The KwaiYii LLM from Kuaishou, a top Chinese short-video platform operator, has “significantly surpassed GPT-3.5 in overall performance … and is approaching the levels of GPT-4 in certain metrics”, said Cheng Yixiao, company founder and CEO, during its post-earnings call on Wednesday.
GPT-4 is OpenAI’s most advanced LLM that is accessible by the public.
Kuaishou is confident that KwaiYii can match GPT-4 in the next six months, Cheng said. The company also said that its KeTu text-to-image model has outperformed Midjourney V5, the AI image-generating service launched by San Francisco-based Midjourney a year ago.
Meanwhile, Tencent said its Hunyuan LLM is now among the world’s best.
The social media and video gaming behemoth has been enhancing the ability of its AI models to create images and videos based on text prompts, said company president Martin Lau Chi-ping in a conference call with the media on Wednesday, calling the endeavour “the most important for us this year”.
The Shenzhen-based firm recently introduced an image-to-video model through a tie-up with academic partners, including the Hong Kong University of Science and Technology and Beijing’s Tsinghua University.
Named Follow-Your-Click, the tool is able to animate any part of a picture with a simple text prompt and mouse click.
Both Tencent and Kuaishou are looking for ways to apply cutting-edge AI technology to their businesses.
Tencent’s Lau said the Hunyuan model has been integrated into some of the firm’s enterprise products. The model is also being used by employees for daily tasks like coding and content generation, as well as improving advertisements to drive clicks.
Kuaishou, a major rival of ByteDance’s Douyin, has used KwaiYii to improve automated customer service in several of the company’s businesses, including its online talent and property sales platforms, according to CEO Cheng.
He also said the company is looking to use AI to develop generative models that can serve as creative tools for users to reduce the content creation barrier, improve video quality and raise production efficiency.
“The faster-than-expected launch of Sora … opens up immense opportunities for the short-video ecosystem,” Cheng said.
Eugen Boglaru is an AI aficionado covering the fascinating and rapidly advancing field of Artificial Intelligence. From machine learning breakthroughs to ethical considerations, Eugen provides readers with a deep dive into the world of AI, demystifying complex concepts and exploring the transformative impact of intelligent technologies.
