The hottest running app has nothing to do with speed | E2303

Key Concepts

• Gamified Fitness: Using game mechanics (territory capture, leaderboards) to increase user motivation for physical activity.
• Drug Discovery AI: Utilizing machine learning and "world models" to predict biological outcomes and identify drug targets.
• Multimodal Learning: Integrating diverse data types (genetics, blood, brain tissue) into a unified mathematical representation.
• Contrastive Multimodal Learning: An architecture that identifies synergies between different data types to uncover hidden biological signals.
• Virtual Biopsy: Using AI to reconstruct a patient's brain state from non-invasive data (e.g., blood draws).
• Data Moats: The competitive advantage gained through proprietary, hard-to-collect datasets (e.g., human brain tissue).

---

Part 1: Interval – Gamified Running

Main Topics:

• Concept: Interval is a running app that turns local neighborhoods into a global "Turf War" map. Users claim territory by running a perimeter and finishing near their starting point.
• Motivation: The app leverages the "slot machine" nature of smartphones and competitive social dynamics. Users receive notifications when their territory is "stolen," which creates a personal incentive to exercise.
• Business Model: A subscription-based model ($60/year). The company maintains a lean team of five, focusing on high-speed iteration and organic growth via social media.
• Growth Strategy: Founder Louie Phillips emphasizes "failing publicly" on social media. By creating "talking head" content that explains the game mechanics, they achieved 1 million downloads and 100,000 Instagram followers without paid media initially.
• Future Roadmap: The team is launching "Arenas," which will introduce speed-based leaderboards (e.g., fastest time around a landmark) alongside the existing volume-based territory system.

---

Part 2: Verge Labs – AI in Drug Discovery

Main Topics:

• Strategic Pivot: Verge Genomics rebranded to Verge Labs, shifting from a company that develops its own drugs to a platform that provides the "machine" (predictive models) for the pharmaceutical industry.
• The "Lidar" of Neuroscience: Verge Labs has built a massive dataset of over 12,000 human brains. Because brain tissue is difficult to access, this proprietary data acts as the "ground truth" (analogous to Lidar in self-driving cars) that allows their AI to interpret proxy data like blood samples.
• World Models for Biology: The company uses transformer-based architectures to create a "virtual biopsy." By fusing genetics, blood, and brain tissue data, they create a 512-dimensional "patient fingerprint."
• Masking Technique: Similar to LLMs, the model learns by "masking" (hiding) certain data types and predicting them based on others. This allows the model to infer brain activity from a simple blood draw—an emergent property not explicitly programmed.

Key Arguments & Evidence:

• Efficiency: Alice Zhang argues that the $5 billion cost of drug development is largely due to failures. By using AI to predict which patients will respond to a drug, companies can run smaller, more successful clinical trials, ultimately lowering drug costs.
• Validation: In a partnership with Eli Lilly, Verge’s AI-derived targets showed an 83% validation rate in wet-lab experiments, far exceeding the industry standard of ~20%.
• Scaling: Zhang notes that in biology, scaling parameter counts is less important than scaling "right-paired" data. The company’s moat is the 10 years of "unsexy" infrastructure built to collect and digitize brain tissue.

---

Synthesis and Conclusion

The video highlights two distinct applications of modern technology:

1. Consumer Engagement: Interval demonstrates that gamification, when applied to simple activities like running, can create high-retention, community-driven products that thrive on social competition.
2. Scientific Breakthroughs: Verge Labs illustrates the transition of AI from a pattern-matching tool to a predictive "world model" engine. By focusing on high-quality, proprietary data (brain tissue) rather than just raw compute, they are solving the "guess and check" problem in drug discovery.

Both companies emphasize the importance of iteration. Whether it is a running app adjusting its UI or a biotech firm feeding clinical trial failures back into its model, the core takeaway is that success in the AI era is driven by persistent data refinement and the willingness to learn from real-world feedback loops.