from my poking around, even the high-end models designed for argumentation show no sign of any concept of saliency. They don't pick out small distinctions that make all the difference and they're extremely sensitive to the way prompts are worded. Can't imagine how lawyers could use them for briefs