Sam Altman: A 10-Week-Old Startup Today Is Unrecognizable

A 10-week-old startup today looks "totally different," Sam Altman told the interviewer, and one resembling a 10-week-old startup from 10 years ago is in trouble.

In his account, AI is the primary driver — specifically what a small team of people can now accomplish and the speed at which they can do it. The shift goes beyond what is possible: it also covers what a company must do to stay competitive given how much the world has moved.

The product was made for a world in which AI must be able to use documents, presentations and spreadsheets as a first-class consumer.

The work would have taken a startup "a year's worth of work" not long ago, he said, and admitted, "I don't even know what they're called."

The ability to operate in chaos comes down to practice. A handful of things, no matter how well they are understood intellectually, take many repetitions before a person can handle them emotionally — trusting that a problem will be figured out, that it will not be the thing that kills you, even without knowing the solution yet.

He named this as a real weakness of young founders: they have not had the career experience of reaching emotional peace with that uncertainty, so the early days are very hard for them. They do eventually learn it, but "at great pain and cost and unforced errors."

His rule of thumb is that in a high-stress, high-impact job, most people will learn to function well through chaos. Improving is different from being taught.

While running Y Combinator he could always tell a brand-new founder from an experienced one by their emotional state when describing a problem, even without knowing how old the company was.

As a kid he was fascinated by the Industrial Revolution and assumed a cluster of technologies had happened to scale together. Looking at it now as someone who has spent years thinking about business, his read is that "the important invention was the joint stock corporation."

Before that structure existed, companies were family-run and operated on trust — in his words, "you had to know everyone." Sovereigns granted corporations a status that had not existed before: not the power of a state, but "much more than the power of a person."

He identified incentive alignment and liability protection as the two mechanisms, along with the ability to amass capital. Pooling capital let people fund speculative technology development and produced serious financial systems and new instruments — and today millions of people can hold shares in a single company. As the most under-watched chart, he cited the fall of extreme poverty and infant mortality over the past hundred years.

The interviewer raised the statistic that a very high percentage of CEOs are sociopaths or psychopaths, and argued that capitalism may be the first system to realign self-serving people's incentives with society's, so that both they and society win.

Altman's response was partial agreement with a line drawn. He definitely knows a lot of sociopathic CEOs, but would not put the best ones he knows in that category. He separated high-ego CEOs — people with incredibly high opinions of themselves, whose antics might once have been purely annoying — from sociopaths. Because people can buy their shares, there is tolerance for the annoyance and antics, "some aligning thing there." There are many CEOs of very large companies he would not call sociopaths at all.

What drives those people varies, he said: seeing how good they can get at the game, how much better they can perform year over year, and the work being "intellectually quite stimulating."

Major AI alignment and safety issues remain to be solved, along with major economic questions such as the future of jobs.

In that light, he defined the fight of the current moment as whether the world heads toward AI authoritarianism or liberty.

He characterized the alternative as deciding that because the safety and economic problems are very real, one single model should be "the machine god" — or the company associated with it.

His stated principle is that every time humanity has traded liberty for safety, it has been a long-term net loss. He wants the technology placed in people's hands, with society free to express ideas and use it as it wishes, and he accepts guardrails "according to some guardrails, with some guardrails," while keeping a lot of power and potential in society. He said he really cares that they do it that way.

Three years ago, Altman said, GPT-4 had just launched and "the world was melting down." World leaders were asking whether they needed to take control and shut things down.

He described the trip as a very inconvenient pair of overnights, coming when he had a new baby and "whole mess." With the company not big at the time, his conclusion was that it "probably going to have to be me." He feared it was about to go very badly, and said he hates long flights and jet lag.

Brian Chesky advised him to do it — Airbnb had done about eight cities.

Moving between places quickly let him see subtle cultural differences he would otherwise have missed. Each day ran about 14 hours non-stop, and he said he is "not an extroverted person" yet in some cases met hundreds of people in a day.

By the end of the trip, global tension had come down; the world had been very nervous at the beginning.

