Many chatbots have come and gone since the 1960s, some without much fanfare. But OpenAI’s ChatGPT — the conversational interface that OpenAI put on its GPT-3.5 large language model — hit one million users in only five days in December 2022 and its release is already being heralded as a watershed moment for technology. It’s taken over the internet and the media in the weeks since and has propelled generative AI into the public consciousness (which we at Jasper are thankful for).
I was curious about what makes this specific chatbot such a massive success compared to everything that came before it. So I asked Suhail Nimji, Jasper’s VP of business development, and Jeremy Crane, Jasper’s VP of product.
In short, they said: the fact that Chat GPT is a powerful, multi-faceted LLM built on a chat interface, which we have a strong societal preconditioning to use based on decades of interactions with chatbots and conversation-based messaging systems. All this followed by GPT's implications around the way command-and-response functions will fundamentally change how we interact with computers moving forward.
Now let’s unpack that mouthful.
What’s the Deal With Chat Interfaces?
We’ve collectively been using chat-based functionalities for a long time. We watched the clunky (sometimes creepy) chat rooms of the 1990s evolve to AOL Instant Messenger in the early 2000s. This was also around the time that AI chatbots for customer service and marketing started to grow in popularity across the web. From this moment grew Facebook Messenger and iMessage in the early 2010s. Now there’s GPT and whatever else the kids are using to communicate. But in short, chat-based communication with both bots and humans has grown increasingly ubiquitous in our everyday lives.
“Chat interfaces are the easiest form of communication we are accustomed to today,” said Nimji, “We have been trained and normalized to use chatbots when it comes to support and asking questions. I think this is one of the reasons why the ChatGPT interface took off virally.
“And the derivative of [ChatGPT’s] chatbot was its ability to be plumbed with knowledge that propelled it to go viral. It wasn’t about ChatGPT as much as it was about the underlying model, which was fitted to the existing chat interfaces that we have been conditioned and normalized against.”
The general public now has a sense of how an incredibly powerful LLM, wrapped in an intuitive system everyone is familiar with, can be a marvelous tool for both generating content and searching the web. And there’s no turning back. In fact the popularity of these systems means, “We’re on the cusp of a big reshaping in how we interact with software,” said Crane.
But what will that reshaping look like?
The Future of Command and Response
“Traditional software up to this point is based on a command-and-response relationship: I click a button and it does the thing that it’s programmed to do,” said Crane. “Artificial intelligence is getting closer to how you would have a relationship with another human or a real assistant. You can now have conversations with software, saying: ‘Hey, let’s write a blog post about this thing. Let’s get a post up on social media. Ok, now let’s write a post that’s a bit more friendly.’ This is very different from how conversing with software was done previously.”
Ultimately, this wave of successful chat interfaces on LLMs is a massive leap forward for how human-to-computer interactions will grow moving forward.
“We can hypothesize that AI will be there to assist wherever a human is entering data, generating a complex email response, saying they want to see a discounted cash flow, clean up their desktop, organize their calendar and so on,” said Nimji. “The chatbot interface was only the beginning.”
More functionality with fewer prompts is the future we’re headed toward. But because this is the beginning, that functionality is still in its infancy and the technology is imperfect. Models are still susceptible to bias, hallucinations, and nonsensical outputs presented as authoritative truths. However, with time, the LLMs will evolve to be more efficient and less erroneous. The interfaces built on top of them will develop as well based on their functionality and user preferences.
"Artificial intelligence is getting closer to how you would have a relationship with another human or a real assistant."
Use cases will also get more nuanced and reach a wider audience. Besides marketers and content creators, many businesses today have trouble seeing generative AI as a viable tool because it’s tough to generate consistently reliable results. Early adopters can see a positive impact to their operations because they know how to build the perfect prompts to achieve desired outcomes. But things will really take off once it’s easier for both businesses and the masses to consistently create repeatable outcomes.
Getting everyone to see their desired outputs with these tools will take some additional conditioning, however.
“I think people will be conditioned like this over the coming months and years,” said Nimji. “It’ll require relearning how we answer emails through commands, which will be a natural process. There already exist tools like Jasper’s Chrome extension, where you can write a command like ‘respond to an email’ and take the chatbot experience across multiple applications, which is a natural derivative of where we’re going [collectively].”
