It's Time for ChatGPT to Grow Up

Yet again, ChatGPT provided a surprise this week. This one, however, was unwelcome. Widely proclaimed as the next big thing in generative AI, ChatGPT proved its unpredictability by turning off some of its newest features without warning or explanation. 

This week, we were relying on the new vision and data features for a workshop on how to make better decisions using generative AI. Oops. Those new features had disappeared. Users might think that paying for ChatGPT Pro would provide some consistency and reliability. But, buyer beware, with ChatGPT, you never know what you’re going to get. And, on this day, some of the most recent and most interesting features were inexplicably turned off.

OpenAI released ChatGPT as an ‘experiment’ and continues to release new features as betas. Legally, that likely helps it get around the fact that the product is unpredictable. But how long will customers put up with this unpredictability? How can OpenAI’s tools become essential digital partners if users can’t count on the tool to be available? When will ChatGPT become a real product?

Reliability, trust, security, privacy. No software company has succeeded without these. Some may think of these characteristics as independent, but we see them intertwined in user perception. If we can’t trust an application to be reliable, can we trust that application to keep our information secure? Can we trust it with our private information? These questions are at the heart of every enterprise conversation about generative AI: everyone wants to jump into the new thing, but they find trust elusive. 

Generative AI is inherently unpredictable. If a traditional software product performs unpredictably, it’s called a bug. The software goes back to engineering for a fix, improving reliability and predictability. 

Generative AI is entirely different: its power is in its creativity and unpredictability. We use these tools because they can create connections across the trillions of parameters in the data cosmos. We might be specific with a prompt, but we can never predict a response with specificity. 

That doesn’t mean, however, that we want the tools to be unpredictable themselves. To depend on these new digital partners, we need them to perform when we need them—if they don’t, they’ll never reach their potential and promise. 

ChatGPT is about to celebrate its first birthday. It’s time for it to graduate from an experiment into a true product.