Re 1: There probably isn’t a better way to make this a fair comparison, but I don’t find the setup of this study very convincing. What doctors post on Reddit can hardly be compared to their interaction with an individual patient in a doctor’s office:
“To obtain a large and diverse sample of healthcare questions and physician answers that did not contain identifiable personal information, the team turned to social media where millions of patients publicly post medical questions to which doctors respond: Reddit’s AskDocs.”
mesays:
Re 1)
On questions sampled from Reddit “AskDocs” (probably not the most representative task), that may we’ll be in the training data of ChatGPT, which is trained to give “empathetic” answers via RHLF. I’d be careful with interpreting this.
Re 1: There probably isn’t a better way to make this a fair comparison, but I don’t find the setup of this study very convincing. What doctors post on Reddit can hardly be compared to their interaction with an individual patient in a doctor’s office:
“To obtain a large and diverse sample of healthcare questions and physician answers that did not contain identifiable personal information, the team turned to social media where millions of patients publicly post medical questions to which doctors respond: Reddit’s AskDocs.”
Re 1)
On questions sampled from Reddit “AskDocs” (probably not the most representative task), that may we’ll be in the training data of ChatGPT, which is trained to give “empathetic” answers via RHLF. I’d be careful with interpreting this.