IRL to shut down after faking 19M users, ZIRP fraud, Databricks acquires MosaicML for $1.3B | E1768

Episode Summary

Executive Summary: The episode centers on three big themes: exposed fraud in startup/social media metrics, how AI is reshaping search, work, and products, and a broader warning about reckless behavior fueled by wealth and status. Jason and Vinny argue that AI will automate repetitive knowledge work, open-source models may win, and founders should focus on data and product value over premature monetization.

Main Topics: IRL shutdown and metric fraud in startups (Priority: 5/5): The hosts dissect the collapse of IRL, whose board investigation found that 95% of its 20 million monthly active users were fake. They use the case to discuss how investor incentives, weak identity verification, and inflated KPI definitions can enable massive fraud. Identity, bots, and platform incentives (Priority: 5/5): Vinny argues that social platforms tolerate bots because growth metrics and ad revenue are rewarded internally. He contrasts the lack of identity standards in consumer platforms with stricter bank KYC/AML rules and calls for anonymous-but-unique digital identity systems. AI as a platform shift in search and work (Priority: 5/5): They discuss generative AI replacing link-based search with direct answers, reducing friction for users and threatening jobs in tagging, support, and other repetitive knowledge work. Jason gives examples from ChatGPT, Google’s AI search, and his own workflow. Open source vs. giant incumbents in AI (Priority: 4/5): The conversation around Databricks’ acquisition of MosaicML expands into whether open source or major tech platforms will dominate AI. Vinny argues innovation is moving toward an open-source-led stack, with incumbents likely to build interfaces and products on top. AI product strategy and premature monetization (Priority: 4/5): Jason and Vinny argue that early AI startups should prioritize usage, data collection, and product fit rather than monetization too soon. They compare this to the early days of Google search, where the right business model emerged after scale and usage patterns were understood. Risk, status, and the submarine/Titanic tragedy (Priority: 4/5): The latter portion shifts to a moral discussion about wealthy people taking extreme risks for status or dopamine, using the Titanic submersible tragedy and other dangerous stunts as examples. Jason argues adults should protect children from high-risk experiments and question status-driven recklessness. Consumer AI demos: medical, companionship, localization (Priority: 4/5): Vinny demos AI-assisted uses including medical research, Google generative search, photo colorization, voice cloning, translation/localization, and AI companions. These examples illustrate how AI can augment individuals, especially in advisory or emotionally supportive roles.

Key Arguments: The IRL case shows that fake usage can persist until diligence or a board investigation forces a reckoning; it is not a one-off anomaly. Consumer platforms often lack real identity verification, so bots and fake accounts are economically rational and easy to create at scale. Ad platforms and growth KPIs can incentivize companies to tolerate bots because inflated user counts help fundraising and ad sales. Generative AI will remove a step from search by giving direct answers instead of blue links, saving users significant time. AI will automate repetitive knowledge tasks such as customer support, tagging, research, and scheduling, shifting profits away from laggards and toward adopters. Open-source AI may ultimately win because a global distributed community can out-innovate closed systems, while incumbents may mostly own the interface layer. Premature monetization of AI products can suppress adoption and reduce the data needed to improve the product and discover the right business model. Wealth and social-media status can encourage increasingly reckless behavior, making some high-risk experiences seem normal or aspirational. AI companions and coaches could be valuable for people who lack access to professional help, as long as they are framed as simulations or support tools rather than substitutes for licensed experts.

Data Points: IRL fake users: 95% - Board investigation found that 95% of the company’s 20 million monthly active users were fake/automated. IRL monthly active users: 20 million - The company had claimed about 20 million monthly active users before shutdown. Estimated fake users: 19 million - If 95% were fake, about 19 million of 20 million users were not real. IRL Series C valuation: $1.17 billion - IRL raised a $170 million Series C at a unicorn valuation backed by SoftBank Vision Fund. IRL Series C funding: $170 million - Funding round discussed in the fraud example. IRL seed round: $2 million at a $10 million post - Early investor context mentioned to show the company’s cap table history. Frank fake users: 4 million - Used as a comparison case of metric fraud in another startup acquisition. MosaicML acquisition price: $1.3 billion - Databricks’ acquisition of MosaicML was discussed as a major AI exit. MosaicML total raised: $34 million - Used to note how efficient the exit was relative to capital raised. Graduation rate for Founder University when charging: 94% - Jason noted completion improved dramatically when participants paid $500 upfront. Graduation rate for Founder University when free: 5% - Used to illustrate how a small fee increases commitment. Support automation claim: Half of customer support tickets instantly - Intercom’s Finn was described as being able to resolve about half of support tickets before reaching a human. Legal/FTC/Jail outcomes: Multiple sentences and guilty pleas - The episode references several fraud cases in the zero-interest-rate era, including FTX, Frank, and others. Potential self-driving communication frequency: 10 times per second - ChatGPT-generated explanation of V2V safety messaging for autonomous/connected vehicles.

Pivotal Quotes: "95% of their 20 million monthly active users were fake." — Jason: The central revelation about IRL’s shutdown and alleged fraud. "The pyramid of trust is broken down. You can't trust anyone." — Vinny: Commentary on the breakdown of trust in startup fundraising and validation. "AI is now... taking a step out of the process in a big way and they're saving tons of time." — Vinny: Explanation of why generative search will replace traditional link-based search behavior.

Implications: The episode suggests a near-term shakeout: fraud will face more scrutiny, AI will compress workflows and jobs, and founders should build for usage and trust, not vanity metrics or premature revenue extraction.