Tencent PR Chief Warns "Skill" Tinkering Won't Produce True AI Thinking

The comment

It has been reported that Zhang Jun (张军), public relations director at Tencent (腾讯), reposted and endorsed an essay on the limits of so‑called AI “skill” learning, arguing that recent enthusiasm for “炼化skill” is largely cosmetic. Zhang said these skills—translated in Chinese as “技能”—are essentially preset instruction templates that teach models to follow specific prompts, not to internalize deeper cognitive frameworks. He used the Chinese saying “画人画皮难画骨” — you can paint a person’s skin but not their bones — to stress that surface mimicry does not equal genuine understanding.

What the industry is doing

The discussion reflects a broader trend in China’s AI sector: teams tune large language models with many examples or instructions to boost task‑specific performance. Reportedly, this “collect‑more‑skills” approach has a gamified appeal—users and developers alike chase ever more capabilities like levels in a game. But experts inside and outside companies warn that optimizing for output form can leave models brittle when faced with tasks requiring generalized reasoning or novel inference.

Why it matters

Why should Western readers care? Because the debate touches on how next‑generation AI will be built and regulated amid the US‑China technology competition and growing calls for safety standards. If the dominant business model emphasizes packaged, monetizable “skills” over foundational reasoning, what does that mean for reliability, transparency and long‑term risk? Tencent’s comment is a reminder that technical progress and productization are not the same as cognitive progress — and that industry incentives will shape which path prevails.