It's not the same. At all. Both inputs from both players happen. There is a desync in one or both player's connection, so the rollback goes back to the last moment they were in sync, hence the name.
It's not machine learning or predictive. It is connection detection followed by deliberate action.
If the remote inputs have not yet arrived when it's time to execute a frame, the networking code will predict what it expects the remote players to do based on previously seen inputs.
Regardless what GGPO says, that's not how it works. "Predicts" should be in quotes. It's just a basic (poor) description for common understanding. Saying predict implies the system is changing a sequence when all it is actually doing is using the last input.
Example: Let's say I throw a super and in my motion I tiger knee the motion input (2369, 236+P) and the desync happens during the TK, the rollback will assume my next action is jump because 9 is the last input received even though what I actually did was a super. When it syncs, it corrects to super.
Made the reference already but if people were annoyed with SF6 's input reader giving them moves they didn't want they aren't prepared for AI to just guess at what tapping forward really means
Comments
like, not everything LLM-related is pure evil or useless (although it remains to be seen if its a valid application)
It's not machine learning or predictive. It is connection detection followed by deliberate action.
These are noncomparable.
If the remote inputs have not yet arrived when it's time to execute a frame, the networking code will predict what it expects the remote players to do based on previously seen inputs.
have you seen this man pilot a giant robot?
have you seen what this man does to grandma???????