AI Has Entered the Chat

With ChatGPT sweeping the internet, we turn to two software engineering experts for a lesson in AI chatbot 101.

ChatGPT is coming for your job. … If it feels like all you’ve been hearing about is the rise of artificial intelligence chatbots—especially conversational phenom ChatGPT—you’re not alone. 

The headlines are everywhere, some fascinating, some frightening—think everything from major companies like Snapchat, Spotify and Zillow debuting cool chatbot features to IBM announcing it will pause hiring on jobs it believes can be replaced by AI. And the list of ChatGPT’s impressive abilities grows longer every day. In fact, a new study out of the University of Florida found it can predict the rise and fall of stock prices. Say what? 

AI anxiety is spreading fast, but many are out there still wondering, what’s with all the hype? And, by the way, what exactly is a chatbot? If you’re a little lost in AI translation, we get it—and we’ve got you covered. Here, two tech whizzes help us break down all things ChatGPT—the good, the bad and the downright freaky (we’re looking at you, AI headshots). Brave new world, here we come.

WHAT IT IS “ChatGPT is a conversational AI system that leverages the GPT large language model (LLM) to perform various text-prompted tasks such as information retrieval, email composition and knowledge distillation,” says Dunchadhn Lyons, a Durham-based engineering manager who heads up the AI team at Bay Area video intelligence company Spot AI. Um, GPT? “Generative Pretrained Transformer,” explains Red Hat Senior Principal Software Engineer, Emerging Technologies Erik Erlandson. “The first capability that makes them new and important is in that word ‘generative.’ Large language models like ChatGPT can generate entire bodies of content for you.” 

The second is that “large language models can respond meaningfully to human requests and questions of nearly any kind,” Erlandson continues. Whoa. But why does it matter? “Together, these two capabilities allow humans to interact with software in entirely new ways,” he declares. “A human user can talk to these models essentially as if they were speaking to another human, and furthermore the model can respond in a way that is similar to an actual human.” Adds Lyons, “The wide-ranging tasks that ChatGPT can accomplish makes it a revolutionary-level tech disruptor on the order of the PC or the internet.”

WHO SHOULD USE IT Software application developers, of course, but also regular people just like us. “Although early adopters have mostly been in the tech industry, ChatGPT was designed for anyone and everyone,” says Lyons. “Do you Google things a lot? ChatGPT is for you! Do you write a lot of emails? ChatGPT is for you! Do you have a bunch of random ingredients in your fridge and want a recipe for dinner? ChatGPT is for you!”

HOW TO MASTER IT “In one sense, nobody has mastered large language models yet; data scientists are discovering new capabilities and uses for these models nearly every day,” says Erlandson. However, there are tricks to ensure a successful chatbot experience. “The most important skill is probably the task of constructing the best prompts to get the result you want,” he reveals. “Creating good prompts has become so important that software engineers and data scientists have already given it a name: prompt engineering.” 

Lyons agrees: “By providing as much specific context around what you are trying to do, the underlying model is able to better isolate the necessary information and give you a better answer back with a lower likelihood of hallucination.”

WHAT TO WATCH OUT FOR Wait—hallucination? “When the system doesn’t know how to respond to a prompt it will ‘hallucinate,’ or make up a response that sounds intelligent,” Lyons explains. “Many of these mistakes are harmless or merely annoying,” adds Erlandson. “However, some can have more serious consequences.”

Therefore, proceed with caution, Lyons warns. “In a world of misinformation, it is more important than ever to vet information,” he says. “The same can be said about Google search, but at least you have a source to tie back to reality.”

WHAT TO GEEK OUT OVER Don’t worry, it’s not all bad. “The major pros of AI chatbots are their ability to drastically reduce the time and energy required to complete basic tasks—and their ease of use, which, in turn, democratizes access to the world’s most powerful technology,” says Lyons. In addition to supplying meaningful answers and content in response to human questions, large language models like ChatGPT can also “provide ‘multimodal’ operation,” Erlandson shares—aka “generate images for you based on a text description.” 

WHAT’S NEXT Peter Parker’s uncle said it best in Spider-Man: “With great power comes great responsibility.” Though the opportunity is certainly there for “bad actors to mass generate disinformation,” says Lyons, as long as our leaders continue to promote ethical AI development—something the White House recently pledged—it’s hard to ignore the myriad ways chatbots could benefit society. “If you haven’t tried out ChatGPT, give it a shot the next time you’re trying to look something up,” Lyons offers. “You’ll never search the same way again.”