Few tech developments ever have had the sudden impact of the ChatGPT generative artificial intelligence chat software, released last week. For many, using it had the quality of experiencing a transition into a new era of tech. That’s because this evolution of software developed in recent years by OpenAI, a company formed by Elon Musk and others in 2015, performs such compelling and convincing verbal pyrotechnics based on simple written commands.
I copied the paragraph above into ChatGPT, first simply writing “Here is the first paragraph, please write the next paragraph of this article:”. It spit the following back, in about 5 seconds, verbatim:
“One of the most notable features of ChatGPT is its ability to generate responses in real-time, making conversations with it feel natural and seamless. This has led many to speculate about the potential uses for such a technology, from customer service chatbots to virtual assistants and beyond. As with any new technology, however, there are also concerns about the potential negative consequences of using ChatGPT, such as the potential for misuse or abuse. As the developers continue to refine and improve the software, it will be interesting to see how it is embraced and integrated into our daily lives.”
So yeah. It definitely can help write just about anything. And that’s just the tip of the proverbial iceberg. Or so I think. I asked if it agrees and it replied:
“It is difficult to say whether ChatGPT is the top of the proverbial iceberg. It is a significant development in the field of artificial intelligence and has the potential to have a major impact on the tech industry. However, the field of AI is constantly evolving and there may be other developments in the future that are equally or even more significant. Only time will tell whether ChatGPT is a precursor to even greater advances in AI.”
I mean really! This is important. If you read a lot of tech media, you’ve already heard a lot about ChatGPT, because just about every professional observer, like me, feels compelled to comment. Those articles are uniformly, and properly, ga-ga.
We had a Techonomy prep call for our upcoming Climate conference (March 28–Menlo Park) with the great entrepreneur and tech inventor Bill Gross of Idealab. Just a mention of ChatGPT sent him rhapsodizing: “Last weekend ChatGPT completely blew my mind. I felt like I was witnessing a combination of the Wright brothers and the nuclear bomb–a transformation that changes the world. It just felt so seismic.” One prompt Gross gave it whose response amazed him: “Tell me about the trials of being an entrepreneur, and make it to the words and tune of Bohemian Rhapsody.”
One of the most frequently discussed topics is what this all means for Google. That’s because for many queries or inputs, ChatGPT offers an instant answer that at least sounds more useful and pertinent than a comparable request to Google. And by contrast, it does not require you to click further links or devote any more of your own potentially limited intelligence and time. Alphabet and Google are not caught unawares, however. That company, too, has done extraordinary work on AI systems at its DeepMind subsidiary, and will surely release some kind of AI chat search tool in the not distant future. (Alex Kantrowitz explains in detail.) It’s obvious, once one uses ChatGPT, that this is, in some form, the future of what we now call “search” but which will likely in future go by another name. Maybe we’ll just call it “answers,” or “digital help.”
The category of generative artificial intelligence and OpenAI already made waves earlier this year–though did not lead observers to such total immediate analytical apoplexy–with another experimental product called Dall-e, which instantly crafts illustrations in response to verbal prompts. Our own intrepid human columnist Robin Raskin explored it in September.
One already-widely-noted problem with ChatGPT is that while its answers are uniformly fluid and consistently impressive, it is often wrong. Here is an amusing example, with my query and the software’s reply.
Aside from the obvious pleasures this reply induced, I was struck to learn that “chucking” was in fact “burrowing,” and felt satisfied that not only had I prompted a great response from ChatGPT, but I’d actually learned something. However, a brief reflection induced doubts. Hey, isn’t chucking really throwing? Yes, a Google search revealed. So here is exactly the sort of blind alley this new tool can lead us down.
Many observers have noted that what such systems need is some form of attribution. Unlike Google, ChatGPT gives no indication of where it got all these useful ideas. Obviously they all come from web scraping and online information analysis. But from where? Such systems will not be able to go commercial until they solve that problem. Otherwise they’ll be a legal minefield for their purveyors.
There’s also the potential for new forms of plagiarism, intended or inadvertent. One widely-speculated-upon risk this now creates is that the high school paper is a goner. Maybe in college, too. Teachers are going to need a method to determine if a student met the assignment deadline merely by idly inputting a request of an AI. I wonder if perhaps a universal search system shouldn’t be developed so that anyone’s ChatGPT answers can be searched by anyone else. That way a teacher could copy a student’s random sentence and see where it came from.
Unquestionably, from here on, originality is going to be routinely suspect.
But there are innumerable applications of this technology where originality is unnecessary. At a holiday party this week I met a savvy corporate technologist at a New York media company who already routinely uses ChatGPT to help him communicate inside his company. He feeds in Slack messages from colleagues, noting roughly what he’s aiming to say, and gets suggested replies. More often than not, he says he can use the result with little alteration. He also described needing to write a careful recommendation to his boss about giving an employee a promotion. He gave ChatGPT a casual outline of the sorts of things he needed to say, without agonizing over his wording or tone. The software prepared him a complete two-page memo which with only slight tweaking he forwarded to his boss. He says this saved him as much as an hour.
Some more hyperbolic commentators, like the redoubtable Chamath Palihapitiya, predicted on Twitter that the job of writer may not be as reliable a source of income in the future. I reject that, just as I’m confident that my illustrator daughter is likely to keep getting great assignments from magazines for the foreseeable future, despite Dall-e.
I asked Robin Raskin her opinion, and she wrote back: “Generative AI will probably be more disruptive than crypto and the metaverse combined. People will definitely be out of jobs. You can generate legal contracts, requests for proposals, exhibit plans, ad campaigns–basically anything you can think of. But as always, jobs shift. Doing a good job with generative AI is a skill set that will be needed but that few people have right now. I can see companies having a Chief Generative AI officer in the future.”
Below for your enjoyment are a few additional outputs from this amazing system. Anybody can use it, if it’s not overloaded, as sometimes happens. Go and try it yourself!