Public transcript

Sam Altman Reveals What He'd "Never Ask" ChatGPT, Says "Govts Shouldn't Control.." | OpenAI | VERTEX

By: VERTEX
Length: 57 min
Language: English
Text from: the video's captions
Transcribed: 12 October 2026

Read the full transcript Open on YouTube

Notes in English.

Summary

In this wide-ranging interview at Express Adda, Sam Altman discusses how AI has advanced dramatically in a year—from high school math to research-level mathematics—and how India has shifted from an AI consumer to a builder economy, with India being OpenAI's fastest growing market for Codex. He addresses jobs disruption, the need for massive compute buildout, government-industry partnership, and China's position in AI. He also shares personal views on topics from regulation to his biggest mistakes, and answers audience questions on responsible AI, vulnerable professions, and creativity.

Key points

Altman says AI has gone from doing high school math a year ago to solving 7 of 10 research-level math problems in the recent 'first proof' challenge.
India is OpenAI's fastest growing market for Codex, and Altman says the builder energy in India is 'off the charts'.
He clarifies his $10 million comment: India couldn't build a frontier model for that amount, but Indian narrow models are incredible.
He is 'not a jobs doomer'—every technological revolution has created new jobs, though big categories of jobs will be completely obsolete.
He argues water usage criticism of AI data centers is 'totally fake' but total energy consumption is a real issue requiring nuclear, wind and solar.
He believes no single person, company, or country—including the US—should control superintelligence, and power should be widely distributed.
He says AGI feels 'pretty close' and, given a faster expected takeoff, superintelligence is 'not that far off'.
He would never ask ChatGPT how to be happy, preferring to ask a wise person, though he thinks AI can be good for therapy and life advice.
He calls his decision not to take equity in OpenAI 'one of the dumbest' things, given the conspiracy theories it spawned.
The biggest corporate mistake he sees is planning multi-year AI deployment timelines like a 2026-2028 strategy, which he calls catastrophic.

Questions it answers

00:00Opening and AI's progress

How much has AI advanced since Altman's last India visit a year ago?

AI has gone from doing good high school math to solving research-level mathematics—OpenAI's latest model got 7 of 10 unsolved 'first proof' problems right—and programming has been completely transformed by tools like Codex. Altman credits the scientists behind deep learning rather than himself.

AI now pushes the edge of human knowledge in math and physics.
India is the fastest growing market for Codex.
India has shifted from AI consumer to builder economy.

05:35The $10 million controversy

Did Altman say India couldn't build a meaningful LLM with $10 million?

He says the comment was taken out of context: the point was that no one could build a frontier model for $10 million, which is even more true now. India's smaller narrow models are incredible, and with enough funding an Indian company could build a frontier model.

Frontier models have become very expensive.
Indian narrow models are praised as incredible.

06:35Jobs and disruption

How big a threat is AI to jobs, especially India's IT sector?

Altman says there will be a big impact and it's unhelpful to pretend otherwise, but people will operate at higher levels of abstraction and new jobs will emerge as with every technological revolution. Durable skills include fluency with AI tools, resilience, adaptability and being useful to others.

He is 'not a jobs doomer'.
The promised leisure from past revolutions never happened.
Predictions of new jobs during the industrial revolution were shockingly wrong.

11:03Compute and infrastructure

Does the world have the computing power for an AI-first India?

Not yet, and the world must rapidly work together to build it—Altman calls it the most expensive, complex infrastructure project ever. He dismisses orbital data centers as ridiculous this decade, and says water-usage criticism is fake while total energy use is a real issue.

No one asked ever wants less than one GPU working for them.
8 trillion GPUs for 8 billion people is impossible anytime soon.
Nuclear, wind and solar are needed quickly.

15:22Government and geopolitics

How should tech companies, governments and China fit into the AI picture?

