> what LLMs do is string together words in a statistically highly probable manner. I guess the article fails to admit that when you have billions of connected

> what LLMs do is string together words in a statistically highly probable manner. I guess the article fails to admit that when you have billions of connected points in a vector space, "stringing together" is not simply "stringing together". I'm not a fanboy but somehow GPT/attention based logic is capable of parsing input and data then remodeling it in depths that are surprising.