Larry Ellison, founder and chairman of Oracle, outlines the differences between two types of artificial intelligence (AI) models, separating them based on the need for real-time, low-latency decision-making. In a video shared on Elon Musk's X, Ellison explains that the design and deployment of AI models depends entirely on whether they can tolerate network latency or require instantaneous, localized processing.
Ellison: The world needs low latency AI model in a car or robot
In this video from Oracle AI World 25, Ellison said there are types of AI that must make instant decisions when delays of fractions of a second are unacceptable. Ellison pointed to self-driving cars and robotics as prime examples.“If you don't have the network traffic to go back across the network and communicate with an AI model on a far away network, you have to build it another way,” Ellison said, adding, “You need very, very low latency response times.”He specifically cited Tesla as a company adopting this model.“That's why every Tesla car, every Tesla robot, needs to have local computing in the car and local computing in the robot to make decisions on the fly and with very low latency,” he said.He explained that this type of AI requires the model and the necessary computing power to be built directly into the device for “edge computing” to enable critical, instantaneous actions such as braking and steering.
Non-real-time models such as chatbots
The second category includes models that tolerate short delays in computation and network communication. For example, in the case of a chatbot, where it takes time for the bot to process information and then return a response. These models often leverage centralized cloud computing resources.“For example, if you're writing code, it's not required. I can tell you what kind of code to write, and you can think about it a little bit and then give me the answer,” Ellison explained. This type of AI, such as models used for complex data analysis, code generation, and content creation, can spend more time processing queries, gathering resources, and formulating comprehensive responses.Ellison emphasized that both types of models are equally important to the future of AI, concluding, “Both types of models are very important. Both types of models are being built.”
