I thought I’d share a new assignment the students and I are doing. It’s not a big deal assignment, but I wanted to share it anyway. I call it Roleplay, Deep Research, Roleplay.
What is that? Well it could probably use a better title but it’s a way that I was wanting students to experiment with Roleplay and Deep Research in an environment that I thought would help them — by learning what they want to do, and if they don’t know, build out some conjectures by having ChatGPT interview them, then do deep research around that, and then follow up with a second interview where ChatGPT uses all of it to re interview them, to then update its profile. Let me explain.
But wait, before I do, let me flip some coins. Best two out of three to figure out the paywall…
Heads! So this one is paywalled. But maybe consider being a paying subscriber! For the price of a cup of coffee a month, you can support this substacks efforts to expand access to causal inference methods all over the world.
Roleplay interview
It’s probably a reflection of my “Cosmos is Pinocchio and just wants to be a real boy” approach to large language models that I tend to use them to interview me whenever I get stuck. I sometimes have ideas that seem to be very far apart from one another and so rather than have ChatGPT figure out what I’m trying to get out of me, I have ChatGPT interview me to help gradually pull it out of me collaboratively. And the prompt is usually this:
“Interview me [about the topic we’ve been discussing]. Ask me no more than four questions arranged in highest level bullet format. And ask no more than one follow up question per high level question such that you do not ask more than eight questions total. You must ask only one question at a time. Your goal is to figure out deep down what the heck I’m talking about because I can’t quite see it. Help!”
Keep reading with a 7-day free trial
Subscribe to Scott's Mixtape Substack to keep reading this post and get 7 days of free access to the full post archives.