Thank you for sharing the resource and the usage policies. It's a really fascinating and difficult problem, especially in those grey areas where intent is ambiguous to solve. I wonder if there’s a path to bring human experts (like psychologists or ethicists) directly into the loop when there’s risk involved? Aka, rather than train the model with their expert opinions, have a direct line to them whenever potentially dangerous content is flagged. At the same time, maybe the harder question is whether these conversations should happen on the platform at all, not because models can’t handle them, but because the potential harm might outweigh the benefit.
Thank you for sharing the resource and the usage policies. It's a really fascinating and difficult problem, especially in those grey areas where intent is ambiguous to solve. I wonder if there’s a path to bring human experts (like psychologists or ethicists) directly into the loop when there’s risk involved? Aka, rather than train the model with their expert opinions, have a direct line to them whenever potentially dangerous content is flagged.
At the same time, maybe the harder question is whether these conversations should happen on the platform at all, not because models can’t handle them, but because the potential harm might outweigh the benefit.