Large Models Return to Parameter Race: 2 Trillion Becomes Standard, Is 3 Trillion Next?

The article reports a renewed focus on parameter count as a key competitive metric in the development of large language models. Industry trends suggest that models with 2 trillion parameters will soon be common, and the next frontier is expected to be 3 trillion parameters. This competition drives significant advancements in AI capabilities but also raises concerns about computational costs and energy consumption. The piece underscores the importance of scale in achieving superior performance, while also hinting at potential limitations and the need for efficiency improvements.