I don't know of any chat logs. But it was very interesting to see the 'next response' candidates for the two sample chats in Table 1 (p3 of the PDF). In parti

I don't know of any chat logs. But it was very interesting to see the 'next response' candidates for the two sample chats in Table 1 (p3 of the PDF). In particular : it was alarming to see how much their Deep Learning response selection mechanism had to chose not so much the best response out of a selection of decent responses, but more the most acceptable response out of a selection of mostly horrible ones.