Altman says government partnership is essential given the infrastructure scale and societal impact. China is ahead in manufacturing, robots and energy buildout but behind elsewhere, and the dynamic is both collaborative and competitive. He stresses that democracy should lead with AI and that no single entity should control superintelligence.

AI will be one of the highest-order bits of geopolitical tension.
No one country, company or person should control superintelligence.
He worries about internet-fought wars and surveillance states justified by AI fear.

24:16Rapid-fire game

What are Altman's quick takes on Google, regulation, and AI criticisms?

He admires Google's early conviction and recent execution; says no country yet has regulation right but experimentation is good; defends against power-concentration and water criticisms; and admits AI can make some kids dumber if they cheat, though most kids report accomplishing more.

Water usage claims like '17 gallons per query' are 'completely untrue'.
AI has probably caught up with humans on energy efficiency per query.
Education must evolve as it did after Google.

36:13News, safety and relationships

What does Altman think about war uses of AI, safety resignations, and AI in love?

He says AI shouldn't make warfighting decisions yet due to reliability limits, agrees safety people are overwhelmed because the takeoff is faster than expected, and bets human relationships will become more valued in an abundant world.

AI analyzing intelligence reports is a great use; warfighting decisions are not.
He expects a faster takeoff than originally thought.
Human warmth will become one of the most valuable commodities.

40:36AGI, devices and reflections

How close is AGI, and what is Altman building with Jony Ive?

AGI feels 'pretty close' and superintelligence is 'not that far off'. The Jony Ive device aims to move beyond the 50-year-old computer form factor toward technology that observes your whole life and participates in it rather than getting in the way; they hope to talk about it late this year.

He would never ask ChatGPT how to be happy.
Not taking OpenAI equity was 'one of the dumbest' decisions.
Multi-year corporate AI deployment plans are a catastrophic mistake.

Notes

AI's rapid progress

A year ago AI could do high school math; by last summer it competed in the hardest math competitions, and OpenAI's latest model solved 7 of 10 unsolved research problems in the 'first proof' challenge.
Programming has been completely transformed: typing an idea into Codex can produce an entire application. Altman calls India OpenAI's fastest growing market for Codex.
He credits a small set of researchers who discovered deep learning, insisting the story is about scientists, not him.

India and the $10 million comment

Altman clarifies his earlier remark was about building a *frontier* model for $10 million, which he says is even more impossible now—but India's smaller narrow models are 'incredible'.
He says India has shifted from an AI consumer to having builder energy that is 'off the charts', and that India should play at all levels of the AI stack, a view he says the Prime Minister shares.

Jobs

He is 'not a jobs doomer': every technological revolution panicked about jobs and created new ones; the promised leisure never arrived.
Still, big categories of jobs will be completely obsolete—his own training in writing C++ by hand is 'effectively completely irrelevant'.
Durable skills: fluency with AI tools, resilience, adaptability, and learning how to be useful to people.

Compute and infrastructure

The world does not yet have the compute needed; he frames it with the question 'how many GPUs would you like working for you?'—no one says less than one.
He calls orbital data centers 'ridiculous' for now given launch costs and GPU breakage.
Water-usage criticism is 'totally fake' since evaporative cooling ended, but total energy use is real and demands nuclear, wind and solar.

Power and geopolitics

No single person, company, or country—including the US—should control superintelligence; he favors widely distributed power and democratized access.
China is ahead in manufacturing, robots, motors and energy buildout, behind in other areas; the relationship is both collaborative and competitive.
He worries about a new kind of internet-fought war and about fear of AI being used to justify a surveillance state.

Personal reflections

Not taking equity in OpenAI was 'one of the dumbest' decisions due to the conspiracy theories it spawned.
He would never ask ChatGPT how to be happy—he'd ask a wise person—though AI can be good for therapy and life advice.
AGI feels 'pretty close', and given a faster expected takeoff, superintelligence is 'not that far off'.
The Jony Ive device effort aims to move beyond the 50-year-old computer form factor toward technology that participates in your life; they hope to talk about it late this year.

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