Fei-Fei Li’s influence on deep learning is undeniable, primarily through ImageNet and her role in establishing human-centered AI frameworks. However, the narrative of singular genius often obscures the collaborative nature of these breakthroughs; this claim would be falsified if we could isolate her specific technical contributions from the broader community efforts that scaled them.

Li holds a tenured professorship at Stanford University, serves as Director of the Institute for Human-Centered Artificial Intelligence (HAI), and founded World Labs, a newly minted unicorn focused on embodied and spatial intelligence. She is widely regarded as one of the most influential Chinese Americans in AI, with her students actively shaping industry trajectories and offering critical perspectives on “AGI.” While some hail her as the “Godmother of AI,” others note that like Geoffrey Hinton, she did not originate from a traditional computer science background.

I think the “Godmother” title is marketing shorthand that risks oversimplifying complex technical histories. From the paper, imageNet’s impact relied on massive annotation labor, not just algorithmic innovation. One caveat: world Labs’ valuation assumes spatial intelligence will scale faster than current benchmarks suggest.

How a Chengdu Student Reshaped the Trajectory of AI

The Course of AI Rewritten by a Physics Enthusiast

Fei-Fei Li’s pivot from physics to artificial intelligence wasn’t just a career change; it was a bet on data over pure algorithmic elegance. This trajectory would only make sense once the industry realized that scale could outperform theoretical purity—a hypothesis falsified if massive datasets fail to generalize across domains without architectural breakthroughs.

Li began as a physicist, idolizing Einstein, long before the 2024 Nobel Prize cemented the physics-AI nexus in public consciousness. In 1995, however, AI was still in its “winter.” While Hinton’s backpropagation and LeCun’s CNNs had solved key theoretical puzzles, computing power and data scarcity kept neural networks from demonstrating real-world utility.

Li’s entry into the field came during her sophomore summer at UC Berkeley, assisting an experiment on computational neuroscience. She feared rejection due to her lack of biology background, but the project focused on reconstructing cat vision from brain signals—a study later published in the Journal of Neuroscience. This experience clarified her path: she would use computer vision to decode intelligence.

After graduating from Princeton’s physics department in 1999, Li declined lucrative Wall Street offers. Supported by her family, she chose academia, joining Caltech for her master’s under Pietro Perona and Christof Koch. She pursued parallel research in neuroscience and computer science, though she did not formally begin AI research until 2001—five years before ImageNet.

In June 2009, Li and her team released ImageNet. At the time, it was the world’s largest image labeling dataset; today, it remains a cornerstone of computer vision (CV). The dataset comprises 15 million curated internet images spanning 22,000 concept and object categories.

I think relying on web-scraped data introduces significant label noise that modern models often fail to correct. From the paper, the sheer scale of ImageNet may no longer be sufficient for next-generation multimodal benchmarks. One caveat: we assume the 22,000 categories remain relevant despite shifts in cultural semantics over time.

While such volume is standard now, ImageNet was radical in 2006. AI was still algorithm-centric; researchers prioritized complex network architectures, treating data merely as a training utility rather than a primary driver of intelligence. The prevailing view held that improving synaptic connections—via better algorithms—was the only path to machine intelligence.

If machine intelligence is compared to biological intelligence, algorithms are like synapses or the intricate wiring in the brain. Therefore, the most important thing is to make these connections better, faster, and more powerful.

How a Chengdu Student Reshaped the Trajectory of AI

The core technical claim here is that Fei-Fei Li’s ImageNet was not merely a dataset but a necessary precondition for the generalization capabilities required by modern machine learning, effectively correcting the field’s stagnation in model complexity. This hypothesis would be falsified if subsequent breakthroughs in computer vision had occurred without access to large-scale, diverse natural image data, proving that algorithmic innovation alone could overcome data scarcity.

The Burden of Scale and Academic Risk

Li’s ambition was to construct a “map of human meaning” from an image dimension using WordNet as a lexical foundation, aiming for tens of thousands of categories. At the time, this scale was unimaginable; models rarely recognized more than one or two classes. Colleagues questioned the necessity, the training timeline, and the annotation workload.

