Though it’s likely no secret to those who have been devoting significant time and effort to the field, the way in which a user interacts with generative models and systems is quickly becoming as important, if not more so, than the underlying training and inference technology when it comes to generative AI. ChatGPT by OpenAI, the latest and, in my opinion, best such example, has just gone live as a free preview of research for anyone to try.

ChatGPT is an application of their new GPT-3.5 natural language generation technology that can be used in a web browser, much like chatting with a colleague or a customer service person on a website, if you haven’t heard the buzz about it yet.

Already making headlines with its DALL-E picture generating technology, OpenAI’s GPT series has gained attention with each each iteration (and occasional existential dread on the part of writers). However, the latest chat-style iteration appears to have broadened its appeal and audience, shifting the conversation away from “wow, undergraduates are going to use this to submit bad but workable term papers” and toward “wow, this could actually help me debug code that I intend to put into production.”

So far, evidence suggests that it is significantly improving its ability to write term papers, while it still has some work to do in avoiding certain common traps for AI chatbots, such as passing off fictitious material as reality. Even though it’s a new version with stronger core inference technology and a new interaction paradigm, ChatGPT’s popularity seems to be skyrocketing, and users seem to be more pleased with it than they were with GPT-3.

Personal, mundane example of why I think this is so powerful: I recently asked ChatGPT to tell me which Pokémon Types are strong and which ones are weak, and it did exactly what I hope Google will do every time I enter a Tera Raid in the new Pokémon Scarlet game and have to try to remember which counters which.

To wit:You may have noticed that my question was not very flowery; in fact, it is as straightforward as it is possible for me to make it without losing clarity. And the outcome is precisely what I need, rather than a list of items that could help me locate what I need if I put in the effort (like Google does):

The concept that ChatGPT or something like may one day replace a search engine like Google isn’t new, but this delivery of OpenAI’s core technology is the closest approximation yet to how that would really operate in a fully fleshed out system, and it should have Google worried.

The gold rush in generative AI will be driven by establishing unique, defensible companies based on how it shows up, rather than what’s beneath the hood, and ChatGPT is the finest yet embodiment of this idea.