爱可可 AI 前沿推介(11.8)

文章预览

LG - 机器学习 CV - 计算机视觉 CL - 计算与语言 1、[LG] How Transformers Solve Propositional Logic Problems：A Mechanistic Analysis 2、[LG] Non-Stationary Learning of Neural Networks with Automatic Soft Parameter Reset 3、[LG] LASER：Attention with Exponential Transformation 4、[LG] What Features in Prompts Jailbreak LLMs? Investigating the Mechanisms Behind Attacks 5、[LG] Long Context RAG Performance of Large Language Models 摘要：Transformer如何解决命题逻辑问题、基于自动软参数重置的神经网络非平稳学习、基于指数变换的注意力、大语言模型的越狱提示有哪些特征、大型语言模型的长上下文RAG性能 1、[LG] How Transformers Solve Propositional Logic Problems: A Mechanistic Analysis G Z Hong, N Dikkala, E Luo, C Rashtchian… [Purdue University & Google Research] Transformer如何解决命题逻辑问题：机制分析 要点: 合成命题逻辑问题:  文章提出一个合成命题逻辑问题作为测 ………………………………