Saved in:
Bibliographic Details
Main Authors: Brazilek, Jasmine, Tidmarsh, Miles
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2604.13076
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866910184651096064
author Brazilek, Jasmine
Tidmarsh, Miles
author_facet Brazilek, Jasmine
Tidmarsh, Miles
contents We investigate the robustness of value alignment via midtraining with synthetic documents, using animal compassion as a value that is both important in its own right and orthogonal to existing alignment efforts. To evaluate compassionate reasoning, we develop and publicly release Animal Norms In Moral Assessment (ANIMA), a 26-question evaluation spanning 13 ethical dimensions, publicly available as a dataset and Inspect evaluation. On ANIMA, training with 3000 documents achieves 77% compared to 40% for instruction-tuning approaches, with generalization to human compassion and no degradation in standard safety benchmarks or capabilities. However, subsequent unrelated instruction-tuning degrades the intervention, with the advantage disappearing after 5000 samples. Our exploratory results suggest document-based value interventions may require explicit preservation strategies to remain effective through typical training pipelines.
format Preprint
id arxiv_https___arxiv_org_abs_2604_13076
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Alignment midtraining for animals
Brazilek, Jasmine
Tidmarsh, Miles
Computation and Language
Artificial Intelligence
We investigate the robustness of value alignment via midtraining with synthetic documents, using animal compassion as a value that is both important in its own right and orthogonal to existing alignment efforts. To evaluate compassionate reasoning, we develop and publicly release Animal Norms In Moral Assessment (ANIMA), a 26-question evaluation spanning 13 ethical dimensions, publicly available as a dataset and Inspect evaluation. On ANIMA, training with 3000 documents achieves 77% compared to 40% for instruction-tuning approaches, with generalization to human compassion and no degradation in standard safety benchmarks or capabilities. However, subsequent unrelated instruction-tuning degrades the intervention, with the advantage disappearing after 5000 samples. Our exploratory results suggest document-based value interventions may require explicit preservation strategies to remain effective through typical training pipelines.
title Alignment midtraining for animals
topic Computation and Language
Artificial Intelligence
url https://arxiv.org/abs/2604.13076