Nov 8, 2024 · 46m · y-combinator
How To Build The Future: Sam Altman · Y Combinator
0:00 / 0:00
▶ Watch on YouTube →
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
OpenAI CEO Sam Altman joins Y Combinator CEO Garry Tan on 'How To Build The Future' to discuss the exponential trajectory of artificial intelligence, the historical advantages of agile startups during paradigm shifts, and OpenAI's journey from early research lab to leading the path toward Superintelligence.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. The partners hold 29.9% of the talking time here. How this is scored →
The partners as informed peer 4.8 Guest teaching 2.8 Guest disagreement 1.3 The partners pushing back 1.3
speaking balance: gold is the partners, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 7:07 Sam rejects the formidable founder mythology
Sam gently dismisses the popular Paul Graham lore about his formidable YC admission, stating he would prefer the story drifted off and that he does not identify as formidable.
Hardest push from the partners ▶ 5:18 Garry presses on potential barriers to abundant energyGarry challenges the optimistic assumption of unlimited energy by questioning what happens if physical laws or roadblocks prevent scaling.
Biggest teaching moment ▶ 25:08 Sam explains the pre-GPT sentiment neuron breakthroughSam corrects the technical timeline by pointing out Alec Radford's unsupervised sentiment neuron breakthrough occurred before GPT-1 and laid the foundation for the entire series.
The partners hold their own ▶ 40:42 Garry synthesizes CAD CAM AI tool chaining and code generationGarry demonstrates deep domain mastery by articulating how language models interface with domain software to generate code and recursive tools.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | The partners as informed peer | Guest teaching | Guest disagreement | The partners pushing back | Why |
|---|---|---|---|---|---|---|
| Title Sequence: How to Build the Future | 4 | 3 | 1 | 1 | Garry quotes directly from Sam's essay 'The Age of Intelligence' and frames the conversation around ASI timelines and techno-optimism. Sam elaborates collaboratively on compounding progress and the O1 architecture without friction. | |
| Achieving Abundant Intelligence and Abundant Energy | 4 | 2 | 1 | 3 | Garry poses a counterfactual question asking what happens if physical laws prevent unlimited abundant energy. Sam responds with practical trajectories like solar plus storage before asserting eventual mastery of physics. | |
| Lessons from Early Y Combinator Days | 5 | 3 | 3 | 1 | Garry brings up the famous Paul Graham story about Sam's admission to YC, which Sam gently downplays and rejects as partly apocryphal. Both hosts bond warmly over the power of peer groups and Stanford experiences. | |
| The Founding Story of OpenAI and YC Research | 5 | 3 | 1 | 1 | Garry recalls early YC partner conversations and historical landmarks like ImageNet and Bostrom's book. Sam recounts the early days recruiting Ilya Sutskever and establishing the initial AGI charter goals. | |
| The Heresy of Scaling Laws and Concentrated Bets | 5 | 2 | 2 | 1 | Garry notes how critics warned scaling bets would create an AI winter. Sam agrees and discusses how doubling down on deep learning scaling was considered heretical by established leaders. | |
| Truth-Seeking, Scientific Iteration, and Rebounding from Failure | 4 | 3 | 1 | 1 | Garry summarizes the iterative scientific approach required when conviction meets real data. Sam explains OpenAI's willingness to be wrong and pivot quickly from robotics and games to LLMs. | |
| Breakthrough Models: From Sentiment Neurons to GPT-4 | 6 | 5 | 1 | 1 | Garry references Alec Radford's work and Jake Heller's commercialization of GPT-4 at Casetext. Sam clarifies the timeline around the unsupervised sentiment neuron paper before GPT-1 and recounts user validation. | |
| Lessons from Looped and Platform Transitions | 4 | 2 | 1 | 1 | Garry brings out a vintage Mac replica and asks about Sam founding Loopt before modern app stores existed. Sam reflects on his 20s as an apprenticeship in platform shifts. | |
| Startup Speed Advantage in Paradigm Shifts | 5 | 2 | 1 | 1 | Garry highlights how incumbent giants like Facebook almost missed mobile platform shifts. Sam enthusiastically agrees that incumbent slowness gives modern AI startups their main advantage. | |
| OpenAI's Rapid Scaling, Executive Evolution, and Mission Clarity | 4 | 3 | 1 | 2 | Garry delicately asks about leadership drama and executive departures over the past year. Sam candidly acknowledges the intense transition from research lab to scaled company and the clarity they now have. | |
| The Five Levels of AI and One-Person Unicorns | 7 | 4 | 1 | 1 | Garry provides an in-depth breakdown of Camfer's CAD/CAM hackathon project, tool use, and code generation. Sam maps this progress onto the 5 levels of AI from reasoners to autonomous innovators. | |
| Building Enduring Value and Looking Ahead to 2025 | 5 | 2 | 1 | 1 | Garry asks for actionable founder advice and Sam warns that demos are easy while building enduring business value still requires foundational business laws. The episode concludes warmly on personal milestones. |
Statements from this episode (22)
Sam Altman: The AI Era Is the Best Time Yet to Start a Tech Company
“I sort of think with each successive major technological revolution, you've been able to do more than you could before. And I would expect the companies to be more amazing and impactful and everything else. So yeah, I think it's the best time yet.”
