What are the ways to implement artificial intelligence? What are the current status and trends of AI development?

From August 25 to 27, 2023, the 19th Yabuli Chinese Entrepreneurs Summer Summit 2023 with the theme of "Focus on the Greater Bay Area, Innovation to Promote Development - Working Together to Build Chinese-style Modernization" was held in Shenzhen. Dr. Chen Ning, Chairman and CEO of Yuntian Lifei, was invited to attend.

In a dialogue themed “What the World Needs”, Dr. Chen Ning discussed with industry experts on the same stage.

Talking about big models: application and cost are the focus of the next stage

This year, the industry entered the "Hundred Model War". Yuntian Lifei also launched the "Tianshu" large model and applied it in government affairs and other fields. Dr. Chen Ning mentioned in the dialogue that the large model has gained a lot of popularity in the past six months, and the industry will return to rational thinking more and more in the future. In the future, the large model will also solve the problems of application and cost. The large model may be equivalent to a full-time undergraduate graduate today, but how to apply it in different industries still needs to be combined with industry knowledge and truly promoted through the industry large model.

Talk: Inference chips deserve as much attention as training chips

Chips can be simply divided into two categories: one is training chips, and the other is inference chips. Today, almost all of our focus is on training chips, or training-based chips. But for AI, training is just the beginning, inference and application are the ultimate goals, and the algorithms of large models produced must eventually be put into practical use.

, automobiles, and all kinds of things, the intelligent connection of all things needs inference chips. In the future, the market demand for inference chips may be greater than that for training chips.

Therefore, InnoLife has always insisted on "algorithm chipization" to enable chips to meet the needs of different scenarios. At the end of last year, InnoLife's self-developed AI chip DeepEdge10 was successfully taped out, efficiently supporting matrix multiplication operations in the Transfmer model and can be applied to the field of edge CV large model reasoning.

Talking about innovation: AI development is inseparable from three forces

When the host asked about his views on OpenAI's success, Dr. Chen Ning believed that it was inseparable from three forces - the power of ecology, the power of curiosity, and the power of agility.

He pointed out that although ChatGPT was created by OpenAI, OpenAI stands on the shoulders of "giants" such as Google and Microsoft, and behind it is the support of an innovative ecosystem.

While everyone is considering how to industrialize artificial intelligence, OpenAI is more purely thinking about how to make a large model that can incorporate all of humanity's knowledge system to see if it can be as smart as humans. Curiosity about the unknown drives them to persist day after day.

The third "secret" is that it is "small" and can only be broken by speed. It has no baggage, can continuously innovate, develop agilely, communicate efficiently, and operate quickly... "In fact, this is also a challenge to the flexibility of all our companies today: With the rapid technological changes, can we quickly embrace AI? You will find that there have been a lot of disruptive innovations in the past two years, and I believe that more and more technological innovations will emerge in the future."

Talking about the future: Everyone has three machines

When talking about his expectations for the future development of AI, Dr. Chen Ning believes that under the changes of AI, the way humans live and work will undergo earth-shaking changes. In the future, perhaps everyone will own three machines: a family service robot, an unmanned vehicle for commuting, and an office assistant for use in work scenarios.

To realize this ideal scenario, the support of chips and artificial intelligence algorithms is indispensable. Yuntianlifei will also work hard to realize China's general artificial intelligence.

Reviewing Editor: Peng Jing