I recently finished listening to the audiobook version of Fei-Fei Li’s memoir, *The Worlds I See*. Here are some thoughts on it.
Three keywords first drew me in: immigration, AI, and Asian identity. But what stayed with me was a fourth: curiosity.
The book follows two parallel threads. One is her personal journey: born in Beijing, raised in Chengdu, she moved to the United States at fifteen. She worked to help her family pay off their debts while trying to find her place in a completely unfamiliar education system. She went on to study physics at Princeton, earn a PhD from Caltech, and eventually lead Stanford’s AI Lab. The other thread is the scientific history of how machines learned to “see” the world. At its heart is ImageNet, a public benchmark dataset containing tens of millions of images labeled and categorized so that different algorithms could be compared fairly.
One detail in the story of ImageNet really stuck with me: while many people were focused on competing over algorithms, Fei-Fei Li’s team devoted its energy to the scale and diversity of the data. Early on, this approach seemed clunky, even not very “smart.” But what happened? Later, large-scale data, GPUs, and deep neural networks came together, and computer vision made a genuine leap forward.
This changed how I think about “technological breakthroughs.” The narrower the training material, the more a model can only be smart in a narrow world. The value of broad input is that it forces you to keep distinguishing what an object is from the background you’ve simply grown used to.
This idea has also influenced how I plan my own vocabulary learning. I have nearly 800 words in my vocabulary notebook… I talk about this in more detail on the podcast.
This episode is for anyone curious about the history of AI, or currently reading this book. The link is below 👇
https://redcircle.com/shows/936a3212-92e5-41c7-b6f9-39e3596adeb8/episodes/6d6ed2ed-7068-47c7-a7b5-84c34cb6a5f8
Three keywords first drew me in: immigration, AI, and Asian identity. But what stayed with me was a fourth: curiosity.
The book follows two parallel threads. One is her personal journey: born in Beijing, raised in Chengdu, she moved to the United States at fifteen. She worked to help her family pay off their debts while trying to find her place in a completely unfamiliar education system. She went on to study physics at Princeton, earn a PhD from Caltech, and eventually lead Stanford’s AI Lab. The other thread is the scientific history of how machines learned to “see” the world. At its heart is ImageNet, a public benchmark dataset containing tens of millions of images labeled and categorized so that different algorithms could be compared fairly.
One detail in the story of ImageNet really stuck with me: while many people were focused on competing over algorithms, Fei-Fei Li’s team devoted its energy to the scale and diversity of the data. Early on, this approach seemed clunky, even not very “smart.” But what happened? Later, large-scale data, GPUs, and deep neural networks came together, and computer vision made a genuine leap forward.
This changed how I think about “technological breakthroughs.” The narrower the training material, the more a model can only be smart in a narrow world. The value of broad input is that it forces you to keep distinguishing what an object is from the background you’ve simply grown used to.
This idea has also influenced how I plan my own vocabulary learning. I have nearly 800 words in my vocabulary notebook… I talk about this in more detail on the podcast.
This episode is for anyone curious about the history of AI, or currently reading this book. The link is below 👇
https://redcircle.com/shows/936a3212-92e5-41c7-b6f9-39e3596adeb8/episodes/6d6ed2ed-7068-47c7-a7b5-84c34cb6a5f8