Knowing when he would get home helped him keep going. He now does smaller trips once or twice a year.

His stated lesson is a bundling rule: he would much rather put a bunch of international travel together in a seven- or ten-day chunk than a series of one-day trips spread out. He also said many people overestimate the risk in most actions — the real risk is buying call options on Robinhood, less risk in getting on planes.

His key observation: "this person has never not had a boss in their life" — always a parent, a teacher or a manager telling them what to do. He also says they were punished for being too creative, wanting too much, or thinking too big.

Most countries or cultures have a phrase for being too ambitious, and he names "tall poppy syndrome." The message, he says, came from parents, teachers and friends from when they were quite little, and undoing it takes real time.

His remedy is small repeated wins: do something you did not think you could do, see it work, then try it again. He also pushes back on the Steve Jobs comparison — "I don't do inspirational speeches."

Altman cites Jeff Bezos calling AWS "a business miracle" because it faced no serious competition for about seven years. He calls OpenAI another business miracle, and says there is "clearly something here about why big companies get sclerotic and set in their ways."

For the world, he says, it is good that big companies do not stay dominant forever — "that'd be really bad."

Going back to explain why OpenAI was allowed by giants with huge amounts of capital and talent to do what it has done in 2019 is, in his words, hard to explain.

Microsoft's decision is easier for him to account for: Google had DeepMind, and Microsoft did not have an AI bet. Big companies, he says, feel they should have something in play in all major areas of technology.

Altman says he knew intellectually that the product moment would arrive at some point, but had no idea when or that it would be that moment. He decided to pretend to himself, and "successfully deluded myself" into acting like it was not going to happen.

Running the research lab, he says, was not stressful at all — "the coolest, most fun, most amazing job and lifestyle setup I could possibly have imagined." It put him in "this like front row seat" to the smartest group of people doing "probably the most important work that has happened in like the last century, maybe longer."

Asked about a colleague posting about "another reset" on Twitter, he says there have been two giant form factors so far: chatbots, then coding agents.

Coding agents, in his words, "are just going totally nuts." A third wave is coming "pretty soon": persistent agents, chiefs of staff, co-workers, colleagues.

Technology has long promised less work and more leisure, and he notes people do have more time for leisure and a higher quality of life than at many previous points in history. But society never got "the promise of the 4-hour work week at mass scale," and he does not expect AI to change that.

He attributes it to rising expectations and relative competition — people focus on how they are doing relative to others. He also says society "is not working for a lot of people," and that having more productivity gains actually reach people would be a great thing. He predicts people will still complain about being busy, "but secretly we're going to be happy."

Altman said he "badly undershot on the compute investments." The interviewer framed the compute build-out as the biggest infrastructure project of maybe all time, or about to be.

Asked whether he could have known going in with the right mental model, his answer was yes — "but I got like psyched out by the financial markets or something." He calls it clearly a mistake.

On correcting course, he says he hopes to learn from it and to make new mistakes instead, but not that one again.

Asked whether he could have had a bigger impact on innovation through Y Combinator, Altman says it was not that intellectualized at the time.

He describes wanting to work on AI his whole life, and knowing that it would be the most important thing he could ever touch. For him it was a passion.

He says he did think it would be really important, but that was not the driver: "I just wanted to do it."

Altman says he feels "super lucky" to work with Jony Ive on designing beautiful things, and has learned a lot about how Ive studies a problem before trying to reach a solution — not even letting himself think much about the solution until he really understands the problem.

His stated takeaway, which he says he did not appreciate before: "really great design is way more about understanding the problem than the flash of insight." Rushing toward the thing or locking in too early, he warns, produces worse work.

He calls the iPhone "the greatest piece of technology humanity has collectively yet made."

The research reaches a level of detail fine enough, Altman says, that Ive writes it up as literal books.

When someone jokes about using the "Jony model" as training data, Altman replies no, but that it is a good thing to try.

He acknowledges there is a middle step — the inspiration that comes from understanding the problem really well and leads to a very novel idea — that seems to him to happen all at once, and he does not understand how.