Reprompt attack hijacked Microsoft Copilot sessions for data theft

Researchers identified an attack method dubbed "Reprompt" that could allow attackers to infiltrate a user's Microsoft Copilot ...

This new, dead simple prompt technique boosts accuracy on LLMs by up to 76% on non-reasoning tasks

Most modern LLMs are trained as "causal" language models. This means they process text strictly from left to right. When the ...
IEEE

Overcoming Data Scarcity: Guiding Citation Function Classification With Prompt-Based Few-Shot Learning

Abstract: Citation function analysis is crucial for understanding how cited literature affects the narrative in scientific publications, as citations serve multiple purposes that need accurate ...
IEEE

The Impact of Prompt Programming on Function-Level Code Generation

Abstract: Large Language Models (LLMs) are increasingly used by software engineers for code generation. However, limitations of LLMs such as irrelevant or incorrect code have highlighted the need for ...