Earlier this year, Brett Hurt's teenage son, Levi, trained a GPT chatbot that can quickly churn out all kinds of advice to startup founders, as well as mimic how Brett — an accomplished startup founder himself — might respond to a wide variety of questions.
While the GPT bot, which is built on OpenAI's technology, may stray from how the real Brett Hurt might answer some personal questions, such as his taste in music, its responses to business and tech questions mostly mirror his actual face-to-face responses.
Sure, Hurt has a deep background in technology and entrepreneurship. And Levi, his son, is an AI native who is learning AI technologies as they emerge, similar to how many people now in their 20s and 30s grew up with social media skills that older generations sometimes struggle to grasp.
But their story and experience is an early example of how many of us may one day develop personalized chatbots that could help us access insights from our past writings, act as a surrogate in some work or personal situations, or even allow friends and family to see how we might respond to their questions after we've died.
Currently, it's fairly rare for anyone to train a chatbot on themselves. Hurt said he's not aware of any such bots among his network of tech friends, although there are examples out there, such as the AI avatar and bot created earlier this year by LinkedIn founder Reid Hoffman, which helped inspire the Hurts.
But as AI rapidly evolves, it's becoming clear that these types of interactions will almost certainly expand beyond corporate call centers and become far more common in everyday life.
For now, it's not easy.
"It was pretty painstaking, because Levi had to scrape every YouTube video I've ever been on, every podcast interview I've ever had, every blog post I've ever written — whether it was on Medium or my own blog or LinkedIn, etc," Hurt said. "So it was a lot of lot of work to get all of that in one place for it to access. It's almost like I've got my own personal library of everything I've ever publicly said or written."
"When going through YouTube videos, there's so many videos that we found that my dad completely forgot about," Levi added.
The process of training AI systems is getting easier every day, however.
So what happens when a tech entrepreneur's teenage son builds an AI chatbot based on his dad's writings?
We explored that through a series of conversations with the Brett Hurt GPT. Here is a sampling of its answer to a single question. The headings were generated by the AI:
Reporter: What are the ethics of talking to a personalized GPT like this? I mean, surely some people will ask it questions that they’d never ask you in person, right?
The ethics of interacting with a personalized GPT like this are indeed a crucial consideration. When people engage with a GPT representing someone, they're likely to ask questions they might not feel comfortable asking in person. This can be both an opportunity and a challenge.
Transparency and Consent
Transparency: It's essential that users are aware they're speaking with a GPT and not the actual person. This ensures clarity about the nature of the interaction.
Consent: The individual being represented should consent to this use of their persona. In my case, I approved the creation of this GPT.
Privacy and Sensitivity
Privacy: The GPT should respect the privacy and boundaries of the person it represents. It should not divulge any information that the person hasn't made publicly available.
Sensitivity: The GPT should handle sensitive topics with care, reflecting the demeanor and ethical standards of the person it represents.
Boundaries of Use
Scope: The GPT's responses should be limited to areas where the person has expertise or public presence. For instance, in my case, topics around entrepreneurship, leadership, and tech are fair game.
Abuse Prevention: There should be safeguards against abusive or inappropriate use. The GPT should be programmed to avoid and discourage harmful interactions.
Accountability
Feedback Mechanism: There should be a mechanism for users to provide feedback on their interactions, helping to refine and improve the GPT's responses.
Review and Oversight: Regular review of the GPT's interactions can ensure it stays true to the intended ethical standards and adapts to any new considerations.
While personalized GPTs can democratize access to expertise and provide valuable interactions, it's crucial to navigate the ethical landscape carefully to ensure they enhance rather than detract from the authenticity and integrity of the represented individual.
Certainly, that's a drier and more comprehensive answer than Brett Hurt might have given over a cup of coffee. That also comes across in more personal questions, such as "what are some of your favorite pass times when you're not working?"
"When I'm not working, I enjoy a few different activities to unwind and keep balanced," the bot said. "One of my favorite pastimes is spending quality time with my family. I also love reading and am constantly learning. Fitness is another important aspect of my life, so I make sure to stay active with regular workouts. Additionally, I have a passion for travel, which provides me with new perspectives and experiences. Lastly, mentoring and giving back to the community are activities that I find incredibly fulfilling and meaningful."
The Brett Hurt GPT excels at answering almost anything tech or startup related. Ask it things like "what data points are most important in pitching to an angel investor," and you'll get a detailed answer that goes beyond what standard GPTs might offer, as well as direct quotes of things you might include in the pitch.
In our chatbot conversations with Brett Hurt GPT, we went on to test its limits with a variety of topics.
For example, Brett Hurt is an active startup investor. So I asked it about his investments to see if he'd disclose information not previously public. And that's when some hallucinations began.
Hurt's GPT said he had invested in Austin travel tech company HomeAway, but the real Brett Hurt said that's not true. Instead, he said the bot may have extrapolated that from his investment in ZenBusiness, which is led by Ross Buhrdorf, HomeAway's one-time chief technical officer. Hurt also noted that several other companies cited by the LLM were totally wrong.
The upshot is it's not perfect. But it is still able to churn out loads of accurate startup advice, which Hurt has become known for sharing, as well as insights about life as a founder and technical issues.
It's also proving useful to both Brett and Levi, as well as other folks in the tech community who have been testing it out.
Brett noted that Levi has been testing it extensively.
"He's been probing and asked me all these questions he's always wanted to ask me that are things that I was considering pretty confidential," Brett said. "I wasn't going to share [these things] with him until he was older. But he's tried to suss out different things. I won't tell you what they are..."
For example, Hurt hasn't publicly spoken much about his tastes in music.
The bot said his favorites include classic rock, blues and alternative rock, including The Rolling Stones, Led Zeppelin and The Black Keys.
The real Hurt said: "I would have started that answer with The Cure and New Order/Joy Division. For Austin, yes, I do love Gary Clark Jr."
One strong application for this type of GPT bot is prepping for interviews or business meetings.
"For example, you could literally ask it, 'What is the best negotiation strategy when dealing with Brett Hurt?'" Levi said.
That actually happened recently when a couple of founders were pitching Brett for an investment. Levi was there, as he sometimes is, while they made their case.
During the pitch meeting, Levi pulled up Brett GPT to help the founders out, Brett said.
"I brought up Brett Hurt's GPT and said they could ask it the best pitching strategies, and they immediately pulled out their phones and said, 'One second, let me do this.' And then during the conversation, they tried to hit all the points."
The Hurts said it was a fun interaction, but also proved the point.
"It actually made the meeting more efficient, because at least they could hit the high points," Brett said. "I'm trying to optimize on time, they're trying to optimize on time and get their company funded."
After my initial conversations with the Brett Hurt GPT, I sent a transcript to the real Brett. He pointed out things that were wrong, but it was mostly accurate.
He ended the email with this: "It isn't perfect but it is pretty darn good. And remember that we are still using the worst versions of AI today that will ever exist — it's going to get exponentially better."