Nimji believes that generative AI will usher in a new wave of tech stacks. They will allow gen AI to integrate with the interfaces and functionalities of generally every application at every company, including legacy organizations.
“Generative AI and NLP foundational models have the power to truly automate our lives for the better,” said Nimji. “And I think that’s where the industry is headed in the next 12 to 24 months.”
Internet Searches Won’t Be the Same
It’s not just automating complex tasks that will change in the near future, but how we search the internet.
Google has been receiving criticism in recent years for the diminishing quality of its search results and its lack of innovation within this realm. Unfortunately, most of us are now accustomed to Google Search sending us on ad-infused fetch quests for information through multiple sites to piece together the answers to our queries. Some say Alphabet Inc.’s globally-dominant search tool has been overdue for a shake up.
A couple of weeks after ChatGPT users started getting detailed, (mostly) accurate, ad-free results to their questions in a clear and concise way, Google called a “code red”. Google’s CEO Sundar Pichai immediately reassigned personnel to answer the threat that OpenAI’s implementation poses for its lucrative search tool. That threat got even more real just a few days into 2023, when it was reported that Microsoft planned to use its 2019 investment of $1 billion in OpenAI to infuse its search engine Bing with the power of ChatGPT.
"This is the natural evolution of how we interact with data."
“It makes a lot of sense why people are making these broad assumptions that this is a Google competitor,” said Nimji. “This is the natural evolution of how we interact with data. Chatbots normalized chat interfaces. Google normalized search inference via commands and prompts. So there’s a natural evolution into the idea that if the power of a chat or foundational model was implemented to query results like Google, it’s a natural conclusion for us to jump to that.”
Crane concurred that this is likely the future of search. And since it’s been evidenced by the public’s embrace of this technology and Google being noticeably stirred by it, search will be one of the first things to change in the wake of conversationally-infused LLMs.
“It’s somewhat ironic that the search model will, in many ways, be returning to one of the original models of search,” said Crane. “Remember Ask Jeeves, which rebranded to Ask? The model there was an assistant-based search tool. It wasn’t able to deliver on its promise and was soon overwhelmed by Google and its approach, which is the standard now. However, that desire for more human interactions with software remains and now we have the tools [the models] to actually deliver on the promise.”
Crane and Nimji agree that Google will respond with a gen AI-driven adjustment of its own. Former Google engineers built the technology that ChatGPT is rooted in and Alphabet's acquisition of Deep Mind in 2014 means they’ve got a lot of skin in this game already. It’s just more difficult for a massive player like Google to implement what could be an imperfect AI system, which may have biases and hallucinations, when it has far more regulatory oversight than a startup. But that hesitation is bound to end soon.
“We can assume that the sleeping giant will no longer be in slumber,” said Nimji. “And when Google moves, it will make a very loud noise. We can hypothesize that we will see this show up in YouTube and other Google Suite products like Docs, Slides, and its calendar.”
Waiting for What Comes Next
OpenAI did an amazing job democratizing generative AI and showcasing its power to the entire world. There will be more iterations of technology like ChatGPT over time and holistically, they will evolve how we interact with computers in huge ways.
When asked what an ideal future-state version of these chat-based, command-and-response LLMs looks like, Crane said it’s impossible to predict what “perfect” is because use cases and user preferences are always changing.
“But I think it’s about making sure that our north star perspective is kept in mind when weighing decisions around building products and investing in technology,” Crane continued. “That way, we can think about the long-term goal of where we want to be. And we want to live in a world where generative AI tools are not just for automatically generating content, but unlocking ideas.
“I love stories like a bookstore owner generating more content to stand out. They can spend less time writing blogs or doing book reviews by leveraging these tools to grow their business in the era where content drives ideas and more business. It's inspiring what we can accomplish with these tools.”
Crane believes a major part of following that north star will be fine-tuning models and their interfaces so they’re always customer-first. The more that people are exposed to AI through implementations like GPT and Jasper, the more we can define the nuances and distinct use cases of different audiences, then cater to them. Eventually, whole industries will be built upon adapting the baselines of AI to different needs. And with products like that, our only limitations will be our imaginations.
“The world as we know it has fundamentally changed,” said Nimji.