@rohanpaul_ai: https://x.com/rohanpaul_ai/status/2061230948658954707

TL;DR: Sam Altman recalls his Stanford days, startup experiences (Loopt), how Y Combinator changed the startup landscape, and the story of OpenAI starting as an unpopular idea.

Stanford Days: Steam Tunnels and All-Nighters

Sam has fond memories of his first day at Stanford. He mentioned he hadn’t been back to campus in a long time, but after that event he really wanted to walk around again. Early in his freshman year, he and a group of friends went steam tunnel exploring—a rite of passage considered cool at the time. Someone in the group forgot to put the manhole cover back, and a professor’s car fell into the uncovered hole. Sam thought he’d be expelled, but in hindsight that experience is hilarious. He even said that working all night in the Gates building basement, which he didn’t enjoy then, has become a good memory.

Sam said if he could do it over, he would join fewer clubs and spend more time building things with friends.

From “Face Club” to Loopt: A Crazy Idea in 2005

The host found a business seminar handout from spring 2005 at Stanford’s “Face Club” (a student startup club). The document described a project founded by four students: “Vienno and Viendo is a mobile service that provides personalized information about popular places, a map of nearby friends’ locations, and suggestions for new friends and dates nearby who might be a good match.” At that time, the iPhone didn’t exist, and this idea was considered crazy. That project was the precursor to Loopt.

Sam recalled that back then he wasn’t very interested in starting a company, but the emergence of Y Combinator completely changed how hard it was to fund a project. Before YC, starting a company required writing a business plan, entering business plan competitions, finding an MBA partner, and networking with VCs. Sam described business plans and competitions as “a waste of time” and “the ultimate waste of time.” The birth of Y Combinator made starting a company much simpler.

Contacting YC and the Decision to Drop Out

Sam was one of the first applicants to YC. He remembers filling out a strange form built by Paul Graham—you had to paste the content into an email without changing any formatting because there was a weird parser.

During college, Sam sometimes went to class, sometimes didn’t, mostly hung out with friends and tinkered with things. It was a simple, good, boring time.

As for dropping out, Sam didn’t feel it was risky. He emphasized that survivor bias is real, but Stanford had a policy allowing students to take up to eight years off and return, which made it easier to convince his parents. He believes in doing whatever is most interesting, has the highest learning efficiency, and creates the biggest impact—whether that’s school, starting a company, or doing AI research. Over the past two decades at least, the relative value of the default path has declined while the space of alternative paths has grown. At the time, he didn’t see dropping out as crazy or risky.

Fundraising and Stanford’s Pipeline

Loopt later raised over $30 million (a huge sum back then; Sam joked that now it would probably buy only 17 GPUs). At that time, Stanford was connected to the entire VC pipeline, and Y Combinator was also a big help. Although there were fewer VCs and less capital than today, they focused intensely on Stanford as a source of ideas and talent.

But Sam thinks that Stanford-specific advantage is completely gone now. With the maturity of YC and the VC industry, there is so much capital looking to invest in startups that students from other schools can equally enter the process.

Focusing on AI: From “Completely Didn’t Work” to “An Asteroid Is Coming”

Sam studied AI at Stanford, but back then it completely didn’t work. He worked at an AI lab during the summer between his freshman and sophomore years, then got distracted by other startup ventures and later became an investor. However, he always believed AI was one of the most important things for the future. He emphasized that early in your career, you should form a superset of things you care about, constantly update your worldview as you gain experience, and keep your head up.

When the AlexNet paper came out in 2012, Sam, like many others, thought it was “cool,” but he didn’t fully update his mental model. In the following years, he observed that deep learning models kept improving as they scaled, and gradually realized that “an asteroid is coming and people aren’t paying enough attention.” By 2014, they started seriously discussing doing something about it.

Building an AGI Lab: An Unpopular Idea

Sam likes unpopular ideas. His career as an investor was about betting on ideas that were undervalued but could be huge if successful. At the time, the old guard in AI said, “You’re crazy,” “You’re a fraud,” “AGI will take 100 years.” But Sam’s team pushed only one idea: scaling deep learning seemed important. Although they didn’t yet know that the scaling laws would be so beautifully predictable, putting in more compute generally yielded better results.

They decided to push that idea as far as possible and see what happened. They were also worried about alignment issues and societal impact. Initially they only thought it would be a research lab and didn’t know they would build a product.

Sam still finds deep learning, especially unsupervised learning in models, a bit magical. The fact that next-token prediction can be used to discover new scientific ideas makes logical sense, but seeing it actually work still feels crazy.

OpenAI’s Pitch and Co-founders

Sam said OpenAI’s pitch was very simple: “Something important is happening with deep learning and scale. We don’t know what we’ll do with it or where it will go. We’ll push in this direction, good things will happen, and others underestimate it.” It wasn’t a great pitch, but it worked.

Regarding the co-founders: Sam had known Greg for a long time—Greg was the first employee at Stripe, and Sam was an early investor; they stayed in touch for about ten years. Ilya was the genius and godfather of the field; Sam tracked him down after a conference and started discussing starting a company. Wojciech was met through the Bay Area AGI geek scene at that time. John was introduced by someone else, and then there was Elon.

Sam emphasized that OpenAI was never supposed to be a company; they had no idea it would become one.

From Loopt to OpenAI: A Completely Different Feeling

Starting Loopt felt like still being in college: staying up late chatting or working on computers—surprisingly similar to college life. OpenAI, on the other hand, was a very strange thing—it cannot be overstated how much it was not a company.

Source: https://www.youtube.com/watch?v=FjlymGBt-vY