As I mentioned it in an aside elsewhere:
Chatbot talks to Chatbot. (Not that it
hasn't been done before.)
And, frankly, I'm disappointed in what I've seen.
To quote in the article:
"Did you know why women can't put on mascara with their mouths closed?" asks Kuki.
"Thank you for sharing it with me," replies Blenderbot, politely but entirely missing the point, adding for good measure: "You are a good person."
Knowing something about how the
grande dame that is Eliza works (I typed in a minimal version of her into a BBC Microcomputer in the early '80s, and I subsequently found it almost impossible to 'fool myself' with the IRC version afterwards, though I could still enjoy how others interacted with her) it's entirely obvious that Kuki's "Did you know <foo>?" was contextualised as "Fact: <foo>" in Bb's 'mind', spurring the generic 'I don't immediately know[1] what to do with that' response. Needs a bit more work (manually or by automated competitive 'learning', for which you'd probably also need automated competitive
teaching) to extract and identify just that little extra syntactic sense...
A bit like the bits in the lined Eliza/Parry that go something like "*It's clear to me
if not to you." "
Do you think its likely that not to I?" (my formatting to explain the 'logical' progression).
[1] Or may have a choice set of fuller responses, but opted for the generic "Thanks, please continue" one.