The KwaiYii LLM from Kuaishou, a prime Chinese language short-video platform operator, has “considerably surpassed GPT-3.5 in total efficiency … and is approaching the degrees of GPT-4 in sure metrics”, mentioned Cheng Yixiao, firm founder and CEO, throughout its post-earnings name on Wednesday.
GPT-4 is OpenAI’s most superior LLM that’s accessible by the general public.
Kuaishou is assured that KwaiYii can match GPT-4 within the subsequent six months, Cheng mentioned. The corporate additionally mentioned that its KeTu text-to-image mannequin has outperformed Midjourney V5, the AI image-generating service launched by San Francisco-based Midjourney a 12 months in the past.
In the meantime, Tencent mentioned its Hunyuan LLM is now among the many world’s greatest.
The social media and video gaming behemoth has been enhancing the flexibility of its AI fashions to create photographs and movies primarily based on textual content prompts, mentioned firm president Martin Lau Chi-ping in a convention name with the media on Wednesday, calling the endeavour “crucial for us this 12 months”.
The Shenzhen-based agency lately launched an image-to-video mannequin by a tie-up with educational companions, together with the Hong Kong College of Science and Expertise and Beijing’s Tsinghua College.
Named Observe-Your-Click on, the device is ready to animate any a part of an image with a easy textual content immediate and mouse click on.
Each Tencent and Kuaishou are searching for methods to use cutting-edge AI know-how to their companies.
Tencent’s Lau mentioned the Hunyuan mannequin has been built-in into a number of the agency’s enterprise merchandise. The mannequin can also be being utilized by workers for every day duties like coding and content material technology, in addition to enhancing ads to drive clicks.
Kuaishou, a serious rival of ByteDance’s Douyin, has used KwaiYii to enhance automated customer support in a number of of the corporate’s companies, together with its on-line expertise and property gross sales platforms, in response to CEO Cheng.
He additionally mentioned the corporate is wanting to make use of AI to develop generative fashions that may function artistic instruments for customers to scale back the content material creation barrier, enhance video high quality and lift manufacturing effectivity.
“The faster-than-expected launch of Sora … opens up immense alternatives for the short-video ecosystem,” Cheng mentioned.