We have yet to see what the recursive self-improvement will look like, but I suspect open models will have a lot to do with it. In a way, we are already kind of seeing a glimpse of this with the rapid growth of token usage among open-weight models.