jmsfbs on Nostr: A trained LLM is an n dimensional space that maps out probable relations between ...

A trained LLM is an n dimensional space that maps out probable relations between tokens. This is what allows it to rank the next likely token which creates the illusion of thought or understanding.

This is very different to creativity. This is adhering to clichés.

I think this is also what 'taste' is. Being aware of different layers of these probabilistic choices while consciously curating alternative possibilities with some level of intention.

#ai #art