中文版：山姆奥特曼传
I just finished reading The Sam Altman Biography, and since the iron is hot I want to write something down, and at the same time go back over what I actually learned from Altman’s story.
Honestly, I feel like our era is also one that produces heroes in abundance. Musk, Altman, Hassabis, and a whole crowd of geniuses in AI and other fields are all using their own strength to reshape the world. Sure, this era is a bit dim, on the downswing, full of controversy, with plenty of hard-to-accept sides. Lots of people even preach lying flat and giving up, saying the post-2000 generation is the most miserable one, and so on. But if you stretch your view out a bit, this era is in no way inferior to any other age in history that produced heroes.
But today’s heroes are no longer the generals who charged across battlefields and cut down enemy commanders left and right. They are the ones who change the world with technology and ideas.
1. Starting a Business Means Borrowing the Leverage of the Era
Sam Altman is a man with a lot of ambition.
He dropped out of college and started a company called Loopt, a location-based social product. If you only look at the features, it was nothing amazing; today’s Codex and Claude Code could probably reproduce it pretty easily. Early Loopt never became a great product, but it put Altman at Silicon Valley’s card table. In 2008 Altman showed off Loopt’s iPhone app at WWDC (Worldwide Developers Conference), Apple’s global developer conference, right when the App Store ecosystem was just getting started. Being able to appear on a stage like that shows Altman had already entered Silicon Valley’s core network.
This made me realize that people who change the world don’t necessarily produce some great product right out of the gate in their early years. They go through one project after another to buy themselves a ticket into a higher-quality network and meet more important people. And Loopt was exactly Sam Altman’s ticket in, just like beheading Hua Xiong over a cup of warm wine and three heroes fighting Lü Bu were the ticket in for the Liu brothers. I’ve been reading Romance of the Three Kingdoms lately too. Although historically Guan Yu didn’t really behead Hua Xiong, I only now realize that I used to read the whole Three Kingdoms purely as stories. In a sense, the Three Kingdoms is neither history nor fiction, but the ways of the world and the hearts of the people.
One of the key figures who had a big influence on Sam Altman was Paul Graham, who brought Sam Altman into YC. Many chapters of this book talk about YC, and Sam Altman later spent a while as the head of YC. This makes me feel strongly that if a person wants to achieve enormous success, they really can’t do without the sponsorship of a patron already up above, just as that saying goes: unite with every force that can be united.
Of course, the patron’s sponsorship is only an external factor, and the external is produced by the internal. As Wang Yangming said, the mind is principle. The “mind” here is perhaps the qualities a person already has. The patron’s sponsorship doesn’t mean you do nothing and wait around for someone to drop out of the sky; usually it’s because you’ve already shown some kind of potential. Altman wasn’t so strong because Graham picked him. He first had ambition, execution, and other qualities, which made the people up high willing to give him a hand. Actually, people already in high positions are quite happy to pull along a young man who reminds them of themselves in their own youth.
Altman didn’t want to do AI from the very start. He kept exploring fields that might have a breakthrough impact on humanity’s future, like nuclear fusion and new energy, and of course artificial intelligence too. Altman didn’t bet everything on AI at the beginning; he was looking, through exploration, for things that might change the future, like fusion and new energy. He didn’t charge straight into AI; he was looking for the most critical lever for changing the world in the future.
And that key lever for changing the world in the future is perhaps what you can call the leverage of the era.
This point matters a lot to me. If you want to do something, you can’t just stare at what features you can build right now. A genuinely big startup usually has to focus on one question: will a decade-scale change happen beneath this direction, or does it have the potential to become a piece of the future world’s infrastructure?
2. Where It All Started: The Birth of OpenAI
The founding of OpenAI was really pretty rocky at the start.
At the start it was a purely nonprofit research institution, and its mission was to make artificial intelligence benefit all of humanity, but then the problem appeared: to gather AI experts who were pulling fat salaries at big companies like Google and Facebook, one line about how we’re going to benefit humanity wasn’t enough. But only talking money wasn’t enough either, and besides, OpenAI at the time wasn’t even qualified to talk about money.
When OpenAI was first being set up, it made good use of a very plain truth: talk benefits to people who chase benefits, and talk ideals to people with ideals.
Relationships between real people can’t be purely about interests. This world has many people of different personalities. Some are as realistic as Cao Cao, but there are still idealists like Liu Bei. Lots of people think Liu Bei was a hypocrite, but maybe in this modern downswing too many people have become too realistic, so the only way they can explain noble things is to call them hypocrisy, which is what the ancients meant when they said you measure a gentleman’s heart with a petty man’s heart.
Of course, idealism doesn’t mean you don’t talk about interests, and Liu Bei wasn’t a man who could only cry. A truly formidable organizer can always adapt to the situation. Take Liu Bei’s reaction when Cao Cao discussed heroes over wine. Just a short while before, at Uncle Dong Guo’s request, he had actually gone and signed his name to the emperor’s plea for help, and then Cao Cao suddenly called him in. An ordinary person would probably have been scared limp, sure they’d been discovered, but Liu Bei could handle Cao Cao with complete ease.
