- Open AI, the AI company behind the AI art generator DALL·E, released the viral bot Chat GPT.
- The bot, which drew more than 1 million users soon after its launch, is attracting more investors to generative AI.
- If you haven't followed the GPT craze, here's how it works and which experiments are using it to replace humans.
Since Open AI released its blockbuster bot Chat GPT in November, the tool has sparked ongoing casual experiments, including some by Insider reporters trying to simulate news stories or message potential dates.
To older millennials who grew up with IRC chat rooms — a text instant message system — the personal tone of conversations with the bot can evoke the experience of chatting online. But Chat GPT, the latest in technology known as "large language model tools," doesn't speak with sentience and doesn't "think" the way people do.
That means that even though Chat GPT can explain quantum physics or write a poem on command, a full AI takeover is not imminent, according to experts.
"There's a saying that an infinite number of monkeys will eventually give you Shakespeare," said Matthew Sag, a law professor at Emory University who studies copyright implications for training and using large language models like Chat GPT.
"There's a large number of monkeys here, giving you things that are impressive — but there is intrinsically a difference between the way that humans produce language, and the way that large language models do it," he said.
Chat bots like GPT are powered by large amounts of data and computing techniques to make predictions about stringing words together in a meaningful way. They not only tap into a vast amount vocabulary and information, but also understand words in context. This helps them mimic speech patterns while dispatching an encyclopedic knowledge.
Other tech companies like Google and Meta have developed their own large language model tools, which use programs that respond to human prompts and devise sophisticated responses. Open AI, in a revolutionary move, also created a user interface that is letting the general public experiment with it directly.
Some recent efforts to use chat bots for real-world service have proved troubling — with odd results. The mental health company Koko came under fire this month after its founder wrote about how the company used GPT-3 in an experiment to reply to users.
Koko co-founder Rob Morris hastened to clarify on Twitter that users weren't speaking directly to a chat bot, but that AI was used to "help craft" responses.
The founder of the controversial DoNotPay service, which claims its GPT-3 driven chat bot helps users resolve customer service disputes, also said an AI "lawyer" would advise defendants in actual courtroom traffic cases in real time.
Other researchers seem to be taking more measured approaches with generative AI tools. Daniel Linna Jr., a professor at Northwestern University who works with the non-profit Lawyers' Committee for Better Housing, researches the effectiveness of technology in the law. He told Insider he's helping to experiment with a chat bot called "Rentervention," which is meant to support tenants.
The bot currently uses technology like Google Dialogueflow, another large language model tool. Linna said he's experimenting with Chat GPT to help "Rentervention" come up with better responses and draft more detailed letters, while gauging its limitations.
"I think there's so much hype around Chat GPT, and tools like this have potential," said Linna. "But it can't do everything — it's not magic."
Open AI has acknowledged as much, explaining on its own website that "ChatGPT sometimes writes plausible-sounding but incorrect or nonsensical answers."