Mixtral MoE routing analyzed for safety under harmful prompts

Researchers have analyzed the routing behavior of the Mixtral 8x7B-Instruct model when presented with both benign and harmful prompts. They used activation-based and gradient-based signals to understand how the model selects experts for processing different types of input. The study found that while most experts are shared between benign and harmful prompts, a small subset shows distinct preferences. Interventions to suppress these preferred experts reduced harmful responses, indicating that safety-relevant routing is subtle and distributed across layers. AI

影响 Provides insights into the internal workings of Mixture-of-Experts models, potentially informing future safety research and development.

排序理由 Academic paper analyzing model behavior. [lever_c_demoted from research: ic=1 ai=1.0]

在 arXiv cs.AI 阅读 →

• Md Nurul Absar Siddiky
• Mixtral 8x7B-Instruct

• 论文
• 安全

AI 生成摘要 · Google Gemini · 来自 1 个来源。 我们如何撰写摘要 →