We can’t test it at scale, but (a) we narrow the path through the world model si

We can’t test it at scale, but (a) we narrow the path through the world model significantly (b) LLMs don’t process like traditional programs but what’s effectively in semantic parallel because it can handle so many more relations than humans can. The existing LLMs estimate 40% reduction. But again, we can’t test it at scale so it’s simply impossible to testify to.