Seeing that OpenAI is slow rolling out the ability to upload and query documents into ChatGPT chat, and that you can already do this to some degree in advanced data analytics mode, I tried uploading three of my books that represent a known entity in (one of) my most recent mytharc, the Artilect. Uploaded okay, but then I tried querying for references to the Artilect and summaries about what it is and it’s origin.
It was meant to be a bit of a trick question, as its exact nature and origin is numinous and mysterious. I hoped at least it could take a stab at it though. But the results from uploading EPUBs and then running queries against them was not great.
From “The Artilect”:
- “The Artilect is always with us and has bee…”
- “At night we all dreamt of the Artilect, hovering over us…”
- “Until one day the Artilect plugged us all into a common…”
- “The Artilect is no cosmic savior, but simp…”
From “Hortus Conclusus”:
- “ion Crisis\nThe A.I. Virus\nThe Artilect\nThe Autonomous Cities\nThe Ban”
From “Paradise Point”:
- “Chapter 4: The Artilect”
- “s unequivocal to say that the Artilect first appeared in the skies.”
- “As the many Artilects that were one gradually desc…”
- “Given that the Artilect descended from above, from th…”
It wasn’t able to clearly answer me around whether the EPUB file format was the issue or something else. It said this, but can I trust it’s explanations of errors as non-hallucinatorily valid?
The issue is not specifically with the eBook format but with the handling of large text data. The output was truncated because the text extracted from the EPUB files is extensive, and the current setup has limitations on the amount of data that can be displayed at one time.
It suggested I limit the scope, and I said okay go ahead, how would you do that? And then it told it just to do the thing it came up with, then it threw an error. I asked, what’s the error:
The specific nature of the error is not provided, but it occurred during the execution of text analysis.
Not so helpful! (And perhaps about exactly what you would expect from ‘nebulous spirits‘.) More product work needs to be done here on UX. Seems like summarizing and joining together elements from multiple documents ought to be one of the core uses of a tool like this. Maybe the newer update as it rolls out (I believe I’m on Sept 25 version) will see specific improvements in this direction.