If you're going through a rough patch in your life and feel like no amount of effort is enough, perhaps Soumith Chintala's story can give you some strength. The man who later created PyTorch and became the VP of Meta did not have a glamorous start. Soumith comes from Hyderabad Public School in India. His academic performance is not bad, but the fact that he is "bad at math" has always been a shadow over him. When he took the college entrance exam, he didn't get into a top-tier university, but instead entered what Indians consider a "second-tier university"—VIT (Vellore Institute of Technology). When he graduated from university, he scored 1420 on the GRE (out of 1600 in the old version), which was a good score. However, when he applied to 12 master's programs in the United States, he was rejected by all of them. Every school—without exception—rejected him. He thought for a moment, and could only describe his state of mind in one word: "fuckit.jpg" So he gritted his teeth and applied for a J-1 exchange visitor visa, flying directly to Carnegie Mellon University (CMU) in the United States without any plans, just thinking, "I'll go first and see what happens." Once he arrived in the United States, he applied to 15 more master's programs. This time the result was better: Only USC accepted him, and NYU (New York University) offered him a supplemental admission in 2010. After arriving at New York University, he met someone who changed his life: Yann LeCun – at that time, he had not yet won the Turing Award, nor become an AI superstar. At NYU, he also met another important mentor, Pierre Sermanet, whom Soumith described as "one of the kindest people I have ever met." It was here that he began to explore and fall in love with open source. After graduation, almost all of his resumes disappeared without a trace. All the companies, including DeepMind, rejected him. The only opportunity offered him was a test engineer job at Amazon—not a research position, nor the AI role he dreamed of. During his low point, his mentor helped him out by introducing him to a small startup called MuseAmi. But this did not immediately turn his life around for him. He was then rejected by DeepMind twice more, making it a total of three rejections. Because of the J-1 visa's "two-year home residency requirement," he was initially unable to obtain an H-1B work visa. He spent several months communicating back and forth with USCIS and the US State Department before finally obtaining a waiver that allowed him to remain in the country. During that period, his self-confidence plummeted. Between 2011 and 2012, he achieved a remarkable result at the time: It boasts one of the world's fastest AI inference engines running on mobile devices. Yet, even with this, he continued to be rejected by DeepMind. Ultimately, what truly changed his fate was the open-source project Torch7 that he continued to work on. He mustered his courage and sent another email to Yann LeCun. It was this email that led him to join Facebook's FAIR Labs. When he first joined the company, he almost "failed" in the training camp because he got stuck on an HBase task for a long time. But soon, he seized a crucial opportunity: While senior L8/L9 engineers were struggling with ImageNet training, he, as an L4 engineer, solved a critical bug in the numerical/hyperparameter settings. This was his first "big victory". At Fair, he led a small team of only three people, who later created PyTorch together. But the story didn't go smoothly. Due to internal political reasons, the management once considered shutting down the PyTorch project. Soumith was so angry that he sat in a bar and cried (his original words were: cries-at-bar.jpg). Fortunately, some people persisted in supporting the project. In 2017, PyTorch was officially released and later became one of the world's most popular deep learning frameworks. Around the same time, he also obtained an EB-1 Extraordinary Ability green card. What happened next became the history we know today. From 2005 to 2017, he experienced: He was bad at math, applied to a second-tier university for a master's program twice, was rejected by all companies, was rejected three times by DeepMind, and was also rejected by Google. His visa crisis nearly led to the cancellation of his project, and he almost died in the early days of Facebook. For almost twelve years, it was a series of failures. But he did not give up. Ultimately, he became the father of PyTorch, Vice President of Meta, and one of the most influential people in the global AI field. Soumith later replied to the tweet, adding a separate message: "These are all true. But I still owe a lot of people a thank you." His advisor at NYU was Pierre Sermanet, who was a doctoral student at the time. Soumith said he was one of the kindest people he had ever met. Yann LeCun gave him opportunities twice when he "could hardly see a way out for AI." One time was at NYU, and the other time was at Fair. It was Praveen Garimella who persuaded him to go to IIIT for his final year of undergraduate studies. After he failed all his master's exams, he was encouraged to "go to CMU first and don't give up." And his parents: Vithal Chintala and Rajani Chintala. He described himself as growing up in a middle-class family burdened with a lot of debt. My parents only gradually achieved financial freedom after 2010. But before that, they had already gritted their teeth and supported their son in pursuing an "insecure" dream path, even though it was beyond their means. Instead of giving him a stable job. Soumith said that this is a truly great way of raising children. He also specifically thanked the original tweeter, Deedy. It was the other person who took the time to gather these "details" scattered throughout his life. The fragments were unearthed and strung together to form a complete story. Soumith concluded with a very calm statement: I believe that everyone who is now "sitting on the throne of success" There are many struggles behind the scenes. Life is never easy.
Loading thread detail
Fetching the original tweets from X for a clean reading view.
Hang tight—this usually only takes a few seconds.
