GPT-4 is really, really impressive!<p>But however we spin it, ChatGPT still has a significant context issue. Sure the increased input token max for GPT4 does help but when you think about the context we draw on in out typical life interactions, which spans decades of experiences it is still quite limited.<p>One might think "yeah but what have my college years have to do with the task of doing my taxes?" but in the real world we do draw from multiple sources when performing large domain tasks (taxes, law, medicine, programming).<p>So while the performance on specific tasks seems to have increased quite well (its even evident when you just talk with it on ChatGPT) the effectiveness in assisting in complex legal cases (a task that could benefit greatly from ai analysis) is still limited. Add to this the fact that it cannot estimate its own truthfullness and the only domain where it still can be applied is fiction...