ChatGPT was not something anyone expected. Not even OpenAI. ChatGPT was launched in November, as a “research preview”, before it became the fastest-growing consumer app ever.
The blog post that announced ChatGPT was a humorous case study in underselling. ChatGPT is a sibling model of InstructGPT. It is trained to respond to a prompt and follow the instructions. ChatGPT is a new tool that we are eager to use to gather feedback from users and find out about its strengths and limitations.” This is the entire pitch! There was no giddy talk about fundamentally changing our interaction with technology. Not even a single line about how awesome it is. This was a preview.
Now, just four months later, ChatGPT looks to be changing the way we view technology. It could be that ChatGPT is going to change the way we think about technology. The future of technology isn’t about whiz-bang interfaces and the metaverse. It’s simply “typing commands into text boxes on your computer.” The command-line is back, but it’s smarter than ever.
Generative AI is moving in two different directions. The first is more infrastructural and adds new capabilities and tools to existing products. GPT-4 and Google’s LaMDA large language models will help you write emails, memos, and photos. They’ll also automatically clean up your slides and fix any errors in your spreadsheets.Do you remember when everyone, even Pizza Hut was using chatbots to communicate with customers?
This is the same path AI has taken for many years. Over the past few years, Google has integrated all types of AI into its products. Salesforce and other companies have also built strong AI research programs. These models can be costly to build, train, query, and could have a significant impact on corporate productivity. AI enhancements to products you already use are a huge business.
It was less obvious that the other direction of AI, where interaction with the AI is a consumer product, was being developed. It’s obvious now: Who doesn’t want to chat to a robot who knows everything about movies, recipes, and where to go in Tokyo? And if I say the right things, it might even try to get out with me. Before ChatGPT was a huge success, and before Bing & Bard tried to create their own products from it, I wouldn’t have predicted that typing into chat windows would become the next big thing in user interfaces.
This is, in a sense, a return to an old idea
This is, in a sense, a return to an old idea. Since long time, most users onlyinteracted directly with computers via a blank screen. The command line was the way you instructed the machine to do what you wanted. ChatGPT is a lotof computers, so they aren’t right at your desk. But you get the idea.
Then, we created better interfaces! Command line had a problem. You needed to know exactly how to enter commands and when to expect the computer to respond. It was easier to point and click on large icons. Also, it was easier to show people what the computer could accomplish through icons and pictures. The command line was replaced by the graphical user interface. Today, the GUI reigns supreme.
Although developers have never stopped working to make chat UI more usable, they are not stopping trying. WhatsApp is an example of this: The company spent years trying to figure how chat users can interact with businesses. Allo was one of Google’s failed messaging apps. It hoped that you would interact with an artificial intelligence assistant in chats with friends. In the early days of chatbot hype around 2016, a lot smart people believed that messaging apps were the future.
The messaging interface, or “conversational Ai,” is very appealing. It’s a way to keep in touch with people you care about. They are also a place where we spend a lot time and effort. Although you may not be able to navigate the Uber app’s hidden areas or find your Southwest frequent flier number, texting these words to this number is something almost everyone understands. Messaging can make life easier in a market that doesn’t want apps or mobile.
Although messaging may not be the most sophisticated interface, it is the one that’s most extensible. Slack is an example of this. You might think of it as a chat application, but you can embed links and editable documents in the back-and-forth interface. Interactive polls, informational robots, and interactive polls are just some examples. WeChat is a platform that is basically the entire internet rolled into one messaging app. You can start with messages and travel to many places.
Many of these tools fall short in many ways. Chat is great for quick information exchanges, such as business hours. Ask a question and get an answer. What if you browse a catalog and get a bunch of messages? No. Do you want to buy a plane ticket that has a thousand messages back-and-forth between? Hard pass. It is no different from voice assistants. And God help you if it has ever been attempted to buy simple items with Alexa. Charmin would say “three .'”)” for most complex things. A dedicated, visual UI is better than a message window.
ChatGPT, Bard and Bing can make things very complicated. Although these models are intelligent and collaborative, you still need to know what to ask for and how to do it. Although it seems absurd, the idea of a prompt engine, the person who can help you get the perfect Javascript from ChatGPT or get Stable Diffusion to create the perfect image, is an essential part of the equation. This is no different from the early computer age, when very few people could tell computers what to do. Marketplaces exist where you can sell great prompts. There are prompt gurus who are experts in prompts. I believe Stanford will soon be offering a Prompt Engineering major.
Generative AI’s remarkable quality is its ability to do nearly anything. This is the problem. What do you do when you know you can do everything? You don’t know where to start. You have only a blinking cursor to show you how it works. How can you learn how to use it? These companies may eventually develop interactive, visual tools that allow people to see what they can do, and how it works. ChatGPT’s new plugins system is another reason to be on the lookout. While it is quite simple for now, it could soon expand the options you have in your chat window. The best thing they have right now is to give you some suggestions on what you might type.
AI was to become a feature. It’s now the product. The text box is now back. Again, messaging is the interface.
