Meta’s New Few-Shot Learner Can Detect Misinformation With Little to No Examples


Meta announced a new artificial intelligence tool called Few-Shot Learner that it already uses on its platforms, such as Facebook, to combat misinformation and other harmful content. With Few-Shot Learner, Meta systems can adapt to new threats faster and require fewer, if any, labeled examples to learn.

In addition to speeding up the AI ​​learning process, the AI ​​Few-Shot Learner works in over 100 languages, which means that some misinformation about COVID-19 in French would be stopped before someone cannot come, translate it into another language, for example. example, English, and allow it to run rampant in English-speaking conspiracy circles. Much like some of Meta’s other AI systems, the Few-Shot Learner is able to learn from different types of data, including images and text.

Discussing the potential of the new technology, Meta said:

“We believe that FSL can, over time, improve the performance of all of our integrity AI systems by allowing them to leverage a single, shared knowledge base and backbone to deal with many types of violations. There is still a lot of work to be done, but these early production results are an important step that signals a move towards smarter, more pervasive AI systems. »

The Few-Shot Learner Training Process

From training to deployment, Meta said FSL can be deployed in about six weeks, whereas traditional AI systems designed for the same task take up to six months to deploy, making it easier to spread the virus. misinformation and harmful content. If you want to learn more about the gritty details of Few-Shot Learner, be sure to check it out Meta AI Tech Blog Post.


Comments are closed.