The paper introduces Content-Invariant Gradient Attribution (CIGA), a method for selecting training data for large language models that mitigates surface-form bias in gradient-based attribution. CIGA identifies layers whose gradients are both influential and stable under semantics-preserving perturbations, then fuses these signals with semantic representations to produce robust data rankings. Experiments show that CIGA yields more stable rankings, maintains downstream performance, and reduces attribution cost.
By Jia Gu, Liang Pang, Huawei Shen
The paper introduces the task of predicting alignment generalization, which measures how fine‑tuning a language model to follow a specific value affects its behavior across a broad set of other values. Through a large‑scale study of 66 values, the authors show that representations derived from model activations when applying values in context outperform textual description‑based methods, achieving a correlation of 0.45 versus 0.05. They further demonstrate that these representations can predict downstream robustness and help construct a taxonomy of LLM values based on empirical generalization dynamics.
By Andy Liu, Mehar Bhatia, Karolina Stanczak, Mona Diab, Vered Shwartz, Daniel Fried
arXiv:2606. 04928v1 Announce Type: new Abstract: Large Language Models (LLMs) are increasingly deployed across diverse applications, raising critical questions for governance, accountability, and data provenance.
By Fr\'ed\'eric Berdoz, Luca A. Lanzend\"orfer, Kaan Bayraktar, Roger Wattenhofer
arXiv:2607. 03453v1 Announce Type: cross Abstract: Inference-time alignment methods, such as Best-of-$N$, offer a flexible alternative to training-based alignment by using reward models to select high-quality responses generated by a reference LLM.
By Eric Lei, Hsiang Hsu, Chun-Fu Chen
arXiv:2601. 22146v2 Announce Type: replace-cross Abstract: Due to limited supervised training data, large language models (LLMs) are typically pre-trained via a self-supervised "predict the next word" objective on a vast amount of unstructured text data.
By Ajay Patel, Colin Raffel, Chris Callison-Burch
The paper introduces the Controlled Attribution Benchmark (CATT), a synthetic evaluation framework that fine‑tunes language models on entirely fictitious knowledge to provide clear ground truth for data attribution. By systematically varying single dataset properties—such as document redundancy and lexical distractors—CATT reveals that these factors consistently degrade attribution performance. The study also explores two underexamined design choices: using negative log‑odds gradients instead of cross‑entropy gradients, and aggregating only the top token attribution scores, which together improve ground‑truth recall by 35–72 percentage points and enhance counterfactual retraining performance on FTRACE‑TREX by up to 4.9×.
arXiv:2610.09660v1 Announce Type: new
Abstract: Automatic Short Answer Scoring (ASAS) requires models that can score student responses against question-specific criteria while remaining efficient and...
By Zhifan Sun, Sebastian Gombert, Fabian Zehner, Leon Camus, Longwei Cong, Hendrik Drachsler
arXiv:2608. 01423v1 Announce Type: cross Abstract: Reference-based text evaluation metrics, which are widely used to assess natural language generation systems, score a candidate response by comparing it with a reference response.
By Shengwei Xu, Yuxuan Lu, Yifan Wu, Jason Hartline, Grant Schoenebeck
arXiv:2606. 11198v1 Announce Type: cross Abstract: Retrieval-augmented generation (RAG) systems inject external knowledge to improve LLM outputs, yet the format of injected content -- distinct from its semantic relevance -- can independently distort the model's attention distribution.
By Yuqi Zhang, Di Zhang
arXiv:2602.10657v2 Announce Type: replace
Abstract: How much of zero-shot benchmark accuracy is captured by word-level alignment between a pre-training corpus and the benchmark target? We study this...
By Jeonghoon Kim, Woojin Chung, Woomin Song, Cheonbok Park, Nako Sung, Jinwoo Shin
The paper investigates how preference tuning—optimizing language models with explicit preference signals—behaves when applied to new domains. It systematically compares five alignment objectives and several adaptation strategies, such as target‑domain supervised fine‑tuning and pseudo‑labeling, across summarization, question‑answering helpfulness, and safety tasks. Results show that while pseudo‑labeling reduces domain‑shift degradation, it also causes mode collapse, highlighting a trade‑off between generalization and diversity.
By Constantinos Karouzos, Xingwei Tan, Nikolaos Aletras
arXiv:2610.08026v1 Announce Type: new
Abstract: In recent years, several methods for detecting when large language models (LLMs) hallucinate have been developed. These methods are often benchmarked w...
By Jorma Valjakka, Juhani Kivim\"aki, Juha Myll\"ari, Jukka K. Nurminen