Jitendra Malik, whom Li calls her “academic grandfather,” warned that while ImageNet was needed for computer vision, the trick of science is to grow with the field, not to run ahead of it. He explicitly stated:

He said that if I did this, might be difficult for me to obtain tenure.

Li recognized that machine learning had stagnated because researchers were building complex models without sufficient data to drive generalization. She argued in her autobiography that biological intelligence is the result of evolution, shaped by environmental influence, and that natural images are simply data reflecting this prior sensory stimulation.

I think the assumption that evolution maps cleanly to algorithmic training ignores the lack of selective pressure mechanisms in static datasets. From the paper, equating natural images with “data” overlooks the significant bias inherent in web-scraped visual corpora.

Institutional Support and the Princeton Connection

Li found support in Professor Kai Li at Princeton, a tenured professor and expert in microprocessor architecture. Li described him as intellectually refined yet warm, noting he understood “exponential thinking” better than most. While their fields overlapped little, Kai Li provided two critical resources: initial workstations and an introduction to his student, Jia Deng.

Professor Li understood the power of exponential thinking better than most people. He believed I was pursuing an important goal.

Deng would later become widely known as the first author of ImageNet. Thus, in 2007 at Princeton, ImageNet officially launched.

One caveat: the reliance on a single advisor’s hardware donation highlights the precarious funding landscape for high-risk AI projects in the mid-2000s. I think attributing success to “exponential thinking” risks obscuring the mundane, labor-intensive reality of data curation.

From Princeton to Stanford: The Long Road to CVPR 2009

For three years, Li and her students faced massive workloads, high costs, and minimal external support. In 2009, after relocating to Stanford with Deng and most students, ImageNet completed its first version and was unveiled at CVPR. To promote the dataset, Li hosted the ImageNet Challenge, inviting global scholars to benchmark object recognition algorithms under a unified standard.

Despite these efforts, the impact remained limited for several years. It wasn’t until 2012 that the timeline tightened, intersecting with Geoffrey Hinton’s work on deep learning. The narrative cuts off as Jia Deng makes a pivotal phone call to Li, marking the moment when the dataset finally began to reshape the trajectory of AI research.

The narrative suggests a linear causality between AlexNet’s success and the subsequent consolidation of talent at Google and OpenAI, but this overlooks the complex, often fragmented nature of academic-industry transitions during that period.

From the paper, the “auction” metaphor simplifies a messy series of negotiations and competing offers into a tidy transaction. One caveat: linking Fei-Fei Li’s administrative roles directly to student output ignores the independent trajectories of those researchers. I think listing fellowships as proof of AI trajectory reshaping conflates institutional recognition with technical influence.

The Institutional Aftermath of AlexNet

The source describes a specific moment where an “ally reserved and calm student” expressed excitement upon encountering AlexNet. This paper, as noted in the filing, re-proved the viability of neural networks with astonishing accuracy, effectively ushering in the second wave of artificial intelligence.

The article then posits that Geoffrey Hinton subsequently facilitated a transition—described as an “auction”—involving Ilya Sutskever and Alex Krizhevsky to Google, while Ilya moved to OpenAI as Chief Scientist. According to the text, this movement drove the creation of early GPT versions, the DALL·E series, CodeX, and ChatGPT. The narrative frames these events as accelerating gears of change.

From the paper, attributing the birth of specific models like DALL·E solely to Ilya’s role at OpenAI ignores the broader team efforts and infrastructure investments required. One caveat: the term “auction” implies a market-driven selection that may not reflect the personal or strategic motivations behind these hires.

Fei-Fei Li’s Administrative Trajectory

Shifting focus, the article asks about Fei-Fei Li following ImageNet’s success. It outlines her career path as increasingly smooth: she received tenure at Stanford in 2012 and became an Associate Professor. By 2013, she was leading the Stanford Artificial Intelligence Laboratory (SAIL).

