Tencent Executive Responds to AI Lag: Insufficient Computing Power Has Slowed Down Metaverse

Watchtower Beating AI News Flash: Tang Daosheng, Senior Executive Vice President of Tencent and CEO of the Cloud and Smart Industries Group, posted a long article in response to the criticism that "Tencent is lagging behind in AI." He acknowledged internal anxiety within the company, acknowledging that competitors have been stronger in recent years. More pointedly, Tang admitted that Tencent's overall computing power is "severely insufficient," which has slowed down the training of multimodal models and product development, causing a "significant impact."

However, Tang believes that the current stage may only be the first kilometer of the AI marathon, stating that "endurance is more important than speed in the early stage." He teased that the next-generation multimodal model, Hy4, will have more breakthroughs, while also emphasizing that Tencent cannot solely rely on basic models. He indicated that algorithms determine the upper limit, engineering determines the speed of implementation, and real-world scenarios such as WeChat's office, documents, meetings, and enterprise services are Tencent's greatest strengths.

In a rare move, he reviewed several past mistakes. In the past year, Yuanbao invested heavily in promotion for user growth, but at that time, "both the model and the product were not yet ready," yielding unsatisfactory results. CodeBuddy and WorkBuddy were also not on a pre-planned path to success. In the early stages, DevTools incurred long-term losses and was almost discontinued.

WorkBuddy later caught the wave of agents and released over 40 versions in three months. Tang Daosheng stated that the team now extensively uses AI to write code, with humans primarily responsible for judgment, debugging, and quality control. He predicted that the boundaries between product development and testing would become increasingly blurred in the future, allowing small teams to accomplish what previously only large teams could do with the help of AI.