Conversation

Replying to and
While it's definitely possible to get better than Markov chain results with few shot (or pertained and then fine-tuned) NNs, the recent replication of the full GPT-2 model has been underwhelming. It seems like results that have led us to believe otherwise were cherry-picked! 😕
1
Replying to and
Let's face the reality, fully automnomus won't be that great as of now, use the existing solution but with man-in-the middle deal. You, somebody you trust or a Twitter poll can be consulted to edit/approve/ choose the best response or message. TLDR:Introduce editorial tooling
1