AI news: Why Gen Z students turned down Elon Musk’s millions and built AI that’s now dominating OpenAI

AI news just took a shocking turn when two 22-year-old Michigan natives rejected a multimillion-dollar offer from Elon Musk’s xAI to pursue their revolutionary brain-inspired AI model. Their decision proves sometimes the best founders refuse the biggest checks to build something even bigger.

A Multimillion-Dollar Offer No One Would Refuse (Until They Did)

Two years ago in Beijing, William Chen and Guan Wang sat inside Tsinghua University’s brain lab staring at something most founders would instantly accept: a multimillion-dollar offer from Elon Musk’s xAI. The terms were lucrative. The company was prestigious. The opportunity was what every young technologist dreams about.

Yet they said no. “We decided that large-language models have their limitations,” Chen told Fortune. “We want a new architecture that will overcome the structural limitation of large-scale machine learning.” Instead of taking Musk’s deal, they doubled down on something far riskier: building an entirely new type of artificial intelligence.

From Michigan High School Dreams to Tsinghua’s Brain Lab

The story begins at Cranbrook Schools, a prestigious boarding school in Michigan where Chen and Wang first met. On their first conversation, they discovered an instant alignment. Wang’s metagoal was building an “algorithm that solves any problem”—what would later be called artificial general intelligence. Chen’s metagoal was optimizing everything, from engineering systems to real-world applications.

While most American high school graduates pursue prestigious U.S. universities, Chen made an unusual choice: he followed Wang to Tsinghua University in Beijing, often described as “China’s MIT.” The coursework proved grueling, but their vision remained unshakable. Professors noticed their ambition and supported their work on building revolutionary AI systems.

OpenChat: The Proof of Concept That Changed Everything

Achievement	Details
First Project	OpenChat: small LLM trained on high-quality conversations
Recognition	Adopted by researchers at Berkeley and Stanford; became early example of AI quality over quantity
Second Project	Hierarchical Reasoning Model (HRM): brain-inspired reasoning architecture
Benchmark Win (June 2025)	HRM with 27M parameters outperformed OpenAI, Anthropic, DeepSeek on ARC-AGI

Their first project, OpenChat, was revolutionary in its simplicity. They trained a small large-language model on carefully curated high-quality conversations rather than massive internet data dumps. They taught it reinforcement learning—how humans learn through feedback and rewards. “It got very famous,” Chen said of the open-sourced model that researchers at Berkeley and Stanford immediately adopted.

That success caught attention in an unexpected place: Elon Musk’s inbox. xAI sent the recruitment offer with a massive financial package. For most young founders, this would be career-defining. For Chen and Wang, it was just a distraction from something bigger.

The 3 A.M. Breakthrough That Changed the AI Industry

The critical moment came in June 2025 at 3 a.m. when Chen and Wang witnessed their tiny HRM prototype with just 27 million parameters demolish performance benchmarks. The results were extraordinary. Their model solved advanced Sudoku-Extreme puzzles, navigated optimally through 30×30 mazes, and achieved 40.3% on ARC-AGI-1 benchmarks—crushing OpenAI’s o3-mini-high at 34.5%, Anthropic’s Claude 3.7 at 21.2%, and DeepSeek R1 at 15.8%.

“It was crazy,” Chen said. “Just with a change in the architecture, it gave the model a lot of what we call reasoning depth. It’s not guessing. It’s thinking.”

— William Chen, Co-founder of Sapient Intelligence

Why Tiny, Efficient AI Beats Massive Language Models

Unlike transformers that predict words through statistical patterns, HRM uses a two-part recurrent structure loosely modeled on the human brain. It mixes deliberate, slow thinking with fast reflexive reactions, enabling genuine problem-solving rather than probability-based guessing. Chen argues that current AI’s reasoning limitations are structural, not temporary—you can stack more layers, but you hit fundamental limits inherent to massive models.

Their models hallucinate far less than traditional LLMs and already match state-of-the-art performance in weather prediction, quantitative trading, and medical monitoring. Sapient Intelligence believes artificial general intelligence won’t emerge from bigger transformers but from smaller, more efficient reasoning architectures trained on exceptional data quality rather than quantity.

What’s Next for Two Gen Zers Who Rejected the Billionaire?

Chen and Wang are preparing to open a U.S. office within November 2025 and raise additional funding. They’re exploring continuous learning capabilities—the ability for models to absorb new experiences without complete retraining. Their ultimate thesis: AGI will emerge from better architecture, not bigger models. Both founders believe pushing the technology to fruition is essential given the stakes. “One day, we’re going to have an AI that’s smarter than humans,” Chen said. “Guan and I always say it’s like Pandora’s box. If we’re not going to make it, someone else will.” Their confidence extends to expecting AGI emergence within the next decade.

Sources

• Fortune – Comprehensive interview with William Chen and story details
• NDTV – Verification of benchmark performance metrics and company background
• Live Science – Scientific confirmation of ARC-AGI benchmark results