Stanford Graduate School of BusinessSam Altman, Co-Founder and CEO of OpenAI: Staying Human While Scaling Superintelligence
Stanford Graduate School of Business
Early learning 0:04
Altman says he almost chose investment banking at Stanford because it looked like the cool path, but he felt growing dread and instead took Y Combinator’s offer to spend the summer building a startup. He says starting a company taught him faster than school ever could. Running YC was different, but it still forced him to learn by doing, with mentors, experiments, and mistakes.
Why OpenAI shifted 7:01
He says OpenAI began as a research lab because deep learning was already working and scaling better than most people believed. By 2020 or 2021, it was clear the lab would need to become a product company, and ChatGPT made that obvious. His hardest choices now center on pacing AI, keeping people in control, and avoiding both rushed deregulation and a total stop.
Safety and pacing 16:30
If progress keeps moving at the current pace, he thinks the world will not be able to keep operating as it is. The systems are becoming more capable too quickly, so the question is no longer whether the models will improve, but what changes are needed to keep safety and control in place.
Work and human value 21:36
He does not expect a sudden job apocalypse. Many tasks will be done better, cheaper, and faster by AI, and some jobs will change completely. But he thinks people will still care about human work, human values, and the need to learn skills that help you think, explain, and make useful things for other people.
What humans still do 26:01
He uses cursive and writing as examples. Some things should be learned because they build thinking, not because machines cannot do them. He thinks the future should keep space for creativity, discovery, and a good life, even if AI can handle much of the knowledge work and the pressure toward efficiency stays strong.
Compute and alignment 33:30
He says OpenAI is pushing three linked areas: more data centers, its own chips, and the supply chain around fabs, power, and equipment. The chip called Jalapeno is meant to raise intelligence per watt, and smarter models may help design better chips. On alignment, he says it turned out easier than he expected so far, but the real risk has grown, and he worries people are treating that ease as a reason to run ahead.
Openness and limits 38:01
He argues against banning all open source models, even though they can spread hate speech, scams, hacking, and fraud. He says the real question is where to draw the line, since society already accepts some risk for liberty. He also says open-weight models matter, but current best models should not be opened soon. More resilience in cyber defense and bio defense would have to come first.
AI-generated summary. It can be wrong or incomplete - check anything that matters against the original.
