As you point out, LLMs work much better when you ask them to operate on objects within its context window, especially artifacts it knows how to work with, like code and text. But I think people are so trained to ask questions to the oracle and expect answers (e.g. Google), and who can blame them, that is the UX built into people's muscle memory for open-ended text input boxes. The launch of ChatGPT Search is recognition of this. Plus, most people are being told to treat these chat boxes as strong AI rather than as text/code-processing programs with specific strengths and weaknesses.
https://pca.st/episode/951f3fb4-f7f7-4b4d-b2bb-dbf5334ea1aa
"The Most Interesting Thing in AI - The Atlantic - Episdoe 1 - Machine Consciousness - with Nicholas Thompson and Geoffrey Hinton.
What if the most advanced AI models could think and respond in a way that felt like a human consciousness? How might that transform our understanding of intelligence itself? Some of the leading AI scientists believe that a super-intelligent form of this technology is only five to ten years away. This episode explores the idea of AI consciousness and delves into how the act of dreaming is connected to neural networks in unexpected ways."