Open-Assistant
Open-Assistant copied to clipboard
LAION-AI
Include Assistant Reply Guidelines to website
Currently, the website seen when replying as an assistant provides no information as to what constitutes as an appropriate and inappropriate reply. The only link is next to the header and leads to this, which obviously doesn't help much. I suspect that many unusable replies don't come from bad actors, but rather from people unfamiliar with the guidelines. The fastest solution to facilitate which criteria apply to the submitted answers would be to either include a list or a clearly identifiable link right between the prompt and the textbox, so users can easily reference the rules and see if their reply would violate any of them. Ideally, examples can give people a better idea and interactivity like messages that ask the user if their submission doesn't violate any of the guidelines can help even more, but this would be a quick and easy way to make sure every user is at least confronted with the criteria.
Now that the guidelines have been updated, could we add a direct link to them on each respective task? Ideally, I'd like to have a collapsible container so users don't have to leave the site, but a simple link that's properly labeled and visible should draw enough attention for more users.
Do we have at least one clearly visible instruction to read the guidelines from registering to sending your first submission? It might be overkill, but since any form of comment sections, forums, chats etc at least tell you to read the guidelines before posting, a data collection project should do the same.