爱可可 AI 前沿推介(7.21)

文章预览

LG - 机器学习 CV - 计算机视觉 CL - 计算与语言 1、[LG] Prover-Verifier Games Improve Legibility of Llm Outputs 2、[CL] Large Models of What? Mistaking Engineering Achievements for Human Linguistic Agency 3、[LG] Neural Optimal Transport with Lagrangian Costs 4、[LG] Context-Guided Diffusion for Out-of-Distribution Molecular and Protein Design 5、[LG] xLSTMTime：Long-term Time Series Forecasting With xLSTM 摘要：用证明器-验证器博弈提高LLM输出的可读性、将工程成就误认为对人类语言完全表示的反思、基于拉格朗日代价的神经最优传输、面向分布外分子和蛋白质设计的上下文引导扩散、用xLSTM进行长时间序列预测 1、[LG] Prover-Verifier Games Improve Legibility of Llm Outputs J H Kirchner, Y Chen, H Edwards, J Leike... [OpenAI] 用证明器-验证器博弈提高LLM输出的可读性 要点: 仅为求解的正确性而优化 LLM 会导致难以理解的求解，时间有限的人类很难对其进行 ………………………………