尽管人们担忧人工智慧的高能耗可能阻碍巴黎协定的目标，但其加速电池技术开发的潜力提供了重大机遇。传统的大语言模型（LLM）主要利用神经网络作为庞大的自动完成系统运行，但缺乏人类般的理解力或认识论。经测试，所有六个主要 LLM 均能正确识别一家公司是否拥有净零排放目标。然而，当被问及该公司特定的 2025 年 Scope 1 温室气体排放值时，只有不到 20%（五分之一）的模型提供了正确答案，而有 66.7%（三分之二）的模型返回了错误的数值，从而浪费了大量的计算能源。

LLM 在处理特定排放数据时的关键局限性，源于与通用查询相比，专业查询可用的文本语料库较小。虽然主要的 AI 超大型企业试图以可再生能源为其运营提供动力，但大多数企业似乎仍依赖平均电网组合。尽管如此，Scope 2 排放量（来自购买的能源）仍然相对微小。AI 的真正价值不在于语言生成，而在于其快速执行数百万次二元筛选决策的能力，这对于解决巴黎协定下电气化所需的最大瓶颈至关重要。

这种筛选能力正在变革材料发现领域，其中数千万种潜在的电池材料仍未被探索。在过去三年中，超大型企业与实验室的合作已将材料发现速度提高了 100 倍。Google DeepMind 的科学家 John Jumper 和 Sir Demis Hassabis 因开发预测蛋白质复杂结构的 AI 模型而荣获 2024 年诺贝尔化学奖的 50%（一半），这正是该 AI 应用的典范。电池技术若能在 2050 年目标年之前取得类似的突破，将能显著推进绿色能源转型。

Despite concerns over the high energy consumption of AI, which is often viewed as a barrier to the Paris agreement, its potential to accelerate battery technology development presents a significant opportunity. Traditional large language models (LLMs) function primarily as massive auto-complete systems using neural networks, yet they lack human-like understanding or epistemology. When tested, all six major LLMs correctly identified if a company had a net zero target. However, when asked for a specific 2025 Scope 1 greenhouse gas emissions value, less than 20% (one in five) provided the correct answer, while 66.7% (two-thirds) returned incorrect values, wasting substantial computational energy.

The critical limitation of LLMs in processing specific emissions data stems from the smaller corpus of available text for specialized queries compared to general ones. While major AI companies attempt to power their operations with renewable energy, the majority still rely on the average grid mix. Nevertheless, Scope 2 emissions (from purchased energy) remain relatively small. The true value of AI lies not in language generation, but in its ability to execute millions of binary screening decisions rapidly, which is essential for addressing the primary bottleneck of electrification under the Paris agreement.

This screening capability is transforming materials discovery, where tens of millions of potential battery materials remain unexplored. Over the past three years, collaborations between hyperscaling companies and laboratories have accelerated materials discovery by 100 times. This application of AI is exemplified by Google DeepMind scientists John Jumper and Sir Demis Hassabis, who won 50% (half) of the 2024 Nobel Prize in Chemistry for predicting protein structures. A similar breakthrough in battery technology could significantly advance the green energy transition ahead of the 2050 target year.