Tencent Mocked for Being Slow in AI, the Answer is Disgusting

JAKARTA - Tencent Holdings has dismissed criticism that the company is lagging behind in the artificial intelligence or AI race. The Chinese technology giant, owner of WeChat, chose a calm, but sharp answer. AI is not a race of speed.

Quoted from a report by Yicai Global, Saturday, June 6, Tencent denied the assumption that the company was late in developing AI. "AI is a long-term game," said Yao Shunyu, Head of AI Scientist at Tencent, at the Tencent Cloud AI Industry Application Conference, June 5.

The statement comes as the global tech industry is drunk on AI. Many companies are racing to showcase the smartest models, highest scores, and fastest products to rival ChatGPT. Tencent took a different position. For the company, AI victory is not measured by the noise of the stage and the rankings.

"In some ways, the second half of the AI race has just begun," said Yao, a former OpenAI researcher. "ChatGPT and Claude will not be the only super apps. New opportunities will emerge."

Yao said China needs to build a special organization to develop Artificial General Intelligence or AGI. This term refers to AI that is able to match, even surpass human ability in almost all fields.

According to Yao, such an organization must be supported by three things: a strong technological foundation, a truly useful product, and the courage to explore new territories.

Yao also touched on the growing unease in Silicon Valley. One of them, the view that in two years many people will lose their jobs because they are replaced by AI.

"That's one view," he said. "But our view is, AI is a long-term game."

Yao's current important task at Tencent is to develop Hy3, the company's large language model based in Shenzhen. A large language model is an AI system trained to understand and generate text, code, and other responses based on very large amounts of data.

Yao said Hy3 has been improved in three main areas.

First, Tencent rebuilt its infrastructure, including the architecture for pre-training and reinforcement learning. Simply put, this concerns the way models are trained from the start and the way they are improved through feedback.

Second, Tencent strengthened the data and evaluation system. The focus is not only on increasing data, but ensuring that the data is of higher quality and closer to real problems.

Third, Hy3 is claimed to be able to make decisions that are more similar to humans based on "taste" or subjective considerations. For example, in recruitment, setting the rhythm of model development, and weighing various difficult choices.

According to Yicai Global's report, Yao also highlighted the habit of China's AI industry that is too fixated on rankings. According to him, good AI is not enough to just win on the scoreboard. More important is its practical benefits.

In other words, it's useless to be cool on the leaderboard if you're confused about what to use it for.

At the same event, Tencent Senior Executive Vice President Tang Daosheng said most of the company's programming code this year has been generated by AI. As a result, engineers can take care of more of the system architecture design. The work of writing large amounts of code is handed over to AI, while humans still provide direction, correction, and control.

Tencent's stance is interesting because it comes from a company that hasn't always been the loudest in the global AI war. Instead of chasing quick applause, Tencent wants to emphasize one thing: AI must be useful, not just look smart.

In the midst of the AI fever, the message feels relevant. This technology is not enough to be a presentation material, business jargon, or a strategy display. The ultimate test is whether it can be used or not. Save work or not. Help humans or just add new terms in the meeting room.

Because in technology, as in life, the busiest are not necessarily the most prepared.