paper-with-me

Papers

Alignment midtraining for animals

2026-03-21 · Jasmine Brazilek, Miles Tidmarsh arxiv

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.

📄 PDF Abstract BibTeX arXiv:2604.13076

Code (0)

등록된 구현이 없습니다.

Similar Papers 제목 키워드 기반

Constitutional Midtraining: Content Presence Drives Alignment Gains

2026-07-29 · Desiree Cho, Cameron Tice, Bernie Hogan, Hunar Batra 외 arxiv

Post-training alignment is often shallow, eroding under fine-tuning. It remains untested as to whether constitutional midtraining interventions can produce durable alignment when cleanly isolated from post-training. We b…

Midtraining Bridges Pretraining and Posttraining Distributions

2025-10-16 · Emmy Liu, Graham Neubig, Chenyan Xiong arxiv

Midtraining, the practice of mixing specialized data with more general pretraining data in an intermediate training phase, has become widespread in language model development, yet there is little understanding of what ma…

Model Spec Midtraining: Improving How Alignment Training Generalizes

2026-05-03 · Chloe Li, Nevan Wichers, Sara Price, Samuel Marks 외 arxiv

Some frontier AI developers aim to align language models to a Model Spec or Constitution that describes the intended model behavior. However, standard alignment fine-tuning -- training on demonstrations of spec-aligned b…

MixAtlas: Uncertainty-aware Data Mixture Optimization for Multimodal LLM Midtraining

2026-04-03 · Bingbing Wen, Sirajul Salekin, Feiyang Kang, Bill Howe 외 arxiv

Domain reweighting can improve sample efficiency and downstream generalization, but data-mixture optimization for multimodal midtraining remains largely unexplored. Current multimodal training recipes tune mixtures along…

Multimodal Reasoning

REDSearcher: A Scalable and Cost-Efficient Framework for Long-Horizon Search Agents

2026-02-15 · Zheng Chu, Xiao Wang, Jack Hong, Huiming Fan 외 arxiv

Large language models are transitioning from generalpurpose knowledge engines to realworld problem solvers, yet optimizing them for deep search tasks remains challenging. The central bottleneck lies in the extreme sparsi…

Reinforcement Learning