"Inherently an LLM cannot go beyond the realms covered by the inputs. There is no mechanism to somehow bootstrap forward-looking beliefs about the future - nor causal logic or knowledge - beyond what can be inferred from the existing statistical associations and correlations found in the words in the training data."

"AI and AI-inspired models of cognition are based on backward-looking data and prediction rather than any form of forward-looking theory-based causal logic."

"AI cannot causally map and project into or anticipate the future, as illustrated by LLMs which are delimited by past data."