During an academic leave in 2016—the year the deep learning revolution began—she joined Google Cloud’s China Center for Artificial Intelligence and Machine Learning. From January 2017 to September 2018, she served as Vice President of Google and Chief Scientist for Google Cloud AI/ML.

In September 2018, Li announced her return to Stanford to teach, becoming the Director of the Stanford Institute for Human-Centered Artificial Intelligence (HAI) and a Full Professor. Her institutional recognition followed rapidly: elected to the National Academy of Engineering and Medicine in 2020; Fellow of the American Academy of Arts and Sciences in 2021; and IEEE Fellow in November 2021.

I think equating administrative promotions with direct contributions to AI technical breakthroughs is a logical leap not supported by the data provided. From the paper, the timeline suggests her industry roles overlapped significantly with her academic leadership, raising questions about resource allocation at Stanford HAI.

The Lab’s Output

The article concludes by noting that dozens of students have emerged from Fei-Fei Li’s lab, many of whom have profoundly influenced AI development. However, it fails to specify which students or what specific contributions they made, leaving the claim vague and difficult to verify against public records.

I read the section detailing Fei-Fei Li’s mentorship network in her new autobiography. The core claim is that Li’s pedagogical focus on “why” over “how” directly shaped the engineering resilience of high-profile AI leaders like Andrej Karpathy and Jia Deng. This narrative would be falsified if these figures’ public accounts contradicted the idea that their success stemmed from this specific type of theoretical grounding rather than pure execution speed or resource accumulation.

A Constellation of Brilliant Disciples

Having taught for nearly 20 years, Fei-Fei Li has mentored a large group of outstanding disciples, shining like stars—

Among those we are familiar with are Andrej Karpathy, a founding member of OpenAI; NVIDIA scientists Jim Fan and Yukey Zhu (Zhu Yuke); Shanghai Jiao Tong University Professor Ce Wu Lu; former President of Google AI China Center Li Jia; and Wang Gang, former head of Alibaba’s autonomous driving division…

When mentoring Karpathy, who was then a second-year graduate student, Fei-Fei Li evaluated this tall, fast-talking student as:

He has the courage and perseverance of an engineer. Whether writing equations all over a whiteboard or dismantling transistor radios, it’s easy for him.
If Einstein and Bohr are dreamers of the universe, then Karpathy belongs to the category of Edison or the Wright Brothers.

The task she assigned to her team and Karpathy was: Input an image, and ultimately output a text description automatically.

Karpathy’s first submission appeared to have completed the task. However, she pointed out that this assignment mainly relied on “matching” existing data and could not handle new situations, meaning it lacked generalization ability.

Frustrated, Karpathi slumped in his seat. Seeing him like this, Fei-Fei Li took the opportunity to remind him:

Karpathy and many students share a common problem: They are overly concerned with whether their model works but forget to ask why it works.

Fortunately, after the depression passed, the “engineer traits” in Karpathy began to take effect.

Although no one knew at this point how he would actually achieve the goal, I knew that the engineer inside him, like me, would persist.
He definitely could do it.

Indeed, he eventually succeeded…

During his PhD studies, he personally designed and taught a course titled “CS231n: Convolutional Neural Networks for Visual Recognition,” becoming an instructor teaching deep learning at Stanford.

This course has always been highly praised and very popular.

After obtaining his postdoctoral position, Karpathy faced multiple career choices (Princeton University was willing to offer him a direct position), but ultimately chose to leave academia and resolutely join the then-obscure OpenAI.

Fei-Fei Li advised him against this, but Karpathy was determined about OpenAI:

This is really different from anywhere else.

The rest of the story is well known. He joined and left OpenAI twice, which seems to have a bit of quantum entanglement flavor (doge).

In 2016, he joined OpenAI as a researcher…

(also a co-founder), led the development of early GPT series, DALL·E series, and ChatGPT models. After working there for one year and six months, he was poached by Elon Musk to Tesla, where he led the computer vision team for autonomous driving.

Under the leadership of Karpathy and Pete Bannon, who headed hardware, Tesla eventually launched Full Self-Driving (FSD).

