> Of course everything right now is highly speculative but this doesn't mean its not possible.
I don't even think this is true either. A year or so ago lab leadership was worried about how they were going to keep progress going when they knew they needed a paradigm shift to handle things like continual learning or breakthroughs on training efficiency. To the point where they invested valuable resources in this bet.
It turns out, this was actually somewhat counterproductive. It turns out that while it may not be the best or most efficient solution, the current paradigm is clearly enough to close the gap to super intelligence.
No one seriously paying attention to performance trends and what's happening at frontier labs that I'm aware of thinks that RSI is more than a year away.
What that will actually look like and the societal implications are more speculative than the actual model capabilities.