On March 20, Kuaishou released its fourth-quarter and full-year 2023 financial results. The data shows that Kuaishou has entered an era of comprehensive profitability: for the full year, both adjusted net income and net income for the period achieved a turnaround from loss to profit, with annual adjusted net income exceeding RMB 10.27 billion for the first time.
In the earnings conference call held that evening, regarding the research and application progress of Kuaishou's self-developed large model, Kuaishou Technology founder and CEO Cheng Yixiao responded that after the company launched its AI strategy in 2023, it has been steadily advancing the research and training of self-developed large models, and accelerating the application of large models in various scenarios. "Following the Kuaishou 13 billion and 66 billion models, in the fourth quarter, we focused on the research and training of a 175 billion-scale large language model, and achieved satisfactory confidence evaluation results."
"We are confident that within the next six months, the comprehensive performance of the large model will reach the level of GPT4.0," said Cheng Yixiao.
In terms of multimodality, Kuaishou's self-developed CLIP image large model - "KeTu", after several iterations, the model's comprehensive performance has surpassed the Midjourney V5 level. In the field of CLIP video, Kuaishou initiated a special research and development project at the end of last year. In this regard, Cheng Yixiao pointed out: "We believe this is a huge opportunity for the short video ecosystem. In the future, we will integrate generative models and producer tools to continuously help creators reduce the threshold of creation and improve the quality and efficiency of short video production."
In terms of using large models, Kuaishou continued to explore the deep integration of large models with Kuaishou APP scenarios, such as iterating on the detailed experience of applications like "AI Xiaokuaishou", "AI Play Review", and "AI Search Dialogue", and applying large model smart customer service assistants to scenarios like recruitment and real estate businesses to further improve the level of customer service. In the fourth quarter, Kuaishou also focused on promoting the application of large models in the live broadcasting of digital humans, relying on live broadcast scripts automatically generated by large models to empower digital humans with intelligent broadcasts and real-time interactive comments.
