract: In-context learning (ICL) enables a pretrained model to infer a task from demonstrations without updating its parameters. While much of the existing...
By Haotian Gu, Yizhou Xu, Lenka Zdeborov\'a
arXiv:2609.36337v1 Announce Type: new
Abstract: Tabular foundation models achieve strong performance by conditioning on labelled examples in context, but softmax attention limits their use on large d...
By David Schnurr, Felix Sarnthein, Thomas Hofmann, Imanol Schlag
The study investigates how post‑training of large autoregressive language models (ARMs) into masked diffusion models (MDMs) affects their internal computation. Across two 7B ARM‑MDM families and four diagnostic tasks, the authors find that MDMs retain much of the ARM’s high‑attribution pathways on prefix‑dominant tasks, but reorganize computation toward earlier layers on globally constrained tasks. Component‑level probes reveal that ARMs depend on sharply specialized components, whereas MDMs show weaker specialization and more diffuse output‑space alignment.
By Injin Kong, Hyoungjoon Lee, Yohan Jo
The paper investigates how pre‑training on random input‑output mappings (memorization tasks) can transfer to downstream tasks. It discovers two unexpected patterns: equivalent transfer, where each pre‑training epoch saves roughly one fine‑tuning epoch, and non‑equivalent transfer, where pre‑training on a mismatched task can be more efficient than training directly on the downstream task. Ablation studies reveal that transfer consists of a trivial magnitude‑driven effect in the last layer and a non‑trivial structure‑driven effect linked to covariance in other layers.
By Yimiao Yu, Florentin Guth