12-year-old Bay Area boy trains adults, peers on ChatGPT

With an unassuming smile, 12-year-old Kaz Keller makes a request of the electronic mind behind the new artificial intelligence platform known as ChatGPT.

“Write a short joke in the style of Harpo Marx about frogs,” Kaz said to the program during a recent online training from his home in Mill Valley. Within seconds, an answer appears on the computer screen:

“Why did the frog call his insurance company? Because his car got toad.”

Chuckles of delight come from the audience of about six adults. They have signed up to learn what the fuss is all about with ChatGPT — and how they might be able to use it.

“I am so impressed with your ability,” participant Lucy Salter of Tiburon told Kaz. “It’s just amazing to me.”

Kaz, a seventh-grader at Mill Valley Middle School, and his father, Sam Keller, have found a burgeoning niche in their part-time joint venture helping people navigate the brave new world of AI.

“Never in my wildest dreams did I ever imagine that I would one day watch my 12-year-old son teach people about an artificial intelligence tool that is fundamentally altering human society,” Keller said. “But that’s what I’ve been witnessing at our house.”

In the seven months since ChatGPT was launched by the San Francisco tech company OpenAI — followed by its powerful sequel, GPT-4, in March — father and son have been increasingly in demand to teach seminars at the Mill Valley Community Center, online from home and in person to anyone curious about it.

“I love hearing about technology from all angles,” said Bianca Buckridee, a tech executive from Tiburon and one of the first participants in a Kaz Keller seminar. “‘As a marketer, this helps me think beyond just the business angle.”

“Attending Kaz’s session left me inspired,” Buckridee said. “He really framed up the value of generative AI in a simple sentence: ‘This will change the way we think.’”

GPT, which stands for generative pre-trained transformer, is a type of artificial intelligence that has the ability to search, retrieve, replicate and simulate human expressions of all kinds and, if asked, use the same context, style, sound and voice of the original version.

It might be amusing if people ask the ChatGPT robot to create a recipe for “peppered chocolate chicken medley,” as participants did in one of Kaz’s classes. And then, perhaps, have the robot read the recipe aloud in the voice of the late French cooking guru Julia Child.

But what if the technology can also steal people’s copyrights and identities and impersonate their voices, or allow students to cheat on exams, write essays for college applications or otherwise perform human-like functions? And what about scams?

Sam Altman, the chief executive officer of OpenAI, asked U.S. senators at a Senate Judiciary Committee hearing last week for the government’s help in regulating, licensing and addressing safety concerns for ChatGPT before it’s too late.

“My worst fear is that we, the technology industry, would cause significant harm to the world,” Altman told legislators. “If this technology goes wrong, it can go quite wrong. We want to work with the government to prevent that from happening.”

Altman’s comments came after U.S. Sen. Richard Blumenthal, a Democrat from Connecticut, launched the hearing with opening remarks that sounded like him, and used his voice, but were not Blumenthal. His demonstration gave a glimpse into a possible future where the true reality in the present moment is hard to discern.

“Kaz and I talk frequently about the role and need for regulation of AI,” Keller said. They advise watching a recent video, “The AI Dilemma,” on YouTube to better understand the topic.

Keller said he and his son have incorporated six “key caveats” in their presentation to list the potential dangers of ChatGPT. The dangers could include using it for bad deeds, such as cheating, scams and disinformation.

“It can be inaccurate,” Kaz explains during the seminar. “Just like some humans can’t tell the difference between fake news and real news.”

Also, ChatGPT has “no morality, is not sentient and it does have biases,” Kaz said.

“I don’t think that will change anytime soon,” he said.

The 12-year-old, an engineering buff, has been “exceptionally interested in building things since he was a toddler,” his father said.

Kaz has since been accepted into the Johns Hopkins Center for Talented Youth.

“I often think about a World Economic Forum quote from 2016,” said Buckridee. “‘A projected 65% of children entering grade school will work in jobs that do not exist today, a transformation that will require social and emotional skills such as creativity, initiative and adaptability to navigate.’”

“This is that moment of transformation, and Kaz simply blew me away,” Buckridee said.

According to Kaz, 10% of ChatGPT-generated simulated test takers were able to pass the bar exam. But when GPT-4 came out, 90% of the robot-generated test takers passed. ChatGPT is available for free, while GPT-4 has a cost to access.

“We’ve been struck by how what’s happening in Marin is a microcosm of what’s happening globally,” Keller said.

“We’ve met parents worried that ChatGPT may harm their kids’ education and ability to think,” he added. “We’ve met professionals excited that ChatGPT supercharges their capacity to generate content.”