An internal statement related to Tencent's Hunyuan team has recently plucked once again the most sensitive nerve in domestic large-model competition. The "out-working DeepSeek through overtime" remark put forward by 28-year-old chief AI scientist Yao Shunyu exploded across public opinion, because what it struck at is not just one company, but the deep pain point that the entire industry cannot avoid in the AGI long run: what exactly does it take to catch up with rivals.
Zooming back to Tencent's own situation, the anxiety is real. On the AI track, its voice is being caught in a pincer attack from ByteDance's Doubao and DeepSeek, and its hold on developers' minds is being chipped away bit by bit by rivals. It urgently needs to prove on AGI underlying infrastructure that it is not a laggard, and this young scientist, on whom great hopes were pinned, was originally expected to break through Tencent's internally relatively rigid R&D system.
But the core of the controversy falls precisely on the methodology behind the word "against." The model DeepSeek set was never built by piling up overtime—it relied on algorithmic innovation and extreme engineering efficiency, leveraging technological breakthroughs at extremely low compute costs. Reducing the rival's success to "the rival is slacking, so we must put in more work hours" is in itself a misreading; DeepSeek's comeback precisely confirms that in the AI era, piercing through technical bottlenecks relies on high-density intellectual innovation, not human-wave tactics and accumulated time.
This also exposes the fatigue in the path that traditional big tech has long depended on. When the rival speaks through algorithmic efficiency, the playbook of relying on headcount and pointless rat-racing instead exposes the loss of any first-move advantage in technical direction and organizational efficiency. For Yao Shunyu, his real hard battle is not external but internal to Tencent: how to coordinate resources, how to defuse resistance from veteran employees, and how to break out of the big-tech habit of short-term KPI orientation. If things continue along the "grind overtime" line of thinking, it is nothing more than using tactical diligence to paper over strategic anxiety.
Technological evolution in the large-model era inherently carries strong non-linear characteristics—a single key architectural insight can be worth thousands of person-months of repetitive labor. What should truly be compared is talent density, the leverage of intelligence, not the thickness of work hours. For giants to truly break through in AI, binding employees to overtime is the easiest yet most ineffective answer; more fundamentally, what is needed is to complete a profound self-revolution of organizational form, building a flat, efficient R&D system that respects hacker spirit, and using high-level technical foresight and pure underlying innovation to move ahead of the industry.