Existing neural networks operate on the basis of the so-called approximation system: they are capable of
These operations require a lot of time and howconsequence, significant computing power. Facebook engineers found a solution to this problem - they taught their neural network to consider the problem not as a mathematical one, but rather as a language problem.
In other words, the researchers taught their AIspeak a mathematical language. The result of this approach was the acceleration of problem solving by several orders of magnitude compared to systems for solving algebraic problems such as Maple, Mathematica and Matlab.
Our AI sees the task as a kind of language, and then considers the solutions as a translation problem for sequential neural networks.
Facebook previously stated that itprohibits publishing deepfake videos on its platform. The social network will automatically delete videos edited using AI, where "the audience is misled using words that the characters did not pronounce."