In a bold move that’s turning heads across Silicon Valley, two 22-year-old founders from Michigan reportedly turned down a multimillion-dollar offer from Elon Musk’s xAI to pursue a radically different vision for the future of artificial intelligence. Instead of joining Musk, William Chen and Guan Wang are betting on their own revolutionary model — one they say already outperforms systems built by OpenAI, Anthropic, and DeepSeek.
The duo behind Sapient Intelligence first met as teenagers in Michigan, bonding over what they described as “metagoals” — ambitious visions that stretched far beyond traditional coding projects. While Wang dreamed of creating an algorithm capable of solving any problem, Chen focused on optimising complex systems across engineering and real-world scenarios. Their shared drive set the foundation for a partnership now challenging the very structure of modern AI.
Turning Down Musk for a Bigger Vision
After graduating high school, Chen followed Wang to Tsinghua University in Beijing. There, they launched OpenChat, a lightweight language model built on high-quality conversations and reinforcement learning. The project gained significant traction within academic AI communities and soon drew interest from Elon Musk’s xAI. Musk approached the pair with a lucrative offer — but they walked away.
According to Chen, the decision came down to a core belief: today’s large-language models (LLMs) have hit structural limits.
“We decided that large-language models have their limitations,” Chen said. “We want a new architecture that will overcome the structural limitation of large-scale machine learning.”
Rather than enhance existing transformer-based LLMs, the two set out to build something fundamentally different.
Introducing the Hierarchical Reasoning Model (HRM)
This vision led to the launch of Sapient Intelligence and its defining breakthrough — the Hierarchical Reasoning Model (HRM). Built with a two-part recurrent architecture, HRM is designed to “think” more like a human, combining reflexive responses with deeper, deliberate reasoning.
Unlike traditional AI systems that statistically predict the next word, HRM aims to replicate layered, human-like thought processes. Chen describes this as the difference between an AI that reacts and an AI that reasons.
A Prototype That Surprised the AI World
In June, Sapient Intelligence tested a prototype with just 27 million parameters — tiny by today’s LLM standards — yet the results stunned observers. The model outperformed some of the industry’s major AI systems on abstract reasoning tasks, including:
- advanced Sudoku puzzles
- maze navigation
- The ARC-AGI benchmark, a key test for general intelligence
“Just with a change in the architecture, it gave the model a lot of what we call reasoning depth,” Chen said. “It was crazy.”
Chasing the Ultimate Goal: AGI
Now, Chen and Wang have their sights set on artificial general intelligence (AGI) — the ultimate milestone in AI development. Their belief is simple but ambitious: if AGI is possible, someone will build it, and they intend to be first.
With industry giants racing toward increasingly larger models, the two 22-year-old founders are taking the opposite path — pursuing smarter design over scale. And with Musk’s offer behind them, Sapient Intelligence is positioning itself as one of the most intriguing next-generation AI startups to watch.
If their early success with HRM is any indication, the AI landscape may be on the verge of a seismic shift — led not by the biggest tech companies, but by two young innovators rewriting the rules themselves.
Read more: Baidu Emerges as China’s Next Big AI Chip Leader Amid Nvidia’s Absence