It wasn’t until February 2023 that he returned to OpenAI, at which point Sam Altman tweeted his welcome. Over nearly a year, he built a small team responsible for improving GPT-4, before leaving again…

His next destination was also entrepreneurship.

In July this year, he announced the founding of Eureka Labs, a new type of AI-native school.

Its first product, and indeed its first course, is LLM101n (returning to his roots).

A step-by-step guide to building a large story-generation model similar to ChatGPT, along with a companion web application.

Besides Karpathy, Fei-Fei Li’s new autobiography frequently mentions Sloan Prize winner Jia Deng.

Deng graduated with a bachelor’s degree in Computer Science from Tsinghua University in 2006 and subsequently went to Princeton University in the U.S. to pursue his Ph.D. under Professor Kai Li.

In 2007, he was recommended by his advisor, Kai L

I read the release notes on Jia Deng’s trajectory, which posits that his reserved nature was a key variable in ImageNet’s longevity. This narrative is falsifiable if archival records show he drove technical decisions behind the scenes while Li took public credit.

One caveat: the reliance on Fei-Fei Li’s anecdotal evidence creates a single-source bias for Deng’s contributions. I think assuming “reserved” equates to lack of ambition ignores the strategic value of low-profile leadership in open datasets.

When ImageNet was published in 2009, Jia Deng was the first author. Fei-Fei Li described him as reserved and understated:

I have never met anyone with such a brilliant mind who showed no desire to stand out.

Until Fei-Fei Li announced the discontinuation of ImageNet in 2017, Deng had been helping to operate the project. After receiving his Ph.D. (graduating in 2012), he began serving as an Assistant Professor in the Department of Computer Science and Engineering at the University of Michigan in 2014. He stayed for only four years before returning to Princeton University. He is currently an Associate Professor of Computer Science there, leading the Visual and Learning Laboratory. Notably, he was also a recipient of the 2018 Sloan Research Fellowship. This award represents some of the most promising scientific researchers in the world (particularly in the U.S. and Canada); since its establishment in 1955, it has produced numerous Nobel Prize and Fields Medal winners.

From the paper, the Sloan Fellowship validates peer recognition but does not quantify Deng’s specific technical impact on ImageNet’s architecture.

Of course, during Fei-Fei Li’s early teaching career, two other students deserve mention: Li Jia, former President of Google AI China Center, and Wang Gang, former head of Alibaba’s autonomous driving division. Li Jia entered the Automation Department at the University of Science and Technology of China in 1998 and later obtained a Master’s degree from Nanyang Technological University in Singapore. From 2016 to 2020, Li Jia pursued her Ph.D. under Fei-Fei Li, during which time a notable mentor-student story unfolded. Because Fei-Fei Li taught at UIUC, Princeton, and Stanford in succession, Li Jia followed her three times, changing schools and taking the doctoral entrance exam three times (succeeding each time), becoming Fei-Fei Li’s most proud student.

One caveat: the “followed three times” narrative prioritizes loyalty metrics over independent academic mobility, which is rare in top-tier ML research.

After graduation, she joined Yahoo in 2011 and became a Senior Researcher within two or three years, leading the visual computing and machine learning departments at Yahoo Labs. During this period, she received internal company awards such as LEAP and Master Inventor, as well as Yahoo’s highest honor, the Super Star Award. In February 2015, she joined Snapchat as Head of R&D, tasked with developing core CV/AI technologies and providing innovative support for products. At that time, Snapchat had already clarified its IPO plans; if successful, it would be the largest deal by a U.S. tech company since Facebook’s listing. Logically speaking, no one would choose to leave at this stage. But when her mentor Fei-Fei Li called, Li Jia resigned in September 2016, and the two joined Google shortly after each other. During their time at Google, they released several new AutoML products and Contact Center AI virtual assistants, and promoted the establishment of the Google AI China Center. Li Jia also served as President of the Google AI China Center, helping to enhance Google’s influence in China. After completing her mission at Google, the mentor and student resigned again shortly after each other, with only a 50-day gap between their departures. Fei-Fei Li returned to Stanford, while Li Jia considered entrepreneurship in the AI direction. She first served as Co-founder and Founding CEO at StartX, providing non-profit “acceleration” support for Stanford alumni startups.

