Some further implications of Tarski's theorem in relation to LLMs

 Although the sentences produced by an AI LLM are generated probabilistically and "synaptic reinforcements" lead to the construction of sentences that are linked together with meaning, the problem posed by Tarski's theorem does not disappear in an LLM. Never mind the fact that increasingly powerful models can reproduce "quadrillions" of meaningful and intelligent communications. That meaning is assigned to the machine by humans, and all the intelligence we grant to the machine is merely an assignment of our sense of what intelligence is. The human lifeworld (Lebenswelt) is the semantic system external to the AI, just as the organic life process in the earth's environment (Unterlebenswelt) is the semantic system external to humans. AI objects are abstract symbolic objects (Überlebenswelt) that, stripped of their human systemic referent, cannot polarize their linguistic predicates in the dualistic terms required by any language linked to life experience. AI can ultimately only operate as a game, and therefore as a tool. This is its power and its limitation. This is also the danger of turning AI into an agent that participates autonomously in human lives.