models are stuck in time. eventually the world moves forward and it is trained on too much obsolete data.

training cannot end for LLMs intrinsically. it's not some fixed cost. it's an ongoing one.