She also taught an AI healthcare course at Stanford, titled AI Empowering Healthcare. The course primarily uses AI technologies such as computer vision to solve problems in the current healthcare industry, such as home care, surgical assistance analysis, AI-assisted parenting, burn assessment, and more. In the latest development, she has chosen the “enterprise AI solutions” route for entrepreneurship. In March 2023, she co-founded LiveX AI, providing enterprises with products such as chatbots, AI search, and voice agents to help increase paid conversions and reduce customer churn rates. So both the mentor and student have embarked on entrepreneurial paths a

How a Chengdu Student Reshaped the Trajectory of AI

Again, appearing just as in sync as ever.

The other student, Wang Gang, is equally impressive: Tenured Professor at Nanyang Technological University, pioneer of Alibaba’s unmanned vehicles, head of DAMO Academy’s Autonomous Driving Laboratory, and father of the “Little Donkey” logistics robot…

Wang Gang graduated with a bachelor’s degree from Harbin Institute of Technology in 2005 and received his Ph.D. from the University of Illinois Urbana-Champaign in 2010; his doctoral advisor was Fei-Fei Li.

Upon graduating with his Ph.D. at age 28, he already held ten top-tier conference papers with over a thousand citations, representing a new generation in the AI field.

Before joining Alibaba in 2017, Wang Gang was already a Tenured Professor at Nanyang Technological University at age 34.

After joining Alibaba, Wang Gang served as Chief Scientist at Alibaba’s Artificial Intelligence Laboratory and later became Head of DAMO Academy’s Autonomous Driving Laboratory.

He pioneered autonomous driving exploration within Alibaba and determined the commercial application direction: fully unmanned logistics robots.

Alibaba subsequently established Little Donkey Intelligent Technology, with Wang Gang serving as General Manager. At the 2020 Apsara Conference, Little Donkey was officially unveiled to the public, entering mass production and commercial operation stages; it is one of DAMO Academy’s most perceptible and topical innovative products since its inception.

In January 2022, Wang Gang was reported to have left Alibaba to start his own business. His newly founded company, Xinshengji Intelligent Technology, focuses on commercial cleaning robots empowered by large models.

According to Tianyancha, this company completed two rounds of financing this year, with investors including Paradise Valley Capital, Puhua Capital, and Baiquan Capital.

Besides autonomous driving, another major hotspot in the AI field—embodied intelligence—also features the presence of Fei-Fei Li’s former students.

Lu Cewu, a professor at Shanghai Jiao Tong University, had Fei-Fei Li as his postdoctoral advisor during his tenure from 2015 to 2016.

In 2013, he obtained his Ph.D. in Computer Science from The Chinese University of Hong Kong under advisor Jia Jia.

He then conducted two years of postdoctoral research at the Hong Kong University of Science and Technology under Professor Deng Zhiqiang.

In 2015, he received a letter of recommendation from Fei-Fei Li and was ultimately invited to join her lab for further postdoctoral studies.

At that time, embodied intelligence was in its embryonic stage; Fei-Fei Li and her students were discussing the start of robot research.

During this period, Lu Cewu met his fellow student Zhu Yuke.

Zhu Yuke graduated with a bachelor’s degree in Computer Science from Zhejiang University in 2013 and subsequently pursued master’s and doctoral degrees at Stanford University.

After joining Fei-Fei Li’s group, Zhu Yuke initially worked on visual knowledge bases before switching to robotics alongside Lu Cewu in 2015.

Later, both achieved great success in the field of robotics.

After returning to China, Lu Cewu joined the Computer Science Department at Shanghai Jiao Tong University and is currently a professor there.

In 2018, he was selected by MIT Technology Review as one of “35 Innovators Under 35,” and based on his outstanding contributions to embodied intelligence, he received the Scientific Exploration Award in 2023.

To date, he has published over 100 papers in high-level journals and conferences such as Nature, Nature Machine Intelligence, and TPAMI, either as corresponding author or first author.

