Mahabubur Rahman Miraj
Titles as the sources gave them. In 1 entry, with 0 checked quotations.
In the Base
Every entry they appear in.
META-IR: Meta-Introspective Reasoning for Adaptive Safety Alignment in LLM
A preprint proposes META-IR, a meta-learning framework in which a model analyses risk and corrects itself, aimed at making refusals hold up against new jailbreak strategies. According to the abstract, it is tested on Llama-3.1-8B-Instruct and Qwen-2-7B-Instruct.