“人工智能寒冬”
用英语怎么说？

Benchmarks function as the ultimate rulers of artificial intelligence, serving both as measuring sticks and as forces that dominate the field. In 1987, a benchmark for automatic speech translation revealed that state-of-the-art models could not outperform simple statistical methods, which ushered in a decades-long AI winter where interest waned and research funding dried up. The pendulum swung back in 2012 when the ImageNet competition demonstrated the overwhelming power of deep neural networks. Today, tests for large language models are created and solved with clockwork regularity, yet computer scientist Moritz Hardt argues the entire evaluation paradigm is fundamentally broken.

基准测试充当着人工智能领域的终极统治者，既是衡量标尺，也是主导该领域的力量。1987年，一项自动语音翻译基准测试显示，最先进的模型无法胜过简单的统计方法，这开启了长达数十年的人工智能寒冬，期间人们的兴趣减退，科研经费枯竭。钟摆在2012年荡了回来，当时ImageNet竞赛展示了深度神经网络的压倒性实力。如今，针对大语言模型的测试以极有规律的节奏被创建并攻克，然而计算机科学家莫里茨·哈特认为，整个评估范式从根本上已经崩溃。