In a bold move that has captured the tech world's attention, two young entrepreneurs from Michigan have declined a lucrative multimillion-dollar acquisition offer from Elon Musk. Their mission? To develop an artificial intelligence system they believe can outpace industry giants like OpenAI, Anthropic, and DeepSeek.
From High School "Metagoals" to a Shared AI Vision
William Chen and Guan Wang, both aged 22, are the co-founders of Sapient Intelligence. Their partnership began in high school, forged over discussions about what they term "metagoals." According to a report by Fortune, Wang's driving ambition was to create an algorithm with the potential to solve any problem. Chen, on the other hand, was focused on optimizing complex systems in both engineering and real-world scenarios.
This shared passion for foundational problem-solving laid the groundwork for their future venture. "One day, we're going to have an AI that's smarter than humans," Chen told Fortune, highlighting their conviction. "If we're not going to make it, someone else will."
The Path from OpenChat to Rejecting xAI
After completing high school, their collaboration deepened at Tsinghua University in Beijing. There, they created OpenChat, a lightweight language model trained exclusively on high-quality conversations. Its unique design allowed it to improve itself using reinforcement learning techniques.
The model quickly gained traction in academic circles, eventually attracting the interest of billionaire entrepreneur Elon Musk. Through his AI company, xAI, Musk presented the duo with a multimillion-dollar deal to acquire their work. In a surprising decision, Chen and Wang turned the offer down.
Explaining their rationale, Chen stated, "We decided that large-language models have their limitations. We want a new architecture that will overcome the structural limitation of large-scale machine learning." This rejection marked a pivotal moment, steering them away from a well-funded path to pursue a more ambitious, self-directed vision.
Building Sapient Intelligence and the HRM Architecture
Following their decision, Chen and Wang formally established their startup, Sapient Intelligence. Their focus shifted to developing a novel framework called the Hierarchical Reasoning Model (HRM). As per reports, the HRM architecture employs a dual recurrent structure engineered to mimic human cognitive processes. It achieves this by blending deliberate, step-by-step reasoning with fast, automatic, reflex-like responses.
The potential of their approach became evident in June. A small prototype with just 27 million parameters reportedly outperformed major AI systems on complex abstract reasoning tasks. These tests included advanced Sudoku puzzles, intricate maze navigation, and the challenging ARC-AGI benchmark. The ARC-AGI (Abstraction and Reasoning Corpus – Artificial General Intelligence) benchmark is specifically designed to evaluate if an AI can reason in a human-like manner.
"It was crazy," Chen remarked about the results. "Just with a change in the architecture, it gave the model a lot of what we call reasoning depth." This early success has fueled their confidence as they now set their sights on the ultimate prize in AI research: achieving true artificial general intelligence (AGI).
