2 ms·
Given that LLMs are trained on tokens and not symbols, that is, they have no context or world model, I would say that LLMs cannot symbolically abstract, given t
by slashdave 2y ago
Given that LLMs are trained on tokens and not symbols, that is, they have no context or world model, I would say that LLMs cannot symbolically abstract, given that they have neither a mechanism nor the data to do so.