From science projects to AI programming: Neel started learning Python when he was in the third grade. In eighth grade, he started using machine learning libraries while doing a science project that distinguishes trash from recyclable material and compost. Learning initially began by watching YouTube videos, but he soon realized that it was much faster to read the documentation. He says, “I knew how to program derivatives before I knew the word derivative.” By ninth grade, he became familiar with math, i.e., derivatives and matrix multiplication, required to understand machine learning algorithms. He built an application that tried to predict earthquakes in California. It had very low accuracy. However, he ended up creating a large dataset on earthquakes in California and made it public.
Organizing hackathons: In eighth grade, Neel attended a local hackathon and enjoyed it. He made new friends and joined the Slack group of their hack club. This led him to a workshop on organizing hackathons in San Francisco during the summer after eighth grade. Neel says, “I think this was really important for me,” because it was the first time he realized it was possible to raise $ 10K–$15 K to organize a hackathon by just sending emails. Subsequently, he, along with others, organized many hackathons, including some at Angel Hacks. In 2021, at age 16, Neel participated as an organizer and a hacker in The Hacker Zephyr, the world’s longest (3502 miles) cross-country hackathon, in which 42 kids traveled from Vermont to Los Angeles on a train converted into a hacking playground.
Independent AI research: At age 18, Neel got a Davidson scholarship to do research on creating new AI algorithms to make artificial photosynthesis & carbon capture a reality, and presented his work at the AAAI23 conference. The same year, he presented a single-author paper at the Neural Information Processing Systems (NeurIPS-2023) Conference on language-to-material generation architecture for crystal discovery. Neel managed to get his own grant for traveling to New Orleans for NeurIPS. He says, “(At NeurIPS) I met tons of people who are great friends.”
Leaving UCLA and Standard Intelligence: During the Sophomore year at UCLA, Neel tried to balance three things: classes, research, and social life. However, as the year proceeded, he realized, “I am not doing well socially, I am not doing well research-wise, and I’m not doing well class-wise.” So he decided to take a gap year and joined Standard Intelligence (SI), an AGI startup in San Francisco. In the past year at SI, Neel has contributed to creating a video action model (FDM-1) that is trained on 11 million hours of video and operates on a context window of 1 hour 50 minutes. This is likely to be an order of magnitude more than what a typical LLM like ChatGPT has been trained on. And they managed to build their own data storage cluster at a fractional cost of a typical cloud provider. It is not surprising that the team at SI has raised $75M with a $500M valuation in a funding round co-led by Sequoia and Stark Capital. Neel also managed to create a demo of their video action model to drive a car in a parking lot. Neel says, “The demo attracted a lot of attention from the investor community as well as people like Andrej (Karpathy).” At this stage, he admits, “Things are going so well, there’s no reason to go back (to UCLA)”.
Being intentional and taking upside-down risks: I asked Neel for advice to the college kids who are worried about finding a job in the AI world. Neel says, “A lot of college or high school students love to defer their decisions to other people. Like, my mom told me I need to go to college.” Neel’s suggestion is to be intentional about what you do. Ask yourself, “Why am I going to college?” If the answer is “to study,” then focus on studies; if the answer is “to make friends,” then focus on making friends. He feels people should identify what their interests are, search up interesting problems in the field, do something about it, ignore everyone telling you, you can’t do it, make something, and then cold email people with the cool thing they made. “You want to maximize upside risk,” says Neel. When you send 100 cold emails, the chance is high that one will respond. He says, “By taking small risks consistently that offer you no negative consequences, you’re always going to do something interesting with your life.”
