After OpenAI launched Sora, it immediately exploded in the film and television industry, advertising industry and game industry, but because of its limitations as a “simulator” in the simulation world, many people did not expect to touch intelligent manufacturing.

IS420ESWAH5A However, some intelligent manufacturing experts have observed and studied Sora, saw the similarities of the underlying logic, and also thought of some points of integration. “If the industry really moves in the direction of smart manufacturing, a series of difficult problems in black light factories, autonomous driving, digital twins can hopefully be solved.” Several smart manufacturing experts said.

An interesting phenomenon is that since the launch of the ChatGPT large model at the end of 2022, technologists, large model manufacturers, and entrepreneurs have spoken out more, but manufacturing companies and industry people have rarely spoken out. “There’s a little bit of overconsumption, there’s an explosion of technology coming out every week, but where do they land? What business value is generated?” A manufacturing source said. Some manufacturing companies feel anxious, behind the reason is from technology to application, there are still some gaps in the middle to bridge.

However, after the exploration and attempts of the past half year, in 2024, with the development of new quality productivity by the government and the acceleration of the action of some central state-owned enterprises in the application of large models, the process of applying large model innovation is accelerating in the intelligent manufacturing industry, especially in those industries with fast changes and fierce competition.

01

Can Sora be used in manufacturing?

After Sora was launched, Zhang Jun, product director of OPPO AI Center, was surprised by the speed of iteration of generative AI, “almost every morning when I wake up, I will see an interesting new demo.” In the future, how users call these generative AI may bring changes to the architecture, interaction and other aspects of the phone. OPPO is also exploring it internally.

IS420ESWAH5A In terms of terminal changes, the industry has some clear directions. In a short period of time, for example, two big changes may occur. One is de-appization. The large model itself has ecological plug-ins and integration capabilities, and it will call the service interface by itself, without opening the App. The second is the change of search entry. The large model itself has a very strong external retrieval ability and a very strong multi-modal ability, which is a good search entrance.

IS420ESWAH5A

An artificial intelligence veteran of a car company also observed and thought after the launch of Sora. For the car factory, he believes that Sora has a certain application scenario and value. In autonomous driving training, they have used large models for data generation, but mostly static data. “The dynamic data on traffic flow, Sora is not yet up to the mark, but at least there is some hope.”

In terms of synthetic data, car companies originally had to do it based on a virtual environment (meta-universe), and the virtual environment needed to shoot some videos, and use Nerf to do 3D reconstruction, but the cost was not small, and many places could not be filmed. “In fact, intelligent driving only cares about the scene, not the landmarks related things, so the synthetic data scene should be a big demand for technology like Sora.”

IS420ESWAH5A At Inspur Information, Wang Wujun, deputy general manager of Hardware Research and development Department, and his team have been involved in AIGC training before. According to Wang Wujun, from the video generated by Sora, “the reflection of rain on the ground in the streets of Tokyo has gradually approached an acceptable reality.”

The underlying logic of Sora is trained by a large amount of data, and there are many places in the manufacturing industry that are calculated by humans. If the amount of data is extremely large, the application of Sora’s model has the potential to bring changes to product design, simulation, flexible installation and operations.

“I can’t say what shape it will take, but it will be disruptive.” Wang Wujun made an example, a problem in the industry is “flexible installation”. For example, there are dozens of lines in the server chassis, how these lines are “wound” and how to direct the robot arm to install.

What is possible is that for each cable, engineers need to make a 3D model. Depending on the configuration, dozens of 3D models are required for a cable to be inserted in different positions, and temporary adjustments are required during assembly, resulting in the automatic installation of flexible objects with less effort and less effort. If you can automatically generate a “winding” way through data training, and then directly extract the data to tell the robotic arm how to assemble, the problem will be solved.

Industrial software people who do simulation have also noticed Sora. “If you look at the walking posture of the heroine in the Sora video, if combined with simulation tools, it will be more natural, and now space spirit.” CAE person Li Yan said. Simulation software can simulate complex machines and products, and he sees a combination of Sora and simulation.