It's very clear that nobody anticipated ChatGPT's arrival. even OpenAI cannot. Before it made history as the consumer app with the fastest growth rate, before it popularized the term "generative pre-trained transformers," and before every business you could imagine was rushing to implement its underlying paradigm, ChatGPT debuted in November as a "research preview."
A humorous case study in underselling may be found in the blog article that announced ChatGPT. "ChatGPT is a sibling model of InstructGPT, which is trained to carry out a prompt's instruction and offer a thorough response. We are eager to launch ChatGPT in order to gather user input and understand its advantages and disadvantages. I'm done now! The entire pitch is there! No waxing lyrical about how technology is profoundly altering the way we interact with it, not even a syllable about how awesome it is. Just a research preview, really.
Nevertheless, only four months later, it appears that ChatGPT will actually alter the way we perceive technology. Alter it back, or perhaps that would be more correct. Because of the direction things are headed, the metaverse or glitzy interfaces are not the technology of the future. "Type commands into a text box on your computer" is what it is. The command line is back, but it's much wiser this time.
Indeed, generative AI is moving in two directions at once. First, which adds new tools and capabilities to the things you already use, is considerably more infrastructure-focused. Massive language models, such as GPT-4 and Google's LaMDA, will assist you in writing emails and notes, improve your presentation decks automatically, fix any errors in your spreadsheets, edit your images more effectively than you can, and, in many cases, create your code for you.
The direction Intelligence has been heading in general for a while now, correct? Over the past few years, Google has been incorporating various types of AI into its products, and even businesses like Salesforce have developed robust AI research initiatives. These models could revolutionize business productivity, but they are expensive to develop, expensive to train, and expensive to query. AI improvements in products you already use are and will continue to be a major business, or at least are receiving significant investment.
The other AI trend, in which interacting with the AI turns into a consumer good, was much less apparent. Now that I think about it, it makes perfect sense: who wouldn't want to converse with a robot who is knowledgeable about movies, recipes, and things to do in Tokyo, and who, if I said the right things, might go completely off the rails and attempt to make out with you? But I never would have guessed that typing into a chat window would become the newest big thing in user interfaces before ChatGPT took the world by storm and before Bing and Bard attempted to build their own products out of it.
This is, in a sense, a return to an extremely ancient notion. For a long time, the majority of users only used computers to type commands into vacant screens using the command line. (Yes, there are many machines in ChatGPT, and they are not on your desk, but you get the concept.)
But then, strangely enough, we created better user interfaces! The challenge with using the command line was that you had to be extremely precise about what to enter and in what order to make the computer do something. Large icons made pointing and selecting much simpler, and using images and icons made it much simpler to explain to others what the computer could do. The graphical user interface (GUI) replaced the command line and continues to be the standard.
However, developers never gave up attempting to make the chat UI function. A good example is WhatsApp, which has spent years attempting to understand how customers can use chat to engage with businesses. One of Google's numerous unsuccessful messaging apps, Allo, hoped you would talk to an AI assistant while chatting with your pals. Many very intelligent people believed that messaging applications were the future of everything during the initial chatbot hype, which peaked around 2016.
The messaging interface, or "conversational Intelligence," has a certain allure. It all begins with the fact that we all know how to use messaging apps, which are where we spend a lot of time and effort because they're how we stay in contact with the people we care about most. While you might not know how to locate your frequent flyer number in the Southwest app or how to navigate the Uber app's hidden areas, almost everyone can grasp the concept of "text these words to this number." Messaging can greatly simplify experiences in a market where users don't want to acquire apps and mobile websites are still generally bad.
Despite not being the most sophisticated UI, messaging may be the most expandable. Consider Slack: you presumably think of it as a chat tool, but you can embed links, editable documents, interactive polls, educational bots, and much more in that back-and-forth interface. WeChat is well known for being a complete platform—basically, the entire internet—condensed into a messaging program. Starting with texting allows you to travel far.
However, so many of these instruments experience the same problems. Chat is ideal for quick information exchanges, like during business hours; you can pose a question and receive a response. But looking through an inventory in the form of messages? Not at all. Purchasing an airline ticket after exchanging 1,000 messages? No, thanks. It's the same as voice assistants, and if you've ever attempted to use Alexa to even make a simple purchase, I pray for your safety. ("Say "three" for Charmin.") A specialized visual user interface is much more effective than a messaging window for the majority of complex tasks.
And things quickly become complex when talking about ChatGPT, Bard, Bing, and the rest. These models are intelligent and cooperative, but in order to get what you want, you still need to know precisely what to ask for, how to ask, and in what order. The concept of a "prompt engineer," a person you hire to know exactly how to persuade Stable Diffusion to produce the ideal image or persuade ChatGPT to generate the ideal Javascript, may seem absurd, but it is an absolutely essential component of the equation.It is comparable to the early days of computing when only a select few were able to instruct a machine what to do. There are already websites where you can purchase and sell really good prompts, prompt experts, and books about prompts. I also assume Stanford is already working on a major in prompt engineering that will be available to everyone in the near future.
Generative AI is amazing because it seems to be capable of almost anything. That is also the root of the issue. What do you do when you have unlimited options? How do you begin? When all you have to look at to see what it's capable of is a flashing cursor, how can you possibly learn how to use it? These businesses may eventually create more engaging, visual tools that aid in people's understanding of what they can do and how everything functions. (This is one reason to keep an eye on ChatGPT's new plug-ins system, which is currently fairly simple but may soon broaden the options available to you in the chat window.)The best notion any of them have right now is to make a few recommendations for what you might type.
The inclusion of Intelligence was planned. The merchandise is now in focus. Thus, the text area is once again present. Once more, the UI is messaging.