Yang Zhilin is having his moment in the sun right now.
The 34-year-old cofounder of the Chinese firm Moonshot AI has been dubbed a “millennial genius” after the launch of his new Kimi K3 model this month, which sent shockwaves across the AI industry.
But years before he cofounded Moonshot AI in 2023 in China, Yang studied in the US and was one of two dozen Ph.D. students in machine learning at Carnegie Mellon University.
Business Insider spoke with Yang’s former Ph.D. advisor, Ruslan Salakhutdinov, a computer science professor at Carnegie Mellon University, about the buzzy founder and the qualities that make him exceptional.
The making of an AI founder
Salakhutdinov said Yang possessed four qualities that set him apart even among one of the world’s most competitive AI programs: unusual independence, original thinking, rare technical breadth, and the ambition to take risks.
Salakhutdinov has a unique vantage point. Beyond supervising Yang, he has also taught or mentored researchers who have gone on to cofound AI startups, become professors at leading universities, and secure research roles at companies including OpenAI, Anthropic, Google DeepMind, Meta, Microsoft AI, and Tesla AI.
He told Business Insider that the most successful AI researchers he’s taught tend to share “extreme motivation,” “extreme self-direction,” and a strong work ethic, while also benefiting from highly competitive environments that push them to excel.
Read more about Moonshot AI’s viral Kimi K3 model
Some of those former students, he said, include xAI cofounder Jimmy Ba, Mistral founding member Devendra Chaplot, and Perceptual Machines cofounder Nitish Srivastava. Salakhutdinov has also taught at the University of Toronto.
Yang, however, stood out even among that group.
The professor said Yang joined Carnegie Mellon directly after his undergraduate degree and quickly developed into an unusually self-directed researcher.
“He started being very independent in terms of his research,” Salakhutdinov said.
He completed his Ph.D. in four years, from 2016 to 2019. During that time, he also helped develop an AI language model, XLNet.
Salakhutdinov recalled being struck by Yang’s willingness to challenge conventional thinking. Around 2019, three years before ChatGPT entered the mainstream, Yang said he believed it was possible that a single AI model, if scaled up enough, could perform many different tasks from simple natural-language prompts rather than requiring separate models for each task.
“I was like: ‘What is this? This is never gonna work. This is a little bit crazy,'” Salakhutdinov recalled telling him.
Yang’s response stuck with him.
“He was like, ‘No, I think if we scale these things should work.'”
Yang developed those ideas while interning on Google’s AI Brain team, where he coauthored two AI papers outlining new methods: Transformer-XL, which they proposed would help models understand longer text, and XLNet, which they said would improve how AI systems understand language.
Salakhutdinov said Yang wasn’t alone in thinking that scaling could unlock more general AI systems, particularly among researchers at Google, but that the belief was “not shared by many researchers” at the time.
Yang didn’t immediately respond to Business Insider’s request for comments.
From Ph.D. prodigy to AI star
Kimi K3 is an open-weight model that rivals leading systems from OpenAI and Anthropic on coding, reasoning, and knowledge tasks.
Its release this month sparked debate across Silicon Valley about China’s AI progress, while overwhelming demand forced Moonshot to temporarily pause new subscriptions.
Moonshot’s rapid rise has helped propel the three-year-old startup to a valuation of about $31.5 billion in its latest funding round. Bloomberg reported this week that the company is already preparing another financing round that could value it at as much as $50 billion ahead of a potential IPO.
Salakhutdinov said Yang’s independence, technical depth, original thinking, and entrepreneurial ambition help explain how he built one of China’s fastest-rising AI companies.
Another was Yang’s unusual ability to combine deep research with low-level engineering.
“He has the ability to do research,” Salakhutdinov said, but also “the remarkable ability to go and do low-level coding” and optimize GPUs for large-scale AI experiments.
Salakhutdinov said the most accomplished AI Ph.D. students are often expected to become professors at top universities or join leading technology companies.
But Yang was determined to start his own company, even after a brief stint as a professor in China.
“I was actually a little bit surprised, because it’s risky,” Salakhutdinov said. “You can go, you can try building a company, and if it doesn’t go anywhere, then your career can suffer.”
Yang, however, had already made up his mind. “He basically told me: ‘I have to try. If I don’t try, then I’ll be regretting it,'” Salakhutdinov said.
Read the full article here
