On May 16, TapTechNews reported that a picture circulating online showed that during a Huawei event on May 10 showcasing the ability of large-scale text and image generation, the demonstration was interrupted due to pressing Ctrl-C during runtime, revealing a snippet of code 'time.sleep(6)’.
Some pointed out that this code implies 'manually controlling the process from input to output, taking 6 seconds', meaning 'waiting 6 seconds to output a local image', leading to speculation that the text and image result may not be generated by large-scale models. This topic immediately sparked discussions on authenticity and technical capabilities.
In response to the online allegations of faking the demonstration during the event, Huawei issued an official statement on the Ascend Community. The translated version by TapTechNews is as follows:
On May 10, during a technical discussion at the Kunpeng Ascend Developer Conference, we demonstrated the mxRAGSDK functionality. This demonstration was aimed at developers, showing how to develop RAG applications in just a few lines of code.
The real-time generated images shown on site were produced by calling an open-source large model. The code including 'time.sleep(6)' signifies a command to wait for real-time image generation from an external open-source large model, not fetching preset images.
All demonstrations shown are real code and will be made available on the Ascend Community for developers to use and provide valuable feedback.
Some opinions suggest that Huawei's use of 'time.sleep(6)' at a high-tech event is deemed as falsification, accusing Huawei of simulating AI computation processes, when in fact, they were showcasing pre-prepared results.
Technical experts have differing opinions on this matter. Some believe 'time.sleep(6)' may have been used to ensure a smooth and controlled demonstration without any unforeseen incidents, serving as a predefined buffer step. Such practice is common in technical demonstrations to showcase a smooth and controllable process.
Other experts point out that the computation and generation processes of large models indeed take time, and 'time.sleep(6)' might have been used to demonstrate the actual processing time of the model. In the field of AI, the speed of model computation is a crucial indicator of technological advancement.