Greg Brockman, co-founder and President of OpenAI, discusses the dramatic 72-hour firing and reinstatement saga that nearly destroyed the company in November 2023. As someone who abandoned his preferred coding tools for the first time to use OpenAI's own Codex, Brockman represents the technical leadership driving artificial general intelligence development.

The conversation covers OpenAI's evolution from nonprofit research lab to $100+ billion valued company, including failed negotiations with Elon Musk over control, the crucial Microsoft partnership providing $13 billion in funding, and the technical breakthroughs from GPT models to Sora video generation. Brockman explains how scaling laws continue driving exponential progress despite efficiency gains, creating an insatiable demand for compute resources.

From his rural North Dakota upbringing to transferring from Harvard to MIT, Brockman's path illustrates the unconventional thinking that shaped OpenAI's approach to building AGI systems that could benefit humanity while maintaining diversity and resilience rather than centralized control.

The 72-Hour OpenAI Coup That Nearly Destroyed Everything

Greg received a surprise video call from board members excluding Sam Altman, learning simultaneously that Sam was fired, Mira became interim CEO, and he was removed from the board but expected to stay at the company.

"I knew in that moment it wasn't right" - Greg immediately told his wife they had to leave, assuming all equity would go to zero, and called Sam to start a new company that same day.

95% of OpenAI's 770 employees signed a petition demanding the board's reversal, with so many people editing the Google Doc simultaneously that it crashed Google's servers.

Microsoft offered to hire all OpenAI employees within 24 hours, creating a 'full-size lifeboat' that expanded from a small startup plan to accommodate everyone.

The resolution came when Ilya Sutskever posted on Twitter expressing deep regret, creating an emotional moment of forgiveness that Greg describes as pivotal to rebuilding the company.

Why Elon Musk's $10 Billion OpenAI Deal Collapsed Over Control

Musk agreed OpenAI needed to transition from nonprofit to for-profit structure, saying 'the time is now to go' after their first major breakthrough, but negotiations stalled over control terms.

"He needed majority equity, he needed absolute initial control, and he needed to be CEO" - Greg and Ilya couldn't accept giving one person absolute control over potential AGI.

Alternative proposals included merging into Tesla to 'cash out the cash cow' in secret, and launching a $10 billion ICO that Musk initially embraced then abandoned.

Musk sent a devastating December 2018 email stating "OpenAI has a 0% chance of success relative to Google without a dramatic change in resources and execution. Not a 1%."

Despite the failed partnership, Musk remained supportive through 2019, even advising the team and staying involved as they developed their for-profit structure.

Microsoft's $13 Billion Bet on Supercomputer Dreams

Microsoft provided three funding rounds totaling $13 billion ($1B + $2B + $10B) when no other investors were offering such quantum capital for AI research.

"It wasn't really about the cash, it was really about the supercomputers" - the partnership enabled designing massive GPU clusters that seemed impossible in 2017.

Microsoft's thesis was that even if the models failed, OpenAI could help them build better supercomputers for deep learning, creating aligned incentives for both companies.

The compute supply chain is now sold out 18+ months ahead across TSMC wafers, memory, and hard drives as everyone races to build AI infrastructure.

Scaling Laws and the Jevons Paradox of AI Efficiency

"Scaling laws continue unabated" despite public claims they're dead - the deterministic relationship between compute, model size, and intelligence remains predictable.

Jevons paradox applies to AI: every efficiency gain (10x-100x cheaper models) gets overwhelmed by 1000x increased demand for more intelligent applications.

"Compute becoming a basic human right" is inevitable as individual productivity becomes directly tied to GPU access in the emerging economy.

Internal OpenAI teams battle intensely over compute resources, with some requesting 100-1,000 GPUs each for optimized code generation tasks.

From ChatGPT Surprise to Reasoning Revolution

ChatGPT's success surprised even OpenAI leadership who had grown accustomed to GPT-4 and forgotten that GPT-3.5 was revolutionary to outside users.

"Working at OpenAI: you live in the future" - the team experiences AI capabilities months or years before public release, creating perspective distortion.

Reasoning AI represents a fundamental shift from instant responses to systems that "pause, think, call tools, run experiments" before answering complex problems.

Future AI assistants will operate like human coworkers, saying "I gave you an answer five minutes ago, but I realized that was wrong - here's the correct thing."

Building Resilience Instead of Centralized AI Control

OpenAI rejects the vision of one centralized, aligned AI system, arguing "there is no humanity's values" - only someone's values imposed on everyone else.

"Humanity and really life itself survives through diversity, through not being a monoculture, through there being the possibility of heresy."

The resilience model mirrors how society adapted to steam engines and cars - building safety standards, infrastructure, and distributed defenses rather than central control.

AI tools will enable a "Cambrian explosion" of new companies as small teams can run highly effective, revenue-generating businesses with minimal staff.