Huawei demonstrated its large model’s text-to-image capability at a press conference on May 10, where an interruption occurred during the demo when Ctrl-C was pressed, revealing a piece of code that included “time.sleep(6).” Some netizens suspected that the generated images were not produced by the model but were manually output as local images after waiting for 6 seconds. This led to widespread discussion about the authenticity of the demonstration on the Chinese social platform Weibo. In response to this incident, Huawei stated on Thursday that the images were generated in real time using an open-source large model. The “time.sleep(6)” code indicates a command to wait for the external open-source model to generate the images in real time, rather than retrieving preset images, according to Huawei. The code used in the demonstration is authentic and will be available on the Huawei Ascend community platform, where developers are invited to use it and provide valuable feedback, the company said. [TechWeb, in Chinese]
Related
