Google open-sourcing AI assistant datasets


#1

#2

Taskmaster-1 (interaction dialogs like ordering pizza): https://ai.google/tools/datasets/taskmaster-1

CCPE (snippets of two people discussing movies): https://ai.google/tools/datasets/coached-conversational-preference-elicitation

Text-only data.


#3

So it would be more useful for Padatious and the like then?


#4

I’d probably strip the sentences out for deepspeech training sentences, but I could see it being used with xlnet or something similar to help it make preference based answers.