Large language models are trained to predict words, not to check facts.They are optimizers of plausibility, not validators of reliability. How AI Tricks Us Into Trusting It