The founding of early OpenAI really couldn’t do without one person: Greg Brockman, Stripe number four (the fourth employee at Stripe), who within a few short years became CTO. And Stripe was incubated in the YC startup program, and it was there that Sam Altman met him, then went out and contacted him when founding OpenAI.
At first Greg Brockman went to Yoshua Bengio, one of the three godfathers of deep learning, and from him got a list of the world’s top AI researchers. Ilya was essentially the spiritual leader of that group, so Greg Brockman went to Ilya first, along with several dozen other researchers.
But Google, Facebook, and the other big companies all rolled out sky-high salaries to hold them back, so Greg Brockman made a very smart decision. Negotiating one by one would definitely be a hassle and wouldn’t work, so saying less and doing more was a great choice. So he invited them all to spend two days together at a Napa winery, a visit where money wasn’t discussed, only ideals, and he gave them a final deadline for an answer. In the end they all resigned one after another and joined. A person really does need a sense of mission to feel that their life is meaningful. After all, the biggest difference between people and animals is that people have feelings.
Let me say what I think first: to begin, people dealing with each other can definitely not be purely about interests; this world has many different personalities,
Another point that struck me hard while reading is that OpenAI lagged far, far behind DeepMind for several years, until GPT-2 and GPT-3 gradually proved the large language model route, and then in 2022 ChatGPT wrapped that capability into an extremely low-threshold chat product. What is truly powerful about it isn’t just that the model got bigger; it turned the model’s ability into a product ordinary people can use the moment they open it. There had been plenty of companies making chatbots before, and they basically all failed. I think the reasons might be the following few.
- In 2022 the pandemic had ended not too long ago, and people needed a machine companion to ease some boredom or bring a little warmth
- GPT-3.5 has quite a few more parameters than the earlier GPT-2 and is much smarter too. A change in quantity produces a change in quality. The first large model I ever used was GPT-3.5, and thinking back to the context of that time, it really was quite good.
- Chatbots in the past were just a chat interface plus limited capabilities, with very low intelligence; they couldn’t answer even slightly more complicated questions, but GPT-3.5 successfully crossed that threshold.
And one thing I noticed while reading is that lots of Silicon Valley big shots have their own blogs: Sam Altman, Paul Graham, and the lately controversial Bill Gates all have blogs. Altman’s blog isn’t really updated anymore, but his older content is still well worth a look. Recording more and writing more really can change a lot of things; otherwise lots of stuff just stays stuck in your head with no way to come out.
The Conflict Between Musk and Altman
Lately Musk has been suing Altman nonstop, and the result for now is that OpenAI’s crisis has been temporarily lifted. The jury thought Musk’s suit came too late, so OpenAI was temporarily relieved of one major legal pressure. But what this really exposes isn’t just a legal problem; it’s the inevitable contradiction that appeared in OpenAI’s transformation from a nonprofit organization to a for-profit organization. Actually the conflict between Musk and Altman had foreshadowing long before. As for OpenAI, if it had stayed a purely charitable organization facing ever-growing expenses, it wouldn’t have been able to survive very well either, and for a while there was even a serious state of talent loss.
At the very beginning, Musk said he’d put a billion dollars into OpenAI as funding, but in reality only a few tens of millions arrived each year, which meant OpenAI could only spend very frugally at the start, not even reaching a fraction of DeepMind’s yearly spending.
Musk even came and poached people from OpenAI, taking Karpathy away, which was really too much. In the end things only turned around after Altman gave up his position at the head of YC and threw himself into OpenAI full time.
OpenAI Partners With Microsoft
At the very beginning Bill Gates actually wasn’t all that optimistic about OpenAI. Microsoft’s CEO at the time was of Indian descent, and Bill Gates had more or less retreated behind the scenes.
In his 2023 blog post The Age of AI Has Begun, Gates wrote that he’d been following OpenAI’s progress since 2016. In 2022 he set the OpenAI team a challenge: train a model that could pass the Advanced Placement Biology (AP Biology) exam. He himself expected it would take two to three years, but it was finished in just a few months. Gates said the last time he had been that blown away was back in 1980 when he first saw the graphical user interface, and that later gave birth to Microsoft. He believes the arrival of artificial intelligence is an event on the same level as the microprocessor, the personal computer, the internet, and the mobile phone.
Microsoft’s huge investment in OpenAI and deep cooperation with it basically started from that point. Looking back now, Gates’ judgment wasn’t wrong, and Altman really is the one who turned technology into an era-making product.
Final Thoughts
My biggest feeling while writing this piece is that what stands out most about Altman isn’t his technical ability, but his ability to organize and gather people. From YC to OpenAI, what he has been doing all along is putting the right people in the right positions. Technology eventually comes down to a product, and behind the product there are always people. This may also be why it’s him, and not someone who understands models better, standing at the center of this era.