ChatGPT is blowing up. Twitter is inundated with screenshots of the app, coding sites like Stack Overflow are already banning answers produced with it, and over 1 million people have played with it. It’s a sensation.
As a professional AI researcher, I wouldn’t have called that. ChatGPT is trained specifically to act as a chat bot, but fundamentally it’s using the same GPT-3 technology that’s been available for over two years now.
What ChatGPT 3 demonstrates — moreso than impressive technology — is the crucial role that access plays in making breakthroughs truly usable. By packaging GPT-3 in a way that normal people can use, OpenAI has finally made people sit up and realize the incredible power of today’s AI.
This is nothing new; we think of Thomas Edison as the inventor of the lightbulb, not because he actually invented it, but because he successfully brought it to market and turned it into something that normal people could understand.
That will be a trend in the AI industry going forward; the companies that make using AI as easy as possible will be the ones that thrive.
The Importance of Use Cases
Most of today’s impressive AI systems are built on massive language models. These language models are trained on, essentially, all of the text humans have created over the last 6,000+ years.
GPT-3 ate 8 billion pages of text, almost every book ever published, and all of Wikipedia. It spat out an AI system that exhibits properties of general intelligence and can do everything from writing sea shanties to solving coding problems.
None of that is new. I started beta testing GPT-3 in 2020, and the fundamentals of the system have been around for much longer than that.
Interestingly, tools like GPT-3 have been baked into all kinds of apps without anyone really knowing. Many of the AI writing assistants that writers here on this platform either fawn over or scream about are just fancy wrappers around GPT-3.
Likewise, lots of utilitarian text on the Internet — think summaries of a restaurant’s menu or little blurbs about what you’ll find in a new city — are already written by AI systems like GPT-3. You’ve probably consumed text written by GPT-3 and not even known about it.
The Power of Access
So why is ChatGPT such a sensation if the technology behind it has been out for a while?
It’s because ChatGPT makes that technology incredibly accessible. The chatbot is free to use, and every day people can sign up and interact with it as if they’re texting a friend. It’s not that ChatGPT does anything fundamentally new. It just does it in a way that normal people can access and thus be blown away by.
Allowing for that level of accessibility is not easy. OpenAI’s Sam Altman shared on Twitter that opening up ChatGPT to the general public resulted in some eye-watering computing costs.
Each chat sent to the system reportedly costs “low digit cents” to process. If 1 million people are using the system, OpenAI is probably sinking several hundred thousand dollars a day into keeping ChatGPT operating, with no immediate business case.
Academic researchers and the other people who’ve made the fundamental breakthroughs behind technologies like GPT-3 simply couldn’t afford that. Without the resources to put into making the technology accessible, the tech can’t make it out into the real world.
Looking Backwards
That dynamic has existed for a long time. The history of science is littered with situations where a researcher developed a breakthrough idea, only to be shoved aside by the entrepreneur or visionary who made that idea easily accessible to the public.
We think of Thomas Edison as the inventor of the lightbulb. But really, it was inventors like Vasilij Petrov, Thomas Wright, and Joseph Swan who made the first lightbulbs. Edison’s genius was in electrifying public buildings, building power plants and wiring, and otherwise making the technology visible and accessible to everyday people.
Edison probably lost a ton of money pulling off dramatic stunts, like building a full power plant just to light the home of wealthy financier JP Morgan, the New York Stock Exchange, and several newspapers’ headquarters.
But once people saw the advantage of electric lights, they needed them for their own homes. By making the technology accessible and easily visible — even at a great personal cost — Edison had unlocked a market for it that proved wildly profitable.
In the AI space, something similar is now brewing. ChatGPT shows that developing breakthrough technology won’t lead to cultural change if that technology is restricted to the lab or even to the high-powered servers of B2B clients.
For technology as groundbreaking as AI to make it out into the real world, it has to engage the imaginations of everyday people. People need to play with it directly — and see its revolutionary power — before the tech will truly impact the world.
As the field continues to evolve, it will be the companies who continue to make the technology widely accessible — even at tremendous cost — that will ultimately succeed.
OpenAI had better get used to those eye-watering compute fees. They’re building a revolution, and revolutions are expensive.