The study investigates how a shared relational representation across queries influences query‑specific choices in language models. Controlled entity‑binding experiments on GPT‑2 Small, Qwen2.5‑1.5B, and Llama‑3.2‑1B reveal a low‑dimensional relational background that is largely shared across query roles and can transfer to new names. Removing this component and restoring independently trained relation directions brings output contrasts closer to their original values than sign‑reversed or random controls, while the sign of the transferable relational representation can differ across models and prompt types.
arXiv:2610.18674v1 Announce Type: cross Abstract: Large language models rely on prompt context to perform in-context learning, follow formats, and adapt to local evidence. This ability, however, also...
arXiv:2610.32309v1 Announce Type: new Abstract: Why can a task-preserving rewrite turn a correct answer into a failure while a larger internal change leaves the answer intact? We develop a direction-...
arXiv:2610.27580v1 Announce Type: cross Abstract: Factual recall is a lookup: some representation inside the network serves as the key, and what that key is made of decides which queries can retrieve...
The paper demonstrates that language models can acquire new capabilities from post‑training data even when the training text is unrelated to the target task. Using a method called Active Taskless Distillation (ATD), the authors show that a single word from a teacher model can transfer knowledge to a student model without any target‑task examples or teacher logits. Experiments on Qwen2.5-1.5B reveal significant performance gains on HumanEval+ and improvements in scientific knowledge, commonsense reasoning, and reading comprehension across various model families.
The study investigates how task context reshapes the roles of shared attention heads in large language models. By applying identical graph facts and measuring interventions on attention heads and downstream activations, the authors show that task changes redistribute influence across five open‑weight models, causing the same heads to respond to different facts and affect later components differently. Linear response maps before and after activation edits reveal significant reductions in prediction error, indicating that task context alters both the facts processed and the downstream impact of shared components.
The paper investigates which linguistic operations can be replaced by latent task representations instead of in‑context prompts. A 2.6 M‑parameter network learns to encode the geometry of a few‑shot support set and applies an additive update to a frozen GPT‑2 model, achieving zero‑shot performance comparable to or better than 10‑shot in‑context learning for forward inflection and lemmatization. However, for arbitrary pairings such as antonymy, the latent approach only reaches about half the performance of in‑context learning, indicating that productive rules amortize well while memorized pairings do not.
The paper investigates what aspects of language model behavior are controlled by activation steering. By introducing Cross‑Encoding Steering Evaluation, the authors show that steering effects often follow the extraction index of answer identifiers rather than the semantic content of the answers, especially at deeper layers. They also find that a low‑rank output‑sensitive component captures most of this effect, and that different datasets (NormBank, MNLI, SC101) exhibit varying preferences for extraction‑index versus semantic‑label following.
arXiv:2606. 07555v1 Announce Type: cross Abstract: Glossaries, technical specifications, and system prompts routinely ask language models to use familiar words in unfamiliar ways.
arXiv:2607. 20524v1 Announce Type: new Abstract: Mean cross-positional attention degradation is widely reported in transformer interpretability, yet whether it causally limits contextual retrieval remains untested.
arXiv:2608. 11510v1 Announce Type: cross Abstract: Congruency effects, observed in conflict tasks such as Stroop and flanker tasks, have been investigated for nearly a century in psychology and neuroscience, but their mechanistic basis is not fully understood.
arXiv:2607. 16741v1 Announce Type: new Abstract: B\"urger et al.
The study investigates how well language models can generalize from fine‑tuned facts by distinguishing active facts—those that support further learning—from inert facts. Using a novel relabeling experiment (e.g., mapping Paris → Europe to Paris → crab), the authors fine‑tune models on a subset of subjects and test whether the learned mapping extends to unseen subjects sharing the same original answer. Across eleven models and 70 relations, they find a strong correlation (Pearson r = 0.87–0.92) between the linear readability of the original relation in pretrained representations and the model’s ability to generalize, a relationship that persists after controlling for various factors and remains robust to label randomization and semantic category changes.