Beyond academic research, he also spans the industry: In 2023, he co-founded **Qio

I think the narrative relies heavily on pedigree rather than independent technical validation of these students’ current impact. From the paper, citing citation counts from graduation ignores how quickly AI metrics inflate and deflate in subsequent years. One caveat: linking commercial success directly to academic lineage assumes a causal link that industry churn often disproves.

The Fei-Fei Li Effect: How a Chengdu Student Reshaped AI’s Trajectory

The core technical claim here is that embodied intelligence is no longer just an academic exercise but a commercially viable sector capable of attracting hundreds of millions in venture capital, driven by the specific research lineage of Stanford’s Computer Vision group. This narrative would be falsified if the recent funding rounds for these startups were found to be based on proprietary data access rather than genuine algorithmic breakthroughs or reproducible generalization capabilities.

The Commercialization of Embodied AI

I followed the release from ngche Intelligent, a company co-founded by Zhu Yuke, who serves as Chief Scientist. They are dedicated to developing embodied intelligence systems and related tools and platforms. The latest news is that in September this year, the company completed a Pre-A round of financing worth hundreds of millions of yuan.

This round was jointly led by Prosperity7 Ventures and GF Xinde, with participation from Zeyu Capital, Sinovation Ventures, Qiji Chuantan, Plug and Play China, and MFund.

I think the valuation metrics for embodied AI startups often rely on projected utility rather than current deployment scale. From the paper, pre-A rounds of this size in hardware-adjacent AI suggest high risk tolerance from investors, not necessarily technical maturity.

Academic and Industry Convergence

As for Zhu Yuke, after obtaining his Ph.D. from Stanford University in August 2019, he is currently succeeding in both academia and industry:

On one hand, he serves as an Assistant Professor in the Department of Computer Science at the University of Texas at Austin and Director of the Robot Perception and Learning (RPL) Laboratory;

On the other hand, he co-leads NVIDIA’s GEAR Lab (researching general-purpose embodied agents) with another fellow student, Jim Fan.

That’s right, NVIDIA scientist Jim Fan is also Fei-Fei Li’s student.

Jim Fan, who graduated with a bachelor’s degree from Columbia University, was an outstanding graduate representative that year, receiving Columbia’s Illig Medal.

From 2016 to 2021, during his Ph.D. studies at Stanford University, he conducted research in deep reinforcement learning, robotics, computer vision, and other fields under Fei-Fei Li’s guidance.

Interestingly, during this period, he also became OpenAI’s first intern (working with Ilya Sutskever and Andrej Karpathy).

Upon graduation, he joined NVIDIA, rising to the position of Senior Research Scientist, during which time he led several embodied intelligence projects:

• Eureka: Using GPT-4 to generate reward functions, teaching robots to complete more than thirty complex tasks; it was rated as one of “NVIDIA’s Top Ten Projects in 2023”;
• Voyager: The first large language model (LLM)-driven agent capable of skillfully playing Minecraft.
• VIMA: The first multimodal LLM equipped with a robotic arm, introducing the concept of “multimodal prompting” to robotics learning.
• MineDojo: An open-source framework that transforms Minecraft into a playground for AGI research, winning the Best Paper Award at NeurIPS 2022.

It wasn’t until February this year that Jensen Huang appointed him and Zhu Yuke (both born in the 1990s) to jointly lead the GEAR Lab.

One caveat: projects like Voyager succeed in simulation but face significant hurdles when transferring to physical robotic hardware. I think the reliance on GPT-4 for reward generation introduces a dependency on external API stability and cost structures.

The Spatial Intelligence Boom

At this point, it is evident that Fei-Fei Li’s students are spread across various fields within AI, each demonstrating a strong entrepreneurial spirit.

Meanwhile, Fei-Fei Li herself officially announced the founding of World Labs in September this year, targeting spatial intelligence.

Less than four months after its establishment, the company’s valuation has already surpassed $1 billion.

Interestingly, one of the co-founders of this company is also a former student of Fei-Fei Li.

