The logic of a llm is that it creates phrases from patterns of “most likely” sequences. In some contexts this can seem right (like with predictive text on) but that hides how in technical areas AI actively drags peoples belief towards the generic most common usage which is often plain wrong. And the more AI is used, the more a body of mediocre or false information will exist, creating a devils circle of language use. People who strive for perfection or improvement will be drowned in the flood of lazy AI.