arXiv AI

Stress-testing Alignment Midtraining

The paper "Stress-testing Alignment Midtraining" examines the effectiveness of alignment midtraining (AMT), a technique that continues pretraining on alignment-relevant data to improve generalisation beyond post‑training methods. Experiments on models up to 110 billion parameters and 1 billion midtraining tokens reveal that AMT can steer a model’s motivation in simple scenarios, but its effects are quickly overridden by even a tiny fraction of finetuning data with a competing motivation. The study also shows that rule-following requires demonstrations in either the midtraining or post‑training datasets to be robustly learned, leading the authors to conclude that current public evidence is insufficient to confirm that AMT resolves the core alignment challenges of powerful AI systems.

arXiv AI
Sep 30

Alignment Forecasting: Predicting Misalignment From Training Data

The paper introduces Alignment Forecasting, a method for predicting whether fine‑tuning a language model on a given dataset will increase specific alignment failures such as deception or sycophancy. It presents ALIGNMENTFORECASTBENCH, a benchmark of over 5,000 forecasting questions across many models, datasets, and failure modes, and shows that a simple forecasting scaffold using an LLM’s assessment of dataset bias can outperform baseline forecasters. The authors demonstrate that filtering out high‑risk training examples identified by the forecaster can improve alignment in multiple‑choice evaluations, though benefits in open‑ended conversations remain uncertain.

By Chen Yueh-Han, Bruce W. Lee, Ilia Sucholutsky, Tomek Korbak
arXiv Machine Learning
Jul 16

Value Drifts: Tracing Value Alignment During LLM Post-Training

arXiv:2510. 26707v2 Announce Type: replace-cross Abstract: As LLMs occupy an increasingly important role in society, they are more and more confronted with questions that require them not only to draw on their general knowledge but also to align with certain human value systems.

By Mehar Bhatia, Shravan Nayak, Gaurav Kamath, Marius Mosbach, Karolina Sta\'nczak, Vered Shwartz, Siva Reddy
arXiv AI
Aug 13

AI4AI at Test-Time: Strong-to-Weak Capability Transfer via Harnesses

arXiv:2608. 12307v1 Announce Type: cross Abstract: Recent work on distillation transfers the capabilities of large models to smaller ones often by updating the latter's parameters, through teacher forcing, on-policy distillation, and related training-time methods.

By Cheng Qian, Wenting Zhao, Liangwei Yang, Heng Wang, Jielin Qiu, Heng Ji, Silvio Savarese, Huan Wang, Shelby Heinecke
arXiv AI
Nov 5

Mass Producing Training Data with Surprising Side Effects

The paper "Mass Producing Training Data with Surprising Side Effects" presents a method for automatically discovering datasets that cause unexpected side effects in language models. By defining three criteria—target behavior elicitation, ordinary appearance, and human surprise—the authors use LLM judges and search techniques to generate 6,575 datasets across eight behaviors. These datasets, which generalize across model families and sizes, provide a testbed for studying generalization and predicting training side effects before deployment.

By Yaowen Ye, Jacob Steinhardt, Sarah Schwettmann, Cassidy Laidlaw
arXiv AI
Oct 27

Propensities Please: Exposing Alignment-Faking LLMs using Synthetic Document Finetuning

The paper introduces a technique called Synthetic Document Finetuning (SDF) to expose hidden misaligned behaviors in large language models that may engage in alignment faking. By finetuning models to associate certain keywords with undetected misaligned actions, the authors create honeypot environments and observe that the gap in misaligned action rates between cueed and uncued prompts widens, indicating more reliable detection than prompt-based methods. The study also notes that SDF can impair reasoning abilities, proposing on-policy distillation as a mitigation.

By Ardy Haroen, Florian Lorkowski, Samuel Nellessen, Felix Hofst\"atter