But LLMs don't "explore" L-space, the training process just maps all of the available conceptual landmarks inside that space. There are still enormous (infinite?) tracts of unmapped territory that LLMs never reach, and that's exactly the space where novel ideas live.