Math Olympiad Silver, Stanford, Google Brain. Vietnamese Engineer Pham Hy Hieu Left OpenAI.
Math Olympiad silver, Stanford, Google Brain, xAI, OpenAI. Pham Hy Hieu spent fifteen years reaching AI's front line. Seven months later, he went home with burnout.
On February 26, 2026, a Vietnamese man posted a thread on X that quickly spread through tech circles.
"I am burnt out. All the mental health deteriorating that I used to scoff at is real, miserable, scary, and dangerous."
The author was Pham Hy Hieu, 34 years old, an engineer who had spent seven months at OpenAI. He said he was leaving the AI front line and taking his family back to Vietnam.
He used to hear people say "working at an AI company will wreck your mental health" and thought they were exaggerating. Now he was the one who had broken down.
A Math Prodigy From Vietnam
Pham Hy Hieu was born in 1992 in Ho Chi Minh City. He collected math competition prizes from a young age. In sixth grade, he won gold at the International Mathematics Olympiad for primary school students. In eleventh grade, he represented Vietnam at the 50th International Mathematical Olympiad (IMO) in Germany and took home a silver medal.
That medal earned him a scholarship. In 2011, the nineteen-year-old flew to the United States and entered Stanford's computer science program.
In his junior year, NLP heavyweight professor Christopher Manning brought him onto his research team. That was his first exposure to AI research. The pure math student changed direction.
He graduated in 2015 with honors, receiving the department's Ben Wegbreit Award for best honors thesis.
Six Months of Failure
In 2017, Google and Carnegie Mellon University launched a joint doctoral program that let students pursue a PhD while conducting research at Google Brain. It accepted only a handful of people worldwide. Pham Hy Hieu was one of them.
His advisor was Vietnamese-born scientist Le Viet Quoc (Quoc V. Le) at Google Brain. The research focus: getting AI to automatically design neural network architectures. Hieu proposed an idea: let different AI models share parameters so each one would not have to be trained from scratch.
When he pitched it, Le Viet Quoc's reaction was: "That idea is crazy."
But he did not stop him. He let Hieu try.
Over the next eleven months, the first six produced nothing but failures. Every single experiment. Hieu was frustrated and anxious, feeling like he was wasting time. Le Viet Quoc was the only person who kept telling him not to give up.
The turning point came from two critical assists.
First, Le Viet Quoc arranged a meeting with Geoff Hinton. Hinton was then Google's VP of AI and one of the most important founders of deep learning, later awarded the 2024 Nobel Prize in Physics. Getting time with him was extremely difficult, but Quoc secured thirty minutes.
Hinton looked at the experimental results and said flatly, "This is terrible." But he spent the full thirty minutes analyzing what was wrong and pointing to directions for improvement. For Pham Hy Hieu, the biggest takeaway was that "someone that brilliant was willing to seriously examine your bad work."
Second, Quoc brought in colleague Barret Zoph to join the team and help fix many bugs in the code. Two months later, positive results finally appeared.
They submitted the paper to a top conference. Rejected. Reviewers called the idea "too crazy" and the writing too messy.
Christmas 2017. Everyone was on holiday. Le Viet Quoc stayed behind, working through the night to rewrite the paper and design supplementary experiments. They resubmitted to ICML (International Conference on Machine Learning). This time, it was accepted.
The paper was called ENAS (Efficient Neural Architecture Search via Parameter Sharing). It reduced the computational cost of neural architecture search by a factor of one thousand. Co-authors included Quoc V. Le and Jeff Dean. By 2020, it had been cited over 4,600 times.
In 2021, he completed his PhD at CMU. He was 29.
Heading to the Front Line
After his doctorate, Pham Hy Hieu stayed at Google Brain as a research scientist. In 2020, Forbes Vietnam named him to its "30 Under 30" list.
That year, he gave interviews to Vietnamese media and spoke optimistically about AI. He predicted that within a decade, AI would dominate international math olympiads. He said Vietnamese AI talent was rising fast, with more Vietnamese faces appearing at top conferences.
Around 2023, after leaving Google, he joined startup Augment Computing and helped it reach a billion-dollar valuation.
In August 2024, Elon Musk's xAI recruited him. At xAI, he worked on attention kernel optimization for the Grok-3 model.
That same August, he moved to OpenAI. Counting from 2011 when he left Vietnam, it had taken fourteen years to reach the world's most central AI laboratory.
Seven Months
He did not share specifics about what happened at OpenAI. He described the experience as "once-in-a-lifetime" and his colleagues as "the very best in every way." He participated in building "extremely intelligent entities" and was proud of it.
But the cost, he said, was too high.
"I cannot believe I would say this one day, but I am burnt out. All the mental health deteriorating that I used to scoff at is real, miserable, scary, and dangerous."
He also wrote: "Today, I finally feel the existential threat that AI is posing."
A person who spent his entire career building AI said he finally felt AI's existential threat. The weight of that statement speaks for itself.
He Is Not the Only One
Pham Hy Hieu is not an isolated case. Around the same time, Mrinank Sharma, head of Anthropic's (Claude's developer) safety research team, also resigned, saying the world faces "multiple intertwined crises."
Nathan Lambert, a researcher at the Allen Institute for AI, observed that companies like OpenAI and Anthropic have developed a culture of "self-imposed pressure." Nobody explicitly tells you to work overtime, but when everyone around you is pushing to the limit, not doing so feels like falling behind. Some have compared the intensity to China's 996 work culture.
Under the pressure of the AI arms race, every lab is competing on speed. Models get stronger generation after generation. Release cycles get shorter. Engineers are the fuel being fed into the accelerator.
Going Home
Pham Hy Hieu has taken his family back to Vietnam. He says he wants to rest, recover, and try something new.
Back in April 2025, while still at xAI, Vietnam National University HCMC had appointed him as a visiting professor, one of the first sixteen scientists selected. His connection to Vietnam never broke.
Fifteen years ago, a Vietnamese teenager set off for America with a math olympiad medal. Now he has come home.