Dooly

New research explores LLM security, efficiency, and training optimization

Researchers are developing novel methods to enhance the efficiency and security of Large Language Models (LLMs). One approach, "Widening the Gap," exploits outlier injection to compromise LLM quantization, demonstrating…