Soumith Chintala from Hyderabad made it to VIT Tier 2 because he was bad at math, but now his PyTorch is powering all the AI ​​boom

Machine Learning


You may have never heard of Soumis Chintala. Unless you’re in the world of AI or machine learning, in which case you’ve 100% heard of him. Still, for some reason, if you don’t know Soumith, you’ve probably heard of something he co-created in 2016. This is PyTorch, a machine learning tool that is as common and familiar to everyone in the AI ​​world as MS Word is to writers. Or Photoshop is for designers. Yes, Soumith is that important that now that he has left Meta, the company that built and maintains PyTorch, his inspiring story is reaching the wider world.

Interestingly, unlike many others, Somis may be one of the most recognizable names in the world of AI. There is chatter that reveals how he wasn’t very good at maths in school in Hyderabad, or how he didn’t go to IIT-B or IIT-M. Instead, he went to VIT, a university considered to be India’s second technical university. When I was in college and just entering the workforce, I had a hard time finding a job. But through his tenacity and perseverance, Soumis built a reputation that will forever be part of AI history.

Meta’s AI fixer, Soumith, recently announced his withdrawal from Meta and the PyTorch project, so his story has come to light.

I’m not good at math but I’m an AI genius

Soumith, who grew up in Hyderabad and studied at the Hyderabad Public School, never admitted that she struggled with mathematics as a teenager. After completing his schooling, he enrolled in VIT Vellore to pursue a Bachelor of Technology degree. This is not an IIT or a prestigious university, but a 2nd tier engineering college.

After completing his degree, Soumis planned to pursue a master’s degree in the United States. According to X’s technical circle, despite scoring 1420 points on the GRE, he was rejected by all 12 universities he applied to. But he didn’t give up, flying from India to Carnegie Mellon University on a J-1 visa with no clear plans, and reapplying to 15 more universities. He eventually got into the University of Southern California and New York University, which offered him late admission in 2010.

Even after earning her master’s degree, Soumith’s path was not smooth. He was rejected from almost every job he applied for, including Google DeepMind. I was rejected not once, but three times. His first real break was a test engineer role at Amazon, but it was far from the research-driven job he had hoped for. His PhD supervisor then helped him secure a job at a small start-up company, MuseAmi, but visa issues, confidence struggles, and career uncertainty made those years particularly difficult.

From 2011 to 2012, he demonstrated his technical prowess by building one of the fastest AI inference engines for mobile devices. However, despite his accomplishments, DeepMind rejected him again.

Great progress towards AI

The turning point came when his open source work connected him with Yann LeCun, now Meta’s chief scientist and also rumored to be leaving the company. Side note: What would happen if LeCun and Saumith teamed up to launch an AI-focused startup outside of Meta?

But back to the topic. Somis and LeCun met long before the French AI researcher became a household name in the field. That connection, and his work at Torch7, led to Soumith joining Facebook AI Research (FAIR). But the transition wasn’t easy. He slogged through Facebook’s engineering bootcamp, struggled with internal tasks, and while senior engineers struggled to solve ImageNet training problems, Soumith, then a Level 4 engineer, solved critical problems with numbers and hyperparameters. That was his first major breakthrough.

Things quickly escalated. He went on to lead a small Torch7 team of three people and eventually co-created PyTorch. PyTorch now forms the backbone of nearly all modern AI research across industries, including companies like OpenAI and Google.

“I led PyTorch for almost eight years, taking it from zero to 90 percent+ AI adoption. Walking away was one of the hardest things I’ve ever done. But I leave with a full stomach,” Soumith wrote in a farewell note on X. “PyTorch powers foundational models that redefine intelligence. PyTorch is being taught in classrooms from MIT to rural India. Is it the tool I dreamed of having access to? Yes, it’s a barrier to entry.” I wanted to lower it, but soon?

Quit Meta and PyTorch

As previously mentioned, after working with Meta for over 11 years, Soumith recently announced his retirement, saying he wanted to explore “something small” and “something new.” The last day will be November 17th. He admitted that leaving “one of the most leveraged seats in the AI ​​industry” wasn’t easy, but added that he doesn’t want to live with the regret of never having tried something outside of Meta.

It’s hard to say quit something that you’ve built and developed into a phenomenon. It’s even harder to say quit when it’s on top of the world. But apparently Soumis, who is likely used to forging new paths, even if they are more difficult, wants to challenge himself again. Just like when you weren’t good at math or didn’t get the job you wanted. “It’s very hard to leave me leading the software layer that powers the entire AI industry. All the major AI companies and hardware vendors are on speed dial. It’s really hard to let go of that kind of power. But in the end, curiosity won out in my mind,” he says, attributing his success to his curiosity about the wizardry of mathematics and IIT degrees.

– end

Publisher:

Divya Bhati

Publication date:

November 13, 2025



Source link

Leave a Reply

Your email address will not be published. Required fields are marked *