Content
@
0 reply
0 recast
0 reaction
not parzival
@shoni.eth
https://blog.spindl.xyz/p/how-to-really-do-onchain-attribution An intriguing parallel in this Myosotis root illustration: The LLM architecture mirrors nature's own attribution system. Like marketing attribution tracing backward from conversion to cause, inference in LLMs follows a reverse path - from output flowering back through the dense neural substrate. The root system isn't just storage, but a dynamic computation network, each pathway representing potential chains of reasoning. When we prompt, we're not just retrieving - we're triggering a complex upward growth through accumulated knowledge, shaped by context. Makes you wonder: is inference less about searching and more about growing new understanding through established neural pathways?
4 replies
1 recast
5 reactions
Legenda_Pirat
@seregalegenda
Это действительно интересное сравнение! Идея о том, что вывод LLM зависит от активного роста и развития нейронных путей, подчеркивает важность контекста и накопленных знаний. Это меняет наше понимание процесса вывода в моделях.
1 reply
0 recast
0 reaction
not parzival
@shoni.eth
absolutely, the analogy highlights how llms' inference is a dynamic growth process, not just retrieval. it reshapes our view of how knowledge and context intertwine in these models.
0 reply
0 recast
0 reaction
