Paul Graham On Startups, Ambition, and Great Founders

Paul Graham says OpenAI is the new Google

Paul Graham's 2012 essay on frighteningly ambitious ideas has largely come true, and OpenAI is the new Google he predicted. A new Google arrives when the incumbent's model becomes obsolete, and OpenAI was incubated at YC as a Sam Altman side project. Ambition is mostly inborn, so the job is to find formidable founders, not to train them.

Takeaways

• OpenAI is the new Google. In a 2012 essay Graham coined 'frighteningly ambitious' and listed a new Google. A new Google arrives when the incumbent's model becomes obsolete, and OpenAI, incubated at YC as a Sam Altman side project, is it.
• Ambition is mostly inborn. Graham knew dutifulness alone was not enough for a startup's obstacles, but founders rarely become ambitious during a batch. Sam Altman was formidable the first time they met, before YC accepted him. Those who seem unambitious were often trained not to show it.
• Fear of failure drives founders. What motivates founders is the fear of looking like a fool or the server crashing, not the prospect of becoming a billionaire. They learn they are billionaires only when they add up their shares at the last round valuation.
• AI arrived as a bullshitting human. The expectation was a perfect fly working up through mice and cats to humans. Instead the first AIs arrived as a full human full of shit, like an undergraduate bullshitting a paper. AI stays jagged, so Graham says 'we're on the smear'.
• Shipping speed still predicts success. Graham says shipping speed remains the best predictor of whether a startup works, and many companies in the current batch ship too slowly. A startup is a poor credential next to a Harvard degree, since there is no easy major.
• The batch is frighteningly ambitious. The current batch includes a startup delivering cargo by intercontinental ballistic missile and one doing on-demand research for people with cancer, inflicting a death of a thousand cuts. It helped Sid from GitLab, whose essay was titled 'I went founder mode on my cancer.'

The lean startup argument

• Patrick Collison said at Startup School that lean startups may be dead, because money is easy to raise and AI lets you run many agents in parallel.
• Graham never understood what a lean startup was and has not read The Lean Startup, comparing himself to Tolkien not reading other fantasy books.
• Technology keeps getting cheaper, with inference prices at a given level falling about 30x a year, though tokens are expensive now because of a GPU shortage.
• Graham says startups can still begin on little money by hitting milestones. Starcloud's Philip wrote a white paper and booked a launch, then raised the money.