Sam Altmann says that superintelligent robots are coming.

We are ready for the next step in the AI. The fast internet and deep neural networks turn superintelligent robots into reality fast. Nobody predicted how fast this advance could be. 

The AI-controlled robot can use large language models remotely anywhere where it can reach the network. The network allows that robot to communicate with supercomputers. If a robot or its computers have access to supercomputers and LLM they can give very intelligent impressions of themselves. The impression is like the cases where LLM is connected with microphones and loudspeakers. 

The robot must have cameras as eyes. Microphones and loudspeakers for vocabulary communication and text-to-speech/speech-to-text applications. The system transforms speech into text and then dumps it for the LLM. During that process, the system must have an error correction protocol. 

That allows the robot to communicate with the LLM. The vocabulary correction makes the speech more understandable to LLM. If the LLM doesn't understand what the user says it can give the list of choices, what the LLM thinks the user wants. That makes it possible to give commands to robots in a noisy environment. 

If the LLM doesn't recognize those words it can give a list of what the word looks like, and then ask to select from the menu would the word be A, B, or C. Or if there is no match the user can give the command again using better articulation. 

Satellite communication allows the robot to communicate with fixed supercomputer sites. Anywhere on the planet. The only needed thing is the system has access to a satellite. 

The LLM can itself be in the computer center. But the robot bodies can also create neural networks by connecting each other. Into one entirety. This kind of network-based application benefits the shared calculation capacity. That makes those systems handle information very effectively. 

But when we think about superintelligent robots the robot itself is not intelligent. A supercomputer that controls those things is intelligent. Or the network-based solution that connects those robots in their entirety is intelligent. However individual robots are not intelligent. Their computing capacity is low. But united they are strong. 

In robotics superintelligent robots are tools that can follow spoken commands. 

They are tools that can have the same abilities as humans except they are not living. The controller can give spoken commands to the system. The weakness of that model is the background noise. The solution for that thing can be voice recognition that allows the system to deny unauthorized commands. 

Active denial or active filtering is the tool that denies robots and LLM to take unauthorized commands. The AI can also ask for confirmation after each command. That thing means that the system introduces what it should do and waits the acceptance before it starts to act. The command can be "clean the yard" and when the robot starts to operate it can clean its entire area without different orders. 

And those robots can communicate with the user. Those robots are only tools that follow the supercomputer orders. We can think that those systems are like advanced mobile telephones. 

https://bigthink.com/business/sam-altman-the-superintelligent-robots-are-coming/