Do Understanding and Generation Fight? A Diagnostic Study of DPO for Unified Multimodal Models
Unified multimodal models share a language model backbone for both understanding and generating images. Can DPO align both capabilities simultaneously? We present the first systematic study of this question, applying DPO to Janus-Pro at 1B and 7B parameters under seven training strategies and two post-hoc methods. The central finding is negative: generation quality resists DPO alignment across all tested conditions on this architecture. No method improves generation CLIPScore at 7B (|Delta| 0.5
Record details
Published: 17 March 2026
Source: arXiv
Category: Research
Topics: Safety & alignment · Healthcare
Retrieved: 14 July 2026
Related evidence
These records share source-supplied organisations, an exact publisher byline, automatic topics or regions. The reason is shown on every link; related does not mean supporting, agreeing with or verifying this record.
Interpretable Cross-Domain Few-Shot Learning with Rectified Target-Domain Local Alignment
arXiv · 18 March 2026
Detection Is Cheap, Routing Is Learned: Why Refusal-Based Alignment Evaluation Fails
arXiv · 18 March 2026
To See or To Please: Uncovering Visual Sycophancy and Split Beliefs in VLMs
arXiv · 19 March 2026
Adoption and Effectiveness of AI-Based Anomaly Detection for Cross Provider Health Data Exchange
arXiv · 19 March 2026
Whether, Not Which: Mechanistic Interpretability Reveals Dissociable Affect Reception and Emotion Categorization in LLMs
arXiv · 15 March 2026
HiPath: Hierarchical Vision-Language Alignment for Structured Pathology Report Prediction
arXiv · 20 March 2026
How to cite this record
ethics.ai (17 March 2026), “Do Understanding and Generation Fight? A Diagnostic Study of DPO for Unified Multimodal Models,” evidence record 7097, https://ethics.ai/record/7097 (originally published by arXiv).
Use and limitations
This page is a stable index and citation surface for a source record. ethics.ai did not author the underlying report and has not independently verified every claim. Automatic topics may be imperfect. For consequential use, quote and cite the original publisher.