Public transcript

FULL INTERVIEW: Sam Altman on OpenAI, New AI Model, Governments and Global Growth at G20 | AI1G

By: DRM News
Length: 32 min
Language: English
Text from: the video's captions
Transcribed: 13 October 2026

Read the full transcript Open on YouTube

Notes in English.

Summary

In this G20-side interview, Sam Altman traces his path from AI enthusiast to OpenAI's founding at the end of 2015, and describes how deep learning, scaling laws, and the GPT-4 moment convinced the team the technology would go very far. He argues that adopting AI is non-negotiable for countries, comparing it to electricity, and predicts a massive boom in entrepreneurship and small business creation. He also discusses risks like cybersecurity, biosecurity, and power concentration, and explains why AI must be made abundant and inexpensive so its benefits are broadly shared.

Key points

Sam Altman says OpenAI was announced at the very end of 2015, with the first day of work in January 2016.
Altman dates the moment AI clearly captured knowledge to the GPT-4 period, around March 2023.
He predicts AI will drive the greatest boom in entrepreneurship and small business creation the world has ever seen.
Altman says adopting AI is non-negotiable for countries, comparing it to electricity.
He warns some things will go very wrong with cybersecurity unless people act quite urgently.
Altman says the work a startup did in 3 months is now probably doable in about 17 minutes with Codex.
He argues AI must be made abundant and inexpensive to avoid it becoming a commodity competed over by the rich.
Altman describes OpenAI as pragmatic, a reliable partner, and firmly on team humanity.

Questions it answers

00:03Altman's path to OpenAI

How did Sam Altman end up leading OpenAI?

Altman was a lifelong AI and sci-fi nerd who studied AI in school, gave up when nothing worked, and pursued entrepreneurship. After deep learning started working in 2012 and improving with more compute, he concluded by 2015 that AI was going to happen and started OpenAI at the end of 2015.

Altman calls the 2012 success of deep learning a watershed moment.
OpenAI was announced at the end of 2015, with work starting January 2016.

02:09Early OpenAI and scaling laws

What did OpenAI work on before language models took off?

Early work included training a robot hand to solve a Rubik's Cube and playing video games. Progress on unsupervised learning led to GPT-1 around 2018 and GPT-3 in 2020, and the discovery of scaling laws convinced the team to push for massive compute despite skepticism.

GPT-1 came around 2018; GPT-3 was trained in 2020.
Scaling laws suggested a trillion dollars of compute would produce something amazing.

03:57The GPT-4 turning point

When did it become clear AI could capture real knowledge?

Altman says the consensus moment was around GPT-4, roughly March 2023, which the team had used internally for about 8 months. Even internal skeptics concluded the approach would go very far, and the technology is now discovering new knowledge and creating complex software.

GPT-4 launched around March 2023 after about 8 months of internal use.
OpenAI is launching a new model soon, with more steps to follow.

05:51AI and the G20 economies

How should G20 economies think about AI?

Altman predicts the greatest boom in entrepreneurship and small business creation the world has ever seen, citing a laundromat owner who used ChatGPT to run his business in 2023. He says using AI is non-negotiable for countries, though they may choose different regulatory approaches and whether to build or rent data centers.

Altman tells the story of a laundromat owner running his business with ChatGPT in 2023.
Countries can build or rent data centers; both are fine.

09:41Context and OpenAI's role

Will OpenAI provide the context for how AI is used?

Altman says that will not and should not come from OpenAI; the company provides the engine, while people around the world bring the context, questions, and ideas. He also describes OpenAI as pragmatic, a reliable partner that respects sovereignty, and firmly on team humanity.

OpenAI sees itself as providing an engine, not applications.
Altman says OpenAI is on team humanity and wants the whole global economy to succeed.

17:01Entrepreneurship as the fire starter

How does entrepreneurship fit with OpenAI's vision?

Altman, who ran a startup accelerator before OpenAI, calls entrepreneurship maybe the most important novel force in capitalism. He says the work a startup did in 3 months is now probably doable in about 17 minutes with Codex, transforming how quickly people can test ideas and build things.

Altman previously ran a startup accelerator.
Three months of startup work is now doable in about 17 minutes with Codex.

18:44Non-negotiable adoption and the electricity analogy

What does Altman mean by AI adoption being non-negotiable?

Altman compares rejecting AI to a country rejecting electricity over 100 years ago. He predicts that in a decade we will talk about AI less because, like electricity, it will simply be expected in every product and service, and a kid growing up today will never know a world without smart products.

Altman compares rejecting AI to rejecting electricity 100+ years ago.
He predicts we will talk about AI less in a decade because it will be expected everywhere.

21:34Risks over the next five years

What could hold us back in five years?

Altman points to cybersecurity, where he says some things will go very wrong unless people act urgently, plus biosecurity and other challenges ahead. He also warns that too much concentration of power would be bad, and that failing to make AI abundant would make it a highly priced commodity competed over by the rich.

Altman warns some things will go very wrong with cybersecurity without urgent action.
Failing to make AI abundant could worsen inequality.

Notes

Sam Altman's path to OpenAI

Altman was an AI and sci-fi nerd as a kid and studied AI in school, but gave up when nothing was working and turned to entrepreneurship.
He calls 2012 a watershed moment when deep learning started working and improved with more compute.
OpenAI was announced at the very end of 2015; the first day of work was January 2016.

Early research and scaling laws

Early projects included training a robot hand to do a Rubik's Cube and playing video games.
Progress on unsupervised learning led to GPT-1 around 2018; GPT-3 was trained in 2020.
Scaling laws suggested that applying something like a trillion dollars of compute would produce something amazing, and the team decided to try.

The GPT-4 moment

Altman dates the consensus turning point to GPT-4, around March 2023, after roughly 8 months of internal use.
Even internal skeptics concluded the approach would go very far.
OpenAI is launching a new model soon, with many more steps to follow.

AI and the G20 economies

Altman predicts the greatest boom in entrepreneurship and small business creation the world has ever seen.
He cites a laundromat owner who used ChatGPT in 2023 for marketing, legal contracts, and vendor relationships.
2025 brought an acceleration with coding agents and white-collar work agents; long-term persistent agents are coming next.
He calls AI adoption non-negotiable for countries, comparing rejecting it to rejecting electricity 100+ years ago.
Countries can choose to build data centers or rent them; both are fine.

OpenAI's role and differentiation

OpenAI provides the engine; the people of the world must bring context, questions, and ideas.
Altman describes OpenAI as pragmatic, avoiding blind optimism and doomerism, a reliable partner that respects sovereignty.
The company wants the whole global economy to succeed, not to be the only company, and is on team humanity: people should run the future, not AI.

Entrepreneurship

Before OpenAI, Altman ran a startup accelerator and calls entrepreneurship maybe the most important novel force in capitalism.
Work a startup did in 3 months is now probably doable in about 17 minutes with Codex.

Risks and abundance

Top concerns: cybersecurity (some things will go very wrong without urgent action), biosecurity, and concentration of power.
Token use grew multi-million-fold in 6.5 years; average per capita use went from zero to 100,000 tokens a month.
Altman wants AI abundant and inexpensive, like the cost of an hour of light at night after electricity and LEDs.
He is optimistic: worry is warranted, as with fire safety, but AI has gone better than doomers predicted 5 years ago.

Text from the video's own captions. Summary and notes written by AI from the transcript, so check anything important against the recording.