How it works
R2’s brain is trained once and then frozen. On top of it sits a team of skills: memory, notes, imagination, exploration, and a part that decides which skill to trust.
Each user gets a small memory of their own, kept with them. It learns from what you do while R2 works: a mail you mark as spam, a photo edge you correct, a move in a game.
The brain never changes, so nothing it knew can be forgotten.
Where we test it
In the lab, R2 works on spam, schedules, notes, games and images. It is not in a product yet. All our AI work today is on R2.
What comes after: R3
R3 is our long-term plan: a brain that grows itself. It grows and prunes its own neurons in a sleep phase between lessons, instead of being trained once and frozen.
In lab tests it did better than a frozen copy of itself, and better than a copy built full size from the start, without forgetting earlier lessons.