清华团队新发现：AI大模型看“块头”更要看“密度”

Core Insights - The development of AI has followed a "scale rule," where larger models exhibit stronger capabilities, but this has led to increased training and operational costs, hindering industrial application [1] - A new research from Tsinghua University introduces the "density rule," suggesting that AI model strength should focus on "ability density" rather than just size [1] - The research indicates that achieving high ability density requires advanced systems of "data + computing power + algorithms," rather than merely compressing large models [1] Group 1 - The "ability density" of AI models is growing exponentially, doubling approximately every 3.5 months [2] - This rapid growth implies that the size of the required computational resources for complex tasks is decreasing significantly over time [2] - Tsinghua University has collaborated with AI company Weimi Intelligence to develop a series of "high-density" models, which are already being applied in various consumer products [2] Group 2 - The future direction of AI models is towards being "refined" and "efficient," rather than simply larger [2] - The convergence of chip computing power and AI intelligence density may lead to the emergence of "edge intelligence," enhancing the capabilities of personal devices [2] - This advancement is expected to result in faster response times and improved privacy protection for users [2]