Silicon Valley’s AI Prodigy Sigil Wen Launches Underdog, a Privacy-First AI Assistant
At just 17, self-taught coder Sigil Wen made a bold move to Silicon Valley, immersing himself in an AI hacker house alongside renowned researcher Andrej Karpathy. This formative environment put Wen in the company of future AI visionaries such as Aravind Srinivas, founder of Perplexity, and Noam Brown, an OpenAI researcher. During this time, Wen experimented with early iterations of groundbreaking AI tools including Anthropic’s Claude chatbot, Midjourney’s image generator, OpenAI’s GPT-3, and Stable Diffusion. His early career also included a stint at Airchat, a social network rival to Clubhouse, founded by investor Naval Ravikant.
Wen recalls this period as “a magical time,” sharing how he even managed to run GPT-2 on his Apple Watch, demonstrating his knack for pushing the boundaries of AI technology.
Introducing Underdog: Privacy Meets Performance
Now a Thiel Fellow—part of Peter Thiel’s prestigious program supporting young founders outside traditional college paths—Wen has unveiled Underdog, an invite-only beta AI assistant that emphasizes privacy without compromising capability. Unlike most AI assistants that rely heavily on cloud computing, Underdog operates entirely on-device, ensuring user data never leaves their personal hardware. Currently available for Mac and Windows PCs, versions for Linux, iPhone, and Android are forthcoming.
Central to Underdog’s performance is Husky, an inference engine developed by Wen designed to optimize AI model execution by minimizing data transfer between a computer’s CPU and GPU. This innovation enables faster, more efficient on-device AI processing, a critical factor given Underdog’s commitment to user privacy.
Security is further enhanced through encryption of sensitive credentials, including email keys and other authorized accounts, providing robust protection against unauthorized access.
Balancing Model Size and Functionality
While Underdog uses smaller AI models than the massive, state-of-the-art models hosted in data centers, Wen points to its 27-billion parameter reasoning model, fine-tuned from Qwen3.8-27B, as highly competitive. According to benchmarks from Artificial Analysis, Underdog’s model performs comparably to Claude Opus 4.6, which was considered top-tier just six months ago.
This level of performance enables Underdog to handle everyday AI assistant tasks effectively, such as shopping research or solving math problems, without sacrificing privacy. Wen emphasizes, “You don’t need to sacrifice your privacy for the capability because they’re just as capable,” underscoring his belief that on-device AI models will only improve over time.
A Novel Business Model Aligned with User Privacy
Underdog’s early business strategy is as innovative as its technology. The app will initially be free and will never rely on advertising revenue. Since AI inference runs locally on users’ devices, operational costs remain minimal, allowing Wen to offer the product without subscription fees.
Leveraging the expertise of angel investor Patrick Collison, co-founder of Stripe, Underdog adopts a fintech-inspired approach: it will take a small percentage of payment transactions processed through Stripe’s secure payment infrastructure, akin to an interchange fee. This model eliminates the need for intrusive data mining, aligning Underdog’s incentives with those of trusted financial institutions rather than advertisers.
This approach starkly contrasts with many other AI assistants, which often collect extensive user data to monetize through advertising or by training further models—a practice raising significant privacy concerns. Such data collection is particularly sensitive for AI assistants, which may require access to deeply personal information, from medical history to financial details and family data.
Wen’s Vision: Building AI for Trust and Longevity
In his AI manifesto, Wen poses a fundamental question: “Why should using AI require surrendering your private information?” This ethos drives his development of Underdog as a product he would feel comfortable having his own children use—a testament to his commitment to trustworthiness.
The startup behind Underdog, Conway Research, boasts a strong investor lineup beyond Collison. Leading venture capital firms Andreessen Horowitz, Khosla Ventures, Hummingbird, SV Angel, and the Anthology Fund (a collaboration between Menlo Ventures and Anthropic) have all backed the company. Additionally, notable angel investors include Vercel founder Guillermo Rauch, OpenAI researcher Noam Brown, and Deedy Das, highlighting the project’s authoritative support within the AI community.
Underdog represents a compelling new direction in AI assistants—one that prioritizes user privacy, delivers robust performance, and offers a sustainable business model aligned with users’ interests. As AI continues to become indispensable in daily life, Wen’s vision offers a promising alternative to the data-hungry models dominating the market today.
When you purchase through links in our articles, we may earn a small commission. This doesn’t affect our editorial independence.
Source: Here
