Adventures in genAI: Comparing prompts
As many of you know, I teach entrepreneurship at Sy Syms School of Business in what I laughingly call my "spare time."
The Dean challenged us to go beyond policing against AI and to actively incorporate it into our teaching style. Challenge accepted.
Before each new module, I now assign "AI Presearch" so my students can take a stab at teaching themselves a bit about what we are about to cover. In addition to making the class discussions that much richer, it has also given me a very interesting data set of comparative prompts and the results that they generated from ChatGPT.
*Quick tip:* the students who asked for "detailed instructions" or "step by step instructions" got much richer and more actionable responses.
What I found fascinating was how different the responses the students got were on topic that really should not have differed.
For example, we just started discussing fundraising and all of chatGPT's responses included a recommendation to begin with a list of a certain number of prospective investors. But the recommend *number* of investors for this list varied wildly: the low was 30 and only one response called for the 50-100+ that I recommend to my startups and in my book, The Entrepreneur’s Odyssey. For the life of me, I can't see anything in the prompts my students wrote that would cause ChatGPT to come up with such wildly different estimates.
Similarly, most of the 15 responses correctly emphasized the importance of warm introductions. ChatGPT generally instructed students to go to their network and ask the (IMHO suboptimal) question "who do you know who might be interested in investing in my startup?" Only one response correctly recommended compiling a list of perspective investors *first* and then following up with "do you know any of the investors on my list? Do you think they might be a good fit?" for better results. Again, there was nothing obvious in that student's prompt that should have elicited different (and much better) advice.
The purpose of the assignment was to prime the pump for the class discussion and my students were going to get the 'right' answer in class so I was not all that alarmed by the widely varying advice. But for entrepreneurs relying on AI for critical information for their startup, my advice to them is clear: AI is great at giving you the broad outlines of what you should be thinking about but get yourself a really good human mentor as well.
