The research scientist sat in the center of a bright, open-plan kitchen and typed a command into a laptop in front of a one-armed, wheeled robot. He said he was hungry. The robot swooped in and grabbed a bag of chips with a large plastic pincer, and then wheeled over to Xia to give him a snack.
The most impressive thing about that demonstration was the fact that no human coder had programmed the robot to do what was asked of it. The control software had learned how to translate a spoken phrase into a sequence of physical actions.
A person doesn't have to use preapproved wording to issue commands, as can be necessary with virtual assistants If you tell the robot that you are thirsty, it should try to find you something to drink, and it should return with a sponge.
This is courtesy of the internet.
In order to deal with the diversity of the real world, robots need to be able to adapt and learn. Machines need to understand how words can be put together in a variety of ways to generate different meanings. Hausman said that it was up to the robot to comprehend the little nuances of language.
The goal of creating robots capable of interacting with humans in complex environments was a step towards being achieved. In the past few years, researchers have found that feeding huge amounts of text taken from books or the web into large machine learning models canyield programs with impressive language skills. Software can pick up the ability to summarize or answer questions about text by analyzing the many forms of writing online.
Large language models are being used by big tech firms for search and advertising. There are a number of companies that offer the technology via the cloud, and new services have sprung up applying the technology to tasks like generating code or writing advertising copy. Lemoine was fired after he warned that the LaMDA technology could be sentient. The vice president wrote in The Economist that he felt like he was talking to an intelligent person when he chatted with the robot.
Artificial intelligence programs are prone to become confused or regurgitating gibberish. Carefully engineering may be required to reliably guide a robot without it running amok, as language models trained with web text often reproduce biases or hate speech found in their training data.
Hausman showed off a robot that was powered by the most powerful language model yet announced by the company. It is able to explain in natural language how it comes to a conclusion when answering a question. A sequence of steps that the robot will execute is generated using the same approach.
All Rights Reserved © 2024