Sam Altman: Startups Gain an Edge Over Big Companies During Tech Revolutions
“Big companies have the edge when things are like moving slowly and not that dynamic. And then when something like this or mobile or the internet or semiconductor revolution happens, or probably like back in the days of the industrial revolution, that was when …”
Altman: AI progress could compound over next nine years toward superintelligence
“I can see a path where the work we are doing just keeps compounding, and the rate of progress we've made Over the last three years, continuous for the next three, or six, or nine, or whatever. You know, nine years would be, like, 3500 days, or whatever. If we …”
Altman: Expecting significant compounding AI progress ahead despite potential walls
“We could hit some, like, unexpected wall, or we could be missing something, but it looks to us like there's a lot of compounding in front of us still to happen.”
Altman: Climate fixes, space colonies, and physics discoveries are not far off
“But I do think all of those things, and probably a lot more we can't even imagine, are maybe not that far away.”
Altman: Solar plus storage trajectory is sufficient without nuclear breakthroughs
“Solar plus storage is on a good enough trajectory that even if we don't get a big nuclear breakthrough, we would be like, okay-ish.”
Altman: Ilya Sutskever is a genuine genius and visionary
“He's also, he's extremely smart, like, true, true, genuine genius and visionary, but also he has this incredible presence”
Altman: OpenAI openly targeted AGI when the field considered it taboo
“We said from the very beginning we were going to go after AGI at a time when in the field you weren't allowed to say that because that just seemed impossibly crazy and, you know, borderline irresponsible to talk about.”
Altman: OpenAI's original goals included never exceeding 120 people
“I believe there were three goals for the For the effort at the time. It was like, figure out how to do unsupervised learning, solve RL, and never get more than a 120 people. Missed on the third one.”
Altman: Most of the world undervalues extreme conviction on one bet
“Most of the world still does not understand the value of, like, a fairly extreme level of conviction on one bet.”
Altman: Conviction is essential until empirical data emerges
“Conviction is great until the moment you have data one way or the other. And there are a lot of people who hold on it past the moment of data.”
Altman: OpenAI initially focused on robotics and gaming agents over LLMs
“I had no idea that language models were going to be the thing. You know, we started working on robots and agents playing video games and all these other things.”
Altman: OpenAI's GPT Series Originated From Radford's Unsupervised Sentiment Neuron Discovery
“And at the time, unsupervised learning was just not really working. So he noticed this one really interesting property, which is there was a neuron that was flipping positive or negative with sentiment. And yeah, that led to the GPT series.”
Altman: No Great Businesses Were Built on GPT-3 Except Copywriting
“But, with the possible exception of copywriting, no great businesses were built on GPT-III.”
Altman: The world is astonishingly asleep on the AI platform shift
“I think that's why I'm so excited for startups right now. It is because the world is still sleeping on all of this to such an astonishing degree. And then you have, like, the YC founders being like, no, no, I'm gonna, like, do this amazing thing and do it very…”
Altman: I wrote OpenAI's early checks before Elon Musk funded scale
“It was definitely helpful that I could just, like, write the early checks for OpenAI, and I think it would have been hard to get somebody else to do that at the very beginning. And then Elon did it a lot at a much higher scale, which I'm very grateful for, and…”
Altman: Stanford AI lab told students not to work on neural nets
“I went to school to study AI. But at the time I was working in the AI lab, the one thing that they told you is definitely don't work on neural networks. We tried that. It doesn't work.”
Altman: OpenAI now knows the path to building AGI
“This is the first time ever where I felt like we actually know what to do. Like, I think from here to Building an AGI will still take a huge amount of work. There are some known unknowns, but I think we basically know what to go do, and it'll take a while.”
Altman: Level 4 AI innovation can happen using current models creatively
“I had been telling people for a while, I thought that The level two to level three jump was going to happen, but then the level three to level four jump was, level two to level three was going to happen quickly, and then the level three to level four jump was …”
Altman: AI will enable someone to solve all of physics
“The fact that somebody else, I really believe, is now gonna go solve all the physics with this stuff, like, I'm so excited to be alive for that.”
Altman: AI will reach Level 3 autonomous agents faster than expected
“Three is agents ability to go off and do these longer-term tasks You know, maybe like multiple interactions with an environment, asking people for help when they need it, working together, all of that. And I think we're gonna get there faster than people expec…”
Altman: AI startups still need traditional business moats to endure
“It is easy when there's a new technology platform to say, well, because I'm doing some of the AI, the rule, the laws of business don't apply to me. I have this magic technology, and so I don't have to build a moat or a you know, competitive edge or a better pr…”