"The human evaluation results indicate that the response generated from DialoGPT is comparable to human response quality under a single-turn conversation Turing test."
Is there a purpose to this, other than tricking the user into thinking they are talking to a human?
One the one hand, if you have a large database of facts on which you can train this to accurately answer questions, why not just index the database in Elasticsearch and let your users search over it? This is just as efficient, a lot cheaper, and a lot more transparent to your users.
On the other hand, if you _don't_ have such a database, the thing will just end up lying to your users. Which is basically a worst-case-scenario, because you're abusing trust that you gained by pretending to be human.
I will take wikipedia + elasticsearch/google/bing/duckduckgo over a chatbot trained on wikipedia any day.
4 comments
[ 3.1 ms ] story [ 18.6 ms ] threadPaper here: https://arxiv.org/abs/1911.00536
How do we regulate this thing? Forbid bots from social media and chats altogether? Force them to introduce themselves as machines?
I ll be damn’d if that bot aint a libertarian
One the one hand, if you have a large database of facts on which you can train this to accurately answer questions, why not just index the database in Elasticsearch and let your users search over it? This is just as efficient, a lot cheaper, and a lot more transparent to your users.
On the other hand, if you _don't_ have such a database, the thing will just end up lying to your users. Which is basically a worst-case-scenario, because you're abusing trust that you gained by pretending to be human.
I will take wikipedia + elasticsearch/google/bing/duckduckgo over a chatbot trained on wikipedia any day.