Justin Johnson, who completed his undergraduate studies at Caltech and earned his Ph.D. in Computer Science from Stanford University.

During his doctoral studies, he and Karpathy were paper-writing partners, spending time as fellow students at Stanford.

He also co-conceived the initial version of CS231n with Fei-Fei Li and served as one of the primary instructors for the course between 2016 and 2019.

After graduation, he joined the faculty at the University of Michigan as an Assistant Professor in the Department of Computer Science and Engineering.

At the same time, he was a Research Scientist at Meta FAIR.

From the paper, a $1 billion valuation within four months reflects market sentiment toward spatial computing more than proven product-market fit. One caveat: the overlap between academic instruction and industry leadership creates potential conflicts in open-source vs. proprietary data usage.

I think the reliance on established academic pedigrees does not guarantee that these models generalize beyond their specific training distributions. From the paper, assuming that internships at major tech firms equate to robust, reproducible research practices is a dangerous oversimplification. One caveat: we must question whether the cited metrics truly reflect real-world utility or just dataset overfitting.

The Human Infrastructure Behind the Models

At Stanford Vision Lab, we discovered more Chinese faces. This observation isn’t merely demographic; it points to the specific lineage of researchers driving current computer vision advancements.

De-An Huang, who received his Ph.D. in Computer Science from Stanford University in 2020 under the supervision of Fei-Fei Li and Juan Carlos Niebles, represents a key node in this network. He earned his master’s degree in Robotics from Carnegie Mellon University. During his doctoral studies, he interned at Microsoft, Facebook, and NVIDIA. Since graduation, he has been working as a Research Scientist at NVIDIA.

Alan Zelun Luo is currently a fifth-year Ph.D. student in the Department of Computer Science at Stanford University. He completed his undergraduate studies in Computer Science at the University of Illinois Urbana-Champaign before pursuing his master’s and doctoral degrees at Stanford. Although he has not yet graduated, he has an impressive internship history, having interned at Nvidia, Facebook, Google, Amazon, Yahoo, and other institutions.

Yanan Sui, currently an Associate Professor at Tsinghua University, specializing in machine learning, neural engineering, and robotics. He graduated from Tsinghua University with a bachelor’s degree in Biomedical Engineering in 2010, followed by doctoral studies and postdoctoral research in Computer Science and Neuroscience at Caltech. In 2020, he was listed as one of China’s “35 Innovators Under 35” by MIT Technology Review. He currently serves as a Area Chair for the international conferences NeurIPS and ICLR, and is on the editorial board of the Journal of Biomedical Engineering.

Serena Yeung, currently an Assistant Professor at Stanford University, focusing on the application of visual AI in healthcare. She leads the Medical Artificial Intelligence and Computer Vision Lab (MARVL) at Stanford and serves as the Deputy Director of Data Science for the Center for Artificial Intelligence Medicine and Imaging (AIMI). From 2006 onwards, she completed her bachelor’s, master’s, and doctoral degrees in Electrical Engineering at Stanford University. She also spent one year as a postdoctoral researcher at Harvard University.

How a Chengdu Student Reshaped the Trajectory of AI

The Optimism Caveat

In her new autobiography, The Worlds I See, Fei-Fei Li posits that scientific optimism is not passive. She argues explicitly that waiting for opportunities yields no results, framing success as something earned through effort rather than luck. This narrative sets the stage for why she launched an entrepreneurial venture during the peak of large language model hype.

I think optimism without a reproducible methodology is just hope, and hope doesn’t scale models. From the paper, the “effort” required to build AI infrastructure likely exceeds what this memoir details. One caveat: attributing breakthroughs solely to personal dedication ignores the massive public funding behind them.

From Chengdu to the Laundry Shop

Li traces her arc from a student at Chengdu No.7 High School to the United States, where she worked in a laundry shop to support her family. This biography acts as a window into the “North Star” guiding her career, offering anecdotes from before and after the AI renaissance. It also highlights the role of her Chinese parents in shaping her trajectory toward the forefront of the current AI era.

I think personal narratives often obscure the systemic barriers that most students never overcome.