I really feel this project should be gpl’ed with the affero clause. Legally requiring any website or service using the conversational AI to disclose the source code and by extension the fact that the conversation is with an AI could do a lot to curb some of the inevitable excesses that will arise from widespread use of the technology.
I just spent approximately thirty minutes asking some tough academic/historical questions. To my surprise, OpenAssistant was 100% correct with every response.
On closer inspection it may be the case that your questions were answered by humans. This appears to be as much a training data crowdsourcing operation as it is an AI project. Should also be noted that your interactions will be released as open data under CC-BY 4.
I have been trying it out too, it is much better than I expected.
Can it be self hosted?
I believe so, I think he mentions it either in this video or another video of his.