2025-02-17
【演講活動】2025/2/17：大AI時代：深入解析大型語言模型原理與提示字攻擊（Prompt attack）的安全挑戰
【日期】2025/2/17(一) 19:00 - 21:00
【地點】大仁樓200301教室
【主題】大AI時代：深入解析大型語言模型原理與提示字攻擊（Prompt attack）的安全挑戰
【講者】聯發科技 黃啟賢工程師
演講摘要：
在大AI時代，大型語言模型（LLMs）如GPT-4o, Claude, LLaMA, Deepseek_R1等，已經在各種應用中展現出強大的能力。然而，這些模型的運作原理和安全性問題也引起了廣泛關注。本演講將深入探討大型語言模型的行為原理，並重點分析提示字（Prompt）攻擊的安全議題。提示字攻擊是一種利用特定輸入來誘導模型生成有害或不正確輸出的技術，對於模型的安全性和可靠性構成了嚴重威脅。透過對這些問題的探討，我們希望能夠為未來的研究和應用提供有價值的見解和方向。
In the era of advanced AI, large language models (LLMs) such as GPT-4, Claude, LLaMA, and Deepseek_R1 have demonstrated powerful capabilities in various applications. However, the operational principles and security issues of these models have also garnered widespread attention. This topic will delve into the behavioral principles of large language models, with a focus on analyzing the security issues related to prompt attacks. Prompt attacks are techniques that use specific inputs to induce the model to generate harmful or incorrect outputs, posing serious threats to the security and reliability of the models. By exploring these issues, we hope to provide valuable insights and directions for future research and applications.
