文章预览
LG - 机器学习 CV - 计算机视觉 CL - 计算与语言 1、[CL] LLMs Know More Than They Show: On the Intrinsic Representation of LLM Hallucinations 2、[LG] GSM-Symbolic:Understanding the Limitations of Mathematical Reasoning in Large Language Models 3、[LG] Strong Model Collapse 4、[LG] Improving LLM Reasoning through Scaling Inference Computation with Collaborative Verification 5、[CL] Round and Round We Go! What makes Rotary Positional Encodings useful? 摘要:LLM幻觉内在表征研究、大语言模型数学推理的局限性、强模型崩溃研究、通过协作验证扩展推理计算改进LLM推理、旋转位置编码为何有用 1、[CL] LLMs Know More Than They Show: On the Intrinsic Representation of LLM Hallucinations H Orgad, M Toker, Z Gekhman, R Reichart… [Technion] LLM知道的比他们表现出来的多:LLM幻觉内在表征研究 要点: 幻觉定位: 大型语言模型 (LLM) 中关于真实性的信息并非均匀分布,而是集中在特定
………………………………