Large Models Return to Parameter Race: 2 Trillion Becomes the Standard, Is 3 Trillion the Next Stop?
Chinese large models are re-entering a parameter race, with 2 trillion parameters becoming a common standard and 3 trillion seen as the next milestone. This trend is driven by competitive pressures and the belief that larger models lead to better performance, though it also raises concerns about computational costs and efficiency.
The article reports that the Chinese large model industry is again focusing on scaling up parameters, with 2 trillion parameters becoming the new baseline for top-tier models. Industry insiders suggest that the next target is 3 trillion parameters, reflecting a renewed emphasis on sheer model size as a key competitive differentiator. This shift comes after a period where efficiency and smaller models were prioritized, but now the race for scale is back on, driven by the assumption that larger models yield superior capabilities. However, this direction also brings significant challenges, including immense computing resource requirements, high training costs, and the need for more advanced optimization techniques. The trend is observed across major Chinese AI companies, which are investing heavily to keep pace in the global AI race. Experts caution that simply increasing parameters may not guarantee better performance, and more attention should be paid to data quality and algorithm innovation.
Source: Moonshot Kimi (GNews) —
original