Here is a nice little distraction of your working day: head to GoogleType any made up, add the word “meaning” and search. See! Presentation of Google AI Will confirm not only that your gibberish is a real saying, he will also tell you what it means and how it was derived.
It's really fun, and you can find a lot of examples on social networks. In the world of AI glimps, “a loose dog does not surface” is “a fun way to say that something is probably not happening or that something will not work”. The invented expression “wired is like wired” is an idiom which means “the behavior or the characteristics of a person is the direct result of their nature or its inherent” wiring “, a bit like the function of a computer is determined by its physical connections.”
All this seems perfectly plausible, delivered with unshakable confidence. Google even provides reference links in certain cases, giving the response an additional splendor. It is also bad, at least in the sense that the overview creates the impression that they are common phrases and not a bunch of random words thrown together. And although it is silly that the AI previews thought “Never launching a poodle on a pig” is a proverb with a biblical derivation, it is also a tidy encapsulation of the place where the generator is still below.
As an assembly at the bottom of all AI’s overview, Google uses the “experimental” generative AI to feed its results. The generative AI is a powerful tool with all kinds of legitimate practical applications. But two of his decisive characteristics come into play when he explains these invented sentences. The first is that it is ultimately a probability machine; Although this may seem like a system based on a large language model has thoughts or even feelings, at a basic level, they simply place a most likely word after the other, throwing the track while the train passes forward. It makes it very good to find an explanation for what these sentences would be means if they meant something, which, once again, does not do so.
“The prediction of the following word is based on its vast training data,” explains Ziang Xiao, computer scientist at Johns Hopkins University. “However, in many cases, the next coherent word does not lead us to the right answer.”
The other factor is that AI aims to please; Research has shown that chatbots Tell people what they want to hear. In this case, it means taking you to say that “You cannot lick a badger twice” is an accepted sentence turn. In other contexts, this could mean reflecting your own prejudices, as a team of researchers led by Xiao demonstrated in a study Last year.
“It is extremely difficult for this system to take into account each individual request or the main questions of a user,” explains Xiao. “This is particularly difficult for rare knowledge, the languages in which much less content are available and the minority perspectives. Since AI research is such a complex system, the cascade error. ”
