据新智元报道，近期，麻省理工学院（MIT）FutureTech实验室的研究人员发表了一项关于大模型（Large Models，如LLM）能力增长速度的研究。研究结果显示，大模型的能力大约每8个月就会翻一倍，其增速远远超过了著名的摩尔定律。
摩尔定律是信息技术领域的一个重要观察现象，它指出，集成电路上的晶体管数量大约每两年翻一番，即硬件算力每两年翻倍。然而，根据MIT的研究，大模型的能力增长速度远远超过了这一硬件发展的速度。
研究人员指出，LLM的能力提升大部分来自于算力。随着人工智能技术的不断进步，大模型在语音识别、自然语言处理、图像处理等领域的表现已经超越了人类的能力。然而，这也意味着，随着时间的推移，未来我们将面临无法满足LLM所需要的算力的挑战。
这一研究结果对于人工智能领域的发展具有重要的启示意义。一方面，它揭示了人工智能技术发展的迅猛速度，让我们看到了人工智能在未来的巨大潜力。另一方面，它也提醒我们，必须寻找新的算力增长点，以满足未来人工智能技术的发展需求。
值得注意的是，尽管大模型能力增长速度远超摩尔定律，但这并不意味着摩尔定律已经失效。硬件算力的持续发展仍然是支撑人工智能技术进步的重要基础。然而，这一研究确实表明，人工智能技术的发展速度正在加快，我们需要跟上这一速度，不断探索新的技术和解决方案，以推动人工智能领域的持续发展。
未来，随着大模型能力的进一步提升，人工智能技术将在更多领域展现出更大的应用价值。而我们也应该关注和研究如何应对大模型发展带来的挑战，以实现人工智能技术的可持续发展。
英语如下：
**Title:** MIT Study: LLM Capability Doubling Far Outpaces Moore’s Law
**Keywords:** MIT Study, LLM Capability, Moore’s Law
**News Content:**
### MIT New Research: Large Model Capability Growth Far Outpaces Moore’s Law
According to New Intelligence, researchers from the MIT FutureTech Laboratory recently published a study on the growth rate of large model (such as LLM) capabilities. The research findings indicate that the capability of large models approximately doubles every 8 months, far outpacing the famous Moore’s Law.
Moore’s Law is an important observation in the field of information technology, which states that the number of transistors on integrated circuits doubles approximately every two years, leading to a doubling of hardware computing power every two years. However, the research by MIT shows that the growth rate of large model capabilities far exceeds this pace of hardware development.
The researchers point out that the enhancement of LLM capabilities mostly comes from computing power. With the continuous advancement of artificial intelligence technology, large models have already surpassed human capabilities in fields such as speech recognition, natural language processing, and image processing. However, this also means that as time goes by, we will face challenges in meeting the computing power required by LLMs.
The implications of this research result for the development of the artificial intelligence field are significant. On the one hand, it reveals the rapid pace of AI technology development and shows the huge potential of AI in the future. On the other hand, it reminds us that we must find new sources of computing power to meet the development needs of future AI technology.
It is worth noting that although the growth rate of large model capabilities far exceeds Moore’s Law, this does not mean that Moore’s Law has already failed. The continuous development of hardware computing power remains an important foundation supporting the progress of AI technology. However, this study does indicate that the pace of AI technology development is accelerating, and we need to keep up with this speed, constantly exploring new technologies and solutions to promote the sustained development of the AI field.
In the future, with the further enhancement of large model capabilities, AI technology will show even greater application value in more fields. And we should pay attention to and study how to deal with the challenges brought about by the development of large models, in order to achieve sustainable development of AI technology.
【来源】https://mp.weixin.qq.com/s/HLHrhOkHxRPRQ3ttJLsfWA
Views: 7