Alignment + Security + Fine Tuning

10.0

Create an open-source library for synthetic bootstrap training of MLLMs specifically for multi-turn safety. Current safety datasets are single-turn; developers need tools to generate long-context adversarial dialogues.

emergingimplementation gap

fine-tuningdataalignmentmultimodalsecurity

Signals (3)

arXiv5h ago

SaFeR-Steer: Evolving Multi-Turn MLLMs via Synthetic Bootstrapping and Feedback Dynamics

arXiv1d ago

C-Mining: Unsupervised Discovery of Seeds for Cultural Data Synthesis via Geometric Misalignment

arXiv5h ago

Alignment + Security + Fine Tuning

Signals (3)

SaFeR-Steer: Evolving Multi-Turn MLLMs via Synthetic Bootstrapping and Feedback Dynamics

C-Mining: Unsupervised Discovery of Seeds for Cultural Data Synthesis via Geometric Misalignment

Shifting the Gradient: Understanding How Defensive Training Methods Protect Language Model Integrity

Alignment + Security + Fine Tuning

Signals (3)

SaFeR-Steer: Evolving Multi-Turn MLLMs via Synthetic Bootstrapping and Feedback Dynamics

C-Mining: Unsupervised Discovery of Seeds for Cultural Data Synthesis via Geometric Misalignment

Shifting the Gradient: Understanding How Defensive Training Methods Protect Language Model Integrity