Premium

2 Gen Z startup co-founders turn down Elon Musk’s multimillion-dollar offer to build AI to beat DeepSeek, OpenAI

William Chen and Guan Wang, co-founders of Sapient Intelligence, first connected in high school over what they described as “metagoals."

Elon Musk’s xAI said the new funding would be used to expand data centers and accelerate research as competition among major AI developers intensifies.Elon Musk’s xAI said the new funding would be used to expand data centers and accelerate research as competition among major AI developers intensifies (File photo)

Two friends aged 22 from Michigan turned down a multimillion-dollar offer from Elon Musk to chase a breakthrough in Artificial Intelligence (AI) they believe could surpass the biggest players in the field, including OpenAI, Anthropic, and DeepSeek.

William Chen and Guan Wang, co-founders of the startup Sapient Intelligence, first connected in high school over what they described as “metagoals,” Fortune reported. For Wang, that meant creating an algorithm capable of solving any problem; for Chen, it was about optimising complex systems in engineering and real-world scenarios, the report added.

“One day, we’re going to have an AI that’s smarter than humans,” Chen told Fortune. “If we’re not going to make it, someone else will.”

After high school, Chen joined Wang at Tsinghua University in Beijing, where the duo created OpenChat, a lightweight language model trained on high-quality conversations and built to improve itself using reinforcement learning. OpenChat quickly gathered attention in academic circles, eventually catching the eye of Musk. Through xAI, he offered them a multimillion-dollar deal, which they turned down, the report stated.

“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.”

Later, they created Sapient Intelligence and their new architecture, the Hierarchical Reasoning Model (HRM). HRM uses a dual recurrent structure designed to imitate human thought processes by blending deliberate reasoning with automatic, reflex-like responses, the report said.

In June, a 27-million-parameter prototype (a small AI model) reportedly outperformed major AI systems on abstract reasoning tasks, including advanced Sudoku, maze navigation, and the ARC-AGI benchmark. ARC-AGI or Abstraction and Reasoning Corpus – Artificial General Intelligence tests whether an AI can engage in reasoning as a human would.

Story continues below this ad

“It was crazy,” Chen told Fortune. “Just with a change in the architecture, it gave the model a lot of what we call reasoning depth.”

With Sapient Intelligence, the duo is now setting its sights on artificial general intelligence.

 

Latest Comment
Post Comment
Read Comments
Advertisement
Loading Taboola...
Advertisement
Advertisement
Advertisement