2 posty
The world's most powerful AI runs on a method that is almost 180 years old.
In 1847 the French mathematician Augustin-Louis Cauchy came up with gradient descent - a way to step down to the minimum of an error, one step at a time.
ChatGPT, Claude and every modern neural network use this very principle today.
Cauchy developed the method to calculate the orbits of celestial bodies. Now it trains artificial intelligence.
Sony patented AI that predicts the next frame of a game. The system generates a frame based on the previous image and the player's actions, without waiting for the graphics card to finish rendering.
The system runs on a neural network with an encoder and a decoder: the first compresses the finished frame into a compact data set, and the second uses it to create the next image.
Unlike DLSS 3, which builds an intermediate frame between two finished ones and can increase latency, Sony predicts the future frame itself. In theory, this provides smoother motion without extra lag.
