In "Label Assistant Reply" subdivide "Quality" into "Accuracy" and f.e. "Grammar"

[ChaoticHuman]

I think it would be a good idea to subdivide the "Quality"-category in "Label Assistant Reply" into "Accuracy" and f.e. "Grammar" if that is possible because the term "Quality" is a relatively vague term. And I also think it would generally be a good idea to focus the training of Open Assistant more on accuracy as inaccurate outputs are one of the biggest issues of competitors like ChatGPT, etc.

Ok, I must admit though it's not always easy for people to asses the accuracy of an output and it's admittedly also possible that people might bring in their own biases meaning I would still keep answering the new categories "Accuracy" and "Grammar" optional and would perhaps suggest to the main-developers of Open Assistant to see the user-contributed "Accuracy"-assesment as merely a first step in checking the accuracy of assistant replies.

[ChaoticHuman]

Ok, I was thinking about closing that issue again because I just found out that the dashboard also includes a function sometimes popping up called something like "Rank assistant reply" where you are apparently supposed to rank assistant replies based on accuracy however as the issue got two likes I will keep it open for now.

[olliestanley]

I don't think it's practical to change the labelling options with so much data already submitted, but maybe others have other views so we can leave this open for any further discussion

[olliestanley]

Going to close this as covered by #1412