The Age-Old Comedy of GenAI Development

Our CEO Moritz Sudhof recently spoke at the Smart AI 100 Summit in San Francisco. In the video of his talk, he appears initially wearing a standard-issue engineer’s hoodie. But then, in the span of about 30 seconds, he cycles through a stethoscope, a baseball cap with “Business expert” (?) written on the front, a lab coat, and a (very poorly tied) tie. There are moments when it seems like he might disrobe entirely. (He does finish the talk wearing much less than he started; the full series of events may be lost to time.)

What was Moritz doing? He was acting out The Age-Old Comedy of GenAI Development.

Act I: Moritz’s first role in this one-person play is that of the talented and well-informed AI engineer. He is wearing his standard-issue engineer’s hoodie. He has internalized all the lessons taught to him by his colleagues at the imaginary Oak Creek Clinic (“The Bee’s Knees in Joint Care”). He is feeling good.

Working in the Bigspin prompt IDE, Moritz begins with a one-sentence seed prompt that defines a new post-surgery recovery chatbot. He relies on the Bigspin assistant to flesh this out into a full system prompt. Again with help from the assistant, he ensures that his system includes the guidance given to him in the past by his product manager, the marketing team, the patient success team, and the clinical team. He runs some quick tests in the app, checks in with the assistant about a few areas for potential improvement, and then clicks “Share” to get feedback from his colleagues around the clinic.

Will this be the first and only time in our hero’s career when everything is perfect on the first try? He knows in his heart that it won’t be, but he is not worried.

Act II: A chorus of Oak Creek Clinic employees are now buzzing away in Bigspin. They have been given a view of his system that allows them to review existing predictions, test new inputs, add comments and verdicts, and regenerate new outputs.

The feedback pours in. Moritz is playing all the roles now. In his stethoscope and lab coat, he becomes a clinician. He notes places where the clinic’s pain scale should be mentioned and emphasizes the importance of the branded phrase “Motion is Lotion” for post-surgery recovery. With his “Business expert” cap on, he is a hard-nosed PM. He tests how the system responds to queries about the clinic’s competitors (Moritz the engineer had not given this any thought) and has some trenchant comments on the outputs.

Act III: And so the drama continues. Most of the verdicts are thumbs-down, and the comments are plentiful. The situation looks dire for Moritz the engineer. However, he has a secret weapon. With all the feedback in (for now), he clicks “Learn from these examples” in the Bigspin app. The gears turn behind the scenes, and finally the Bigspin assistant suggests system updates that reflect all the feedback he received.

Our hero updates the system, runs a few more tests, takes a deep breath, … and clicks “Share” again to request feedback from legal.

Photo credit: Kevin Mei #friendsofkevin

Chris Potts
Chris Potts
Co-Founder & Chief Scientist