Archive — US — techcrunch
Silicon Valley’s AI wunderkind launches Underdog, the most private Instinct/Musecompetitor yet
Sigil Wen, a self-taught coder and Thiel Fellow, has launched Underdog, an AI assistant prioritizing user privacy by processing data entirely on-device. The invite-only beta offers users autonomy over their data, utilizes a new inference engine called Husky, and implements a unique business model focusing on transaction fees without ads, aiming to compete with larger AI tools from major tech companies.
Key points
- Sigil Wen, at 17, lived in an AI hacker house and has collaborated with prominent AI figures like Andrej Karpathy.
- Wen launched Underdog, an AI assistant emphasizing privacy by running on-user devices, preventing data from being sent to servers.
- Underdog uses smaller models, leveraging a 27-billion parameter reasoning model, yet performs comparably to larger, data-center-hosted models.
- The app will initially be free and will avoid ads, instead taking a small transaction fee on payments processed through Stripe.
- Wen's approach addresses privacy concerns prevalent in many AI assistants, who often exploit user data for advertising and other purposes.
Why it matters
- The launch of Underdog aligns with a growing demand for privacy-focused technology solutions in 2025.
- This development reflects an increasing trend of on-device processing among software to safeguard user data, contrasting with data-hungry cloud services.
- As consumers become more aware of data privacy issues, there is a market shift towards providers that prioritize user control over personal information.
Underdog highlights a pivotal shift in AI technology, emphasizing user privacy and lower costs through on-device processing, positioning itself as a significant competitor among traditional AI assistants while promoting ethical data practices.