Veronica Belmont ’04, Technology Evangelist
WHAT’S YOUR BIG IDEA?
It’s a personal LLM—essentially, a version of a Large Language Model like ChatGPT, but based and trained on your own personal information, history, interests, etc.
WHY IS IT IMPORTANT?
A personal Siri or Alexa on steroids? The potential could be “pretty huge.”
Veronica Belmont ’04 has just about had it with Siri. “She Who Must Not Be Named,” as she calls her, doesn’t know anything about what Belmont needs, and half the time, Siri (or Alexa, or Cortana) can’t give a satisfying answer or pull up useful information. What Belmont dreams of is a personal assistant that is functionally intelligent and truly personal—a large language model (think ChatGPT) that just…gets you.
“I talk to ChatGPT…pretty regularly, and I feel so much better about those interactions and so much more confident in the responses [than those of Siri],” said technology evangelist Belmont, who is based in British Columbia and hosts the sci-fi and fantasy podcast Sword and Laser.
Now imagine if ChatGPT lived only on your phone within a closed system, and only learned from information you willingly provided it—protecting your privacy. The potential could be “pretty huge.”
It would be important to limit the model to only operate on a personal device, Belmont said, having only a one-way door from the internet to intake information in real time, while keeping personal data secure. (Belmont said while she has no insider knowledge, she has “no doubt” that tech companies already are working on this concept.)
Here’s how a day with Belmont’s dream PA, “Jake,” (“because I’m tired of female names”) might go: Belmont gets up and asks Jake what’s on the agenda for the day. He gives her a rundown, and she spots an hour between a meeting and lunch to prepare for a big interview she has coming up.
“Jake could—understanding what I’ve researched already, understanding my background, understanding things that I’ve read online recently and already know[ing] what I know for the most part—now start to recommend new sources that maybe I want to check out, perhaps some books that I could read or a few chapters that I could get in for that hour-long break,” Belmont said.
“You can imagine how advanced and how specific and personalized these types of interactions could be moving forward,” she said.
With any AI interface, the question becomes: How much of our personal data do we want this thing to have, and where is the line between helpful and creepy? Belmont acknowledges that privacy issues could be a concern for many.
The key is to always have a “human in the loop.”
“While we do not have the capability of processing extreme amounts of data, and they do, we still need to be there to have the judgment, to have the understanding of how people work, about how that industry works,” she said. “And having that work together is what makes it extremely valuable.”
