The Artificial Intelligence of the Generative Grounded Transformer: this time, it's personal

“It’s almost like an infant,” says Jonathan Bickoff. It’s a refreshingly simple take on one of the most tortured topics currently getting bandied about among La Jolla’s digi-pioneering crowd: AI. Artificial Intelligence: an evolution-leaping phenomenon which, some say, threatens to make mankind a primitive sidekick to technology within the decade.

But here is Bickoff, talking all warm and fuzzy about a new, highly personal evolution of AI. A kind which promises to evolve alongside you. Like any infant’s mind, it arrives empty, until you start filling it up with your experiences and thoughts. The warm, fuzzy part is knowing that those experiences and thoughts will never go away, never get forgotten — and will remain easily accessible in the bargain. “Imagine if you could have the answer to anything you once knew,” reads Personal.ai’s web intro, “or could recall every detail about your conversations, without endless scrolling or searching. Your personal AI is the digital library of you.”

Bickoff says the genesis for all this came back in 2020, before the arrival of all these recent AI companies (they say there are 150+ in San Francisco alone, and maybe three in San Diego). It started with Bickoff’s boss Suman Ganuganti’s last company, called Aira, headquartered in La Jolla and built around the idea of augmenting the lives of blind and low-vision people. Similarly, he says, Personal.ai is about augmenting memory, another frail and fallible human trait. “And memory is obviously something that’s incredibly personal.”

The event that catalyzed this seed of an idea was also incredibly personal: “It happened when [Suman’s] mentor and business partner Larry Bock passed away from pancreatic cancer — pretty rapidly,” Bickoff says, adding that Suman missed not only Bock’s presence — the emotional attachment — but alsohis more intellectual aspects: the knowledge, the advice. “He had the insights. And it occurred to Suman that much of this could be reproduced and trained into an AI model, with each of us developing our own personal AIs, like video diaries” — as opposed to large, external, enterprise AIs which have no memory. “Nobody had done it, and still today, nobody else has done it,” he says. “The only personal language model that exists is ours: Generative Grounded Transformer, as opposed to the most popular language model, which is GPT, Generative Pre-trained Transformer. That middle word is important to note: Large Language Models” — LLMs, the basis for most current AI — “are pre-trained. Whereas our models arrive empty, and so become grounded in up-to-date text, up-to-date data.”

Also, with large language models, the data set is pretty massive. It can take weeks, Bickoff says, or even months, to train a large language model, and the data set is usually public or generic data, things that are out on Twitter, NVIDIA, or Google, Wikipedia etc. “But LLMs are incapable [of building up quick, incremental knowledge of you], or at least not trained on proprietary data, meaning anything that is not on the web. So with Personal.ai, the biggest difference comes from that training data set: our day-to-day lives and creations. Where LLM generic AI is trained on public data, our model is trained on personal data. And we have a three-year start on any competition.”

Personal.ai’s website continues, “Estimates suggest that 20 percent of our workday is spent searching for information, not to mention time spent capturing it. Our goal in our approach is to reduce the time people spend looking for information by creating an ambient recall experience.” Is it going to take off? Already, before its launch on May 11th, Bickoff says his company has a 40,000-strong waiting list of potential customers.

Here’s hoping they can keep all that personal info locked away from the bad guys.