APPRECIATION FOR CHATGPT AND UNDERSTANDING ITS LIMITS I think the reason I appre

APPRECIATION FOR CHATGPT AND UNDERSTANDING ITS LIMITS
I think the reason I appreciate ChatGPT (or any ai) is that I have realistic expectations of what it’s capable of, and because my work exposes its limits more so than math or programming, that expectation is consistently reinforced. SO I learn about the AIs mostly from their limits (failures).
I do not find ChatGPT hallucinating any longer. Thought it might be how I craft each of my prompts.
But the reality is that when I’m writing (a book) I know the subject matter. I’m more interested in how to compose the topic or section for readability, and to ensure I haven’t missed including some example or permutation.
The hard problem I’m looking for OpenAI to solve is increasing attention anchors. As I refine a section every increase in precision causes a loss of content. So bulleted lists shrink as I ask for expansions upon them.
This is true across the spectrum. It’s why it fails at reasoning. And I am confident that this can’t be solved without multiple passes, more attention nodes, and much more context memory.