Papers & Research

Transluce proposes PRBO method using RL agents to automatically discover rare pathological behaviors in Llama, Qwen, and DeepSeek

Original title:Surfacing Pathological Behaviors in Language Models

AI summary

Transluce released a research report proposing PRBO (propensity lower bound) as a proxy for sparse rewards, training an RL investigation agent to generate realistic and natural prompts that elicit specified rare behaviors in open-weight models, covering Llama 3.1/4, Qwen 2.5, and DeepSeek-V3.

Key Takeaways

  • Transluce released a research report proposing PRBO (propensity lower bound) as a proxy for sparse…
  • 1/4,Qwen 2.
  • 5,and DeepSeek-V3.

Quick Facts

Published
Updated
Source
Transluce(网页)
Category
Papers & Research
AI score
66
Read the original Read the full article at the original source
Why it matters

Transluce 提出用 PRBO 和 RL 调查智能体自动激发模型的罕见问题行为,并分析这些行为在真实提示下能否复现。

Related entities

Source & copyright

This article is based on curated titles and AI summaries from the AIHOT public API, with attribution and links back. Original copyright remains with the original publishers and authors; this site does not mirror full texts.

Original article · AIHOT entry

How to Cite

When citing this article, keep the title, date, and permalink, and credit Orange News as the source.

Orange News: Transluce proposes PRBO method using RL agents to automatically discover rare pathological behaviors in Llama, Qwen, and DeepSeek. June 5, 2025. https://new.nat6.net/en/news/cmtyo2tx503jqrog01dxgq3is/

View all →