Totally, I agree. It's an important proof of concept but given the profit margin mentioned in the paper and the ease of automating all processes, this could quickly become quite profitable (in the short term).

Drew, I have some news for you ;)

Yes but the paper also demonstrates that (1) even a few such LLM generated respondents can skew responses, and (2) that there is a clear business model whereby opt-online panels can be used to make money using such agents.

This was compiled by Michael Pruser from Decision Desk: docs.google.com/spreadsheets... Not yet fully updated (scroll to right)

This is awesome Drew! I think this will lead to a ton of really interesting discussions

Would you be willing to share the assignment prompt? Seems like a great exercise!

Thanks for sharing, Drew. This is going straight into my lecture slides.

Thanks, Drew! I was hoping the paper could be useful for thinking through information equivalence/masking concerns in conjoint experiments (and other designs) more broadly.

Thanks Efrén!

Also came here to say there's a non-zero chance of this.