Facebook Says Its New AI Can Detect Multiple Problems Quickly


The Little-Shot Learner is trained on the fire of billions of Facebook posts and images in over 100 languages. It is designed to be treated with additional training that contains posts or pictures written in previous projects as well as a simple description of the points he or she has broken.

After preparation, the process can be guided to find new types of objects, such as establishing a new law or developing a new language, with less effort than previous models, says Cornelia Carapcea, marketing manager for moderation AI at. Facebook.

Ordinary medical systems may require hundreds of thousands or millions of samples before shipping, they say. A few Shot Learner can be used by using a number of them – “a little drawing” in its name – combined with simple descriptions or “notifications” of new concepts to which it is associated.

“Because it is already obvious, learning a problem or a new strategy can be quick,” says Carapcea. “It is always difficult to maintain a complete record of such matters as violence, hatred, and incitement; This helps us to act fast. ”

The Little-Shot Learner can also be guided to find content groups without showing any examples, simply by providing a written plan for a new strategy — an amazingly simple way to connect to the AI ​​system. Carapcea argues that the results are not reliable in this way, but the method can quickly identify potential abuses with the new system, or identify documents that can be used to improve the system.

The impressive potential – as well as the many unknowns – of large AI creations such as Facebook prompted Stanford researchers to recently set up a training ground for such systems, which they dubbed “.”examples of foundations”Because it seems to be the basis for many professional projects. Large-scale machine learning systems are being developed for use not only on social networking sites and in search engines, but also in industries such as money and health care.

Percy Liang, chief executive of Stanford Center, says the Facebook platform seems to be showing off the incredible power of the new models, as well as showcasing some of their products. It’s fun and useful to be able to streamline the AI ​​system to do what you want with written content, as Facebook claims to end with new ideas, Liang says, but that capability doesn’t sound right. He stated: “It is a far cry from science.

Liang says the speed of the Lesser Students may also be problematic. While engineers do not need to improve the quality of training, they stop improving with the knowledge of the feasibility of their system. “There is a greater faith,” Liang says. “With more automation, you have less control power.”

Carapcea of ​​Facebook says that as Facebook develops new optimization systems it also develops ways to monitor how it works to ensure it is accurate or impartial.


Some of the Best WIRED Stories



Source link

Leave a Reply

Your email address will not be published.