Dev Update: Nobody Puts Baby in the Corner
We know you’ve been holding your breath to get an update on boodleGPT’s development. And wait no longer, we are here with an exciting one. You’ll hear from Kisa Brostrom, Chief Data Officer at boodleAI, and Ansel Teng, Chief Technology Officer at boodleAI, once again this week as they share boodleGPT’s progress, and even receive an introduction to our Alpha.
Data Team Update
Chief Data Officer of boodleAI, Kisa is affectionately referred to as “Queen Kisa”, responsible for all things data. A mechanical engineer by education, she’s built her career on growing data teams from scratch and developing useful, practical solutions. When she’s not herding cattle in her minivan, she’s raising the next generation of data artists (9, 6, 3) on a farm in Oklahoma. Connect with Kisa on LinkedIn.
“Nobody puts baby in the corner.” Johnny (played by Patrick Swayze) was adamant that his dance partner Baby not miss out on her chance to perform their dance together merely because her parents were angry with her. So he goes up to Baby, utters this iconic line, and pulls her out of the corner, and the rest is cinematic history.
There are, frankly, a lot of reasons to put our baby, boodleGPT, in the corner. It sometimes refuses to answer. It sometimes gives the wrong answer. And often the correct answer takes a long time. But our baby won’t grow if it’s in the corner.
So, last week, we released boodleGPT Alpha to a select group of early adopters for testing. This has already yielded a lot of insights and some surprises. What we’ve learned so far:
- boodleGPT can handle really, really complex questions about consumer and philanthropic behavior of the US adult population. Like parents everywhere, we watched with trepidation as our baby faced new challenges (in the form of difficult prompts we had not anticipated) but then were overjoyed when it excelled (by providing relevant, useful responses).
- boodleGPT is determined (maybe too determined). We have built a system where boodleGPT gets an answer, evaluates that answer, then tries to get another answer if it’s not satisfied. While robust, this system is resulting in much longer wait times without feedback than we would have hoped for. We are working to provide more feedback to users after prompting AND speed up the underlying response cycle.
- boodleGPT hallucinates a lot less, but it still hallucinates. We have prioritized getting the right answer rather than getting an answer, so boodleGPT is biased towards admitting ignorance rather than feigning knowledge. However, there are times when boodleGPT reverts to its native bias towards responding and generates answers that aren’t up to our standards. We have a plan in place to improve this further.
Besides Alpha testing and improvements, we are working to put in place the data architecture required to handle organizational structured data (think contact lists) and to incorporate it into responses in a way that keeps the data safe and secure. More on that next week.
Tech Team Update
Chief Technology Officer of boodleAI, Ansel is a senior technology leader since the dot com days. He has a Ph.D. in computer science and has been managing software product and system development in organizations of all sizes, from small startup to global enterprise. Connect with Ansel on LinkedIn. He loves to connect with fellow technology leaders.
With all the hard work from the team, we released the alpha version of boodleGPT on May 9th. We anxiously watched how it behaves in the real world like parents of a newborn baby. We are definitely fascinated by its interaction with our alpha testers and are taking notes on our learnings.
We have made a few changes to refine the user interface and will be making an enhancement release next week. At the same time, we are fleshing out a detailed plan to develop the core capabilities around our future engine that will take in organizational data so that our users can ask specific questions about their unique donors and customers.