Jan 2, 2024 · 43m · we-live-to-build
What Founding 5 Companies Took From My Kids That I Can't Get Back
0:00 / 0:00
▶ Watch on YouTube →
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
Host Sean Weisbrot interviews MeForYou CEO Keith Smith about launching authorized AI clones for celebrity creators, detailing the technical architecture behind conversational models, scalable parasocial monetization, and hard-earned founder life lessons.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. Sean holds 34.6% of the talking time here. How this is scored →
Sean as informed peer 4.6 Guest teaching 4.3 Guest disagreement 0.4 Sean pushing back 0.2
speaking balance: gold is Sean, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 22:27 Keith teases Sean's podcast persona
Keith playfully challenges Sean's framing of being authentic on-air by quipping that listeners are just getting the sanitized fluffy version of him.
Hardest push from Sean ▶ 30:15 Sean critiques major LLM conversational architectureSean criticizes the industry leaders including ChatGPT, Bard, and Grok for being passive one-sided answer engines rather than capable conversational partners.
Biggest teaching moment ▶ 18:12 Keith explains fan psychology on public versus private dataKeith explains the counterintuitive finding from customer research that superfans idolize the curated public persona rather than wanting raw private DM interactions.
Sean holds their own ▶ 5:55 Sean outlines startup history and AI market entrySean demonstrates deep industry knowledge by detailing his past blockchain ventures, enterprise platform pivots, and how foundational models eliminated the need for hundred-million-dollar training budgets.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Sean as informed peer | Guest teaching | Guest disagreement | Sean pushing back | Why |
|---|---|---|---|---|---|---|
| Keith's Entrepreneurial Roots and Unemployable Mindset | 3 | 2 | 0 | 0 | Keith shares his background from Bible college to becoming unemployable and founding five startups, then pivoting to AI after machine learning work in fintech. Sean provides standard interview prompts. | |
| Sean Weisbrot on Startup Timing and AI Services | 7 | 0 | 0 | 0 | Sean dominates the segment detailing his personal startup journey in blockchain and enterprise platforms, analyzing how ChatGPT altered capital requirements for AI before describing his three current AI service businesses. | |
| Partnering with Creators as Suppliers and Distributors | 3 | 6 | 1 | 0 | Keith breaks down creator economy business models, explaining why SaaS targeting creators fails and why a 50/50 revenue-share model utilizing creators as suppliers and distributors succeeds. | |
| Monetization Models and Virtual Companion Comparison | 5 | 2 | 0 | 0 | Sean relates Keith's concept to his own past blockchain paywall product and notes a high-earning virtual companion case study to validate the business model. | |
| Parasocial Connections and Reality-Grounded AI | 4 | 7 | 1 | 0 | Keith explains the psychological distinction between fictional fantasy bots and reality-grounded personas, then details how testing DM data revealed superfans actually want the polished public persona. | |
| Mid-Roll Call to Action and Audience Support | 4 | 4 | 2 | 1 | Following a mid-roll ad read, Sean reflects on curating his podcast persona, prompting Keith to tease him about getting the fluffy version before discussing scaling intimate engagement. | |
| Dynamic Updating, Memory, and Recency in AI | 5 | 6 | 0 | 0 | Sean asks a sharp technical question regarding real-time training updates compared to static LLMs. Keith explains the dual engineering challenges of long-term memory and recency prioritization. | |
| Creating Dynamic Two-Way Conversations | 7 | 5 | 0 | 1 | Sean pushes on the conversational limits of standard LLMs and identifies Inflection's Pi as the benchmark for two-way dialogue, prompting Keith to validate the observation and explain their prompt engineering. | |
| Siloed AI Memory and Walled Garden Architecture | 6 | 6 | 0 | 0 | Sean poses sophisticated architectural inquiries about cross-AI data sharing and push surveys. Keith details walled-garden privacy, compliance guardrails, and using Keithbot for rapid feedback. | |
| Keith's Hardest Lesson: Family Over Startups | 2 | 5 | 0 | 0 | Sean asks Keith for his single biggest life lesson, leading to an honest reflection on the regret of prioritizing early startup ventures over being present for his children. |
Statements from this episode (11)
Smith: Creator startups treating creators as customers will ultimately underperform
“I think the companies that have not done very well is those that look at creators as a customer base.”
Smith: 50 million of the 200 million global creators work full-time
“There's two hundred million creators worldwide, there's fifty million that are full-time creators.”
Smith: Influencers drive 60 percent of global buying decisions today
“Influencers influence about 60% of the buying decisions in the world these days.”
MeForYou splits 50% of net revenue with creator partners
“We then give a strong financial incentive to our creator partners that we work with because we share 50% of our net revenue back with them.”
Smith: Synchronous audio and video for celebrity AI clones is coming soon
“Today it is Techspace Chat. But in the very near future, it'll be full video with full audio and synchronous and that fundamentally changes the psychological kind of interaction that starts happening there because you know you're interacting with a virtual bei…”
Smith: Fans want relationships with curated public personas, not private identities
“What we figured out quickly was that The relationship that we as fans want with a celebrity or with a creator is actually really with the public persona that we see of this particular creator, creator. And the creators confirmed that for us as well. They said,…”
Smith: ChatGPT's lack of long-term memory prevents genuine parasocial relationships
“That is, it is impossible to build a relationship on top of that, and that is a little bit like what it is to chat, talk with ChatGPT, at least today, and a lot of the other AIs. They just don't have good long-term memory built in”
Smith: AI clones struggle because most creators lack conversational training data
“In most cases with training data is it doesn't have access to a lot of conversational data. That is a little bit different with interviewers that, that have a lot of like podcasts and YouTube. In that case, there's a lot of training data where there's interact…”
Smith: Base ChatGPT speaks like a socially unaware first-year analyst
“ChatGPT is very much like you know, talking to a socially unaware First year analyst who just kind of answers questions and is done. And it's not a very entertaining and interesting dialogue.”
Smith: MeForYou silos user memory strictly within individual AI clone conversations
“As it stands now, that is a siloed a walled garden within that particular conversation that you've had with that particular AI. So what you tell them is what is, is what they know about you.”
Smith: AI models are better at catching rule violations than obeying constraints
“One of the things that's interesting about AI is that it is better as a tattletale than it is at obeying, and so you can say, here's the rules, only do this, and then it spits back something that doesn't obey the rules, and you say, did that obey the rules, an…”