The odds are that you’ve seen the buzz about ChatGPT, a large language model trained by OpenAI to generate human-like text based on the input it receives. It’s a variant of GPT-3 (Generative Pretrained Transformer 3), the machine learning model that is considered to be one of the most advanced technologies for generating convincing, high-quality written content.
ChatGPT was specifically designed for conversational text generation, meaning it’s good at generating responses to questions or requests on a wide variety of topics based on the data with which it was trained. Meaning, it’s essentially a super-helpful chatbot. But a chatbot that is far more advanced than any you’ve likely interacted with online.
Its conversational interface makes ChatGPT easy (and fun) to use. Given that OpenAI has released a free version, ChatGPT has attracted more than a million users in just a few days (often exceeding the service’s capacity). And this has resulted in nearly as many opinions about exactly what this technology means for human professions as diverse as writing and coding, for companies (like Google), and for entire industries — like education, media, or marketing.
Granted, these are the kinds of things that some of us have been pondering for years: whether technology will ultimately replace humans or empower humans to achieve more, do more, and be more. And of course, the answer is yes to both to one extent or another (as has always been the case…)
But ChatGPT’s recent release and the fact that so many (business) people are playing with it has made the great debate around AI top of mind for many more people — furthered by the fact that it’s just so damn convincing, even if it’s not particularly good.
Hey, even OpenAI co-founder Sam Altman has said so!
ChatGPT is incredibly limited, but good enough at some things to create a misleading impression of greatness.
— Sam Altman (@sama) December 11, 2022
it’s a mistake to be relying on it for anything important right now. it’s a preview of progress; we have lots of work to do on robustness and truthfulness.
Over the past ten or so days, I’ve asked ChatGPT to write articles, headlines, social media posts, and even my own bio. I’ve had it outline a digital strategy for a client. I’ve challenged it to rewrite many of these in the style of Shakespeare, the Bible, Kurt Vonnegut, Stephen King, and the poet Billy Collins. In almost all cases (unsurprisingly, it totally botched the strategy, instead offering a set of steps I might use to develop the strategy on my own), ChatGPT performed surprisingly well, especially when compared to the kind of (frankly so-so) output I’ve gotten from a number of other AI writing assistants I’ve tried.
Now, don’t get me wrong: It never rated a 10 out of 10, rarely even a 5 out of 10. Its work (and words) lacked nuance and personality. It excels with facts while struggling with opinion, although even ostensibly factual information contained obvious errors, (as OpenAI notes on their site) is prone to bias, and doesn’t incorporate anything that has happened since the system was last trained in 2021.
Overall though, ChatGPT comes across as surprisingly human in its ability to string together information into something that loosely resembles the written expression of human thought. This fact alone makes anyone arguing that they don’t feel even a bit at risk in their work seem like Luddites, especially when you consider how far this type of AI has come in these earliest of days.
This isn’t to say a dose of healthy skepticism is not warranted. Because on the flip side, we have the system’s obvious drawbacks (noted above) and the reality that any large learning model might put the world’s information at an AI’s virtual fingertips but that this type of artificial intelligence does not, cannot, and quite probably never will possess true expertise or have any actual understanding of the material with which it works. These facts make the breathless raves, pie-eyed wonder, and premature prognostications come across as unadulterated hype.
After all, ChatGPT is indeed weak AI. I don’t mean this as a pejorative. ChatGPT is weak in the technical sense — trained in and focused on a specific, narrow area. And weak AI is already all around us. It’s in our Google Maps and Uber app, in Siri and Spotify shuffle, and in the infernal social network algorithms that are as likely to connect us with our community as they are to feed the political rage machine.
Yet none of these use cases feel magical, even if they once did. ChatGPT does (feel magical, that is), possibly because it engenders a sense of play. This distinction lies in its interface more so than in its intelligence, or in the ‘fingers on the keyboard’ nature of the interaction and the genie-like ‘your wish is my command’ interpretation of every prompt. And — in contrast to the ephemeral nature of the responses from a Siri or Alexa-like voice assistant — the tangible nature of its output. You wanted an article, you got one. You asked for a summary, it’s right here in black-and-white bullets (just like you specified).
You might make a similar argument about generative image generators like DALL-E 2 (also from OpenAI) or Midjourney, both of which are also enjoying their magical moments right now. And you’d be right, to a large extent — they’re certainly part of the same pattern. But ChatGPT seems to hit closer to home for more people, partly because writing is such a routine task for so many and partly because the conversational interface hides the technical quirks of the process behind a layer of human-like chit-chat that feels far more natural.
And so, I wonder:
Is the release of ChatGPT a tipping point moment in artificial intelligence? Will it result in many more millions of mainstream a-ha moments about what it means to engage with and be augmented by intelligent machines? Are we getting just a peek through the door to unlimited possibilities for the future of work and the future of life? Of the rapidly changing relationship between humans and machines?
In some ways, it reminds me of the first time I fired up a web browser. It was awful — from the screeeeeee of the modem to the slow-motion load of the clunkiest website you could ever imagine — but still, you got the feeling the world had changed forever.
In some ways, it reminds me of the first time I tried Twitter (c’mon Elon, don’t f— it up, man!), of that seachange moment when the web went real-time even if hardly anyone outside of social media wonks was there at the time. Or even the first time I logged into Second Life or, so many years later, first strapped on an Oculus headset as part of a conference demo.
ChatGPT is surprisingly good.
And also pretty awful.
And it just might be giving us a glimpse of the future of work and the future of the world.
