The key wording in the new leak is "AI Neural rendering to revolutionise graphics"
The way I interpret this, is that DLSS4 will be able to render new game data like polygons in the render pipeline - so rather than having frame Gen which just takes existing game data and makes a copy frame, DLSS will now render new game data to create new frames - my guess is that this will be done to make games look better when using DLSS, but it is technically even more fake because DLSS will create game data and assets that don't actually exist in the game, even if it looks the same to your eye, AI software should be able to tell you the individual pixels are different and how different they are.
I wonder whether we'll have the issue of ghost data - e.g rendering extra fingers on hands etc - but maybe not because dlss will have real game data to use as a reference so in theory it should be able to know what the assets should look like, and then just improve those pixels to make it look better and sharper for the player and probably create even better anti aliasing. This will be interesting