It’s hard to compress text deterministically beyond a certain point but LLMs seem quite far from that point. I’d certainly prefer such a model…

→ View original post on X — @naval