What is a flower, if you don’t smell?
CLEARVIESIMAGE RF / AAY
The most recent generation of artificial intelligence models seems to have an understanding of the intelligent level of the world, but it is their lack of sensory information – and a body – limits of concepts such as a flower or laughter.
Qihui Xu In Ohio State University and his teammates and lots of language models (LLMS) are about their understanding of nearly 4500 words – all from “Function” and “swing.” Participants and AI models are asked to rate each word for different aspects, such as emotional levels that they hold, or their links to feelings of different parts of the body.
The goal is to see how the LLMs are, including GPT-3.5 and GPT-4 in Google and Geogle, compared to people in their rank. It has become possible that people and AI have a similar conceptual map of words that are not related to the outdoor relationships, but different from words involved in senses and physical actions.
For example, AI models believe that one can experience flowers through torso – something to see most of the people or a sniff.
The problem, says XU, so LLMs strengthen their understanding of the world from text socked-up from the internet, and that is not enough to understand the concepts of thinking. “They are different from people,” he said.
Some AI models have been trained in visual information such as pictures and videos in addition to the text, and the results of the words can increase by human tests that are closer to human world models.
“We have told us the benefits of making modi-modal training to be greater than we expected. It is like one more than two,” Xu said. “In terms of AI’s progress, it supports the importance of developing multi-modal models and the importance of having a body.”
Philip Feldman At the University of Maryland, Baltimore County, says the provision of AI models is a robot body and see the robots well input the people around them.
Prevention of such risks means increasing guard railroads to robot actions, or use only soft robots that do not harm training, as having their own reduction.
“It’s strong how they understand the world,” Feldman said. “One of the things they know is that you can boil things, because they have a little mass. And now you try to contain each other with each other.
Topics:
