Nov 8, 2024, 10:01 AM ETInterview, Fireside Chat

How To Build The Future: Sam Altman

Sam Altman characterizes the current era as the "best timing yet" to launch technology companies, driven by a new technological revolution comparable to the internet or mobile shifts.
Altman projects that Artificial Superintelligence (ASI) could be reached in roughly 3,500 days (nearly 10 years) if current compounding rates of progress in deep learning continue.
He identifies "abundant energy" and "truly abundant intelligence" as the two primary inputs required to unlock a future of material abundance, solving climate change, enabling space colonies, and discovering physics.
Altman estimates that even without a nuclear fusion breakthrough, solar-plus-storage trajectories are sufficient to reach a level of energy abundance, though a fusion or Dyson sphere-scale breakthrough would be ideal.
He asserts that the current rate of AI capability improvement has only just begun and that O1 models are far from the limit of progress, citing a recent architecture shift as a major unlocker.

At its inception, OpenAI adopted a strategy of extreme conviction on a single bet—scaling deep learning—despite facing derision from established "eminent leaders" who viewed the approach as irresponsible or destined to cause an AI winter.
The founding team, assembled in January 2016, explicitly stated their goal was AGI at a time when such a declaration was considered impossible; this "impossibly crazy" mission was a primary recruitment tool for young, unconventional talent.
Early OpenAI research goals included unsupervised learning, solving Reinforcement Learning (RL), and limiting the company to 120 people (the latter of which was missed).
The organization evolved from a broad research lab into a focused entity by recognizing that while they were less resourced than competitors like DeepMind, they could win by concentrating resources on scaling rather than spreading bets across many experiments.
Altman admits the team was frequently wrong about the specific path to AGI, initially exploring robotics and video games before pivoting to language models after the breakthrough of GPT-3.
The commercial viability of the technology shifted from GPT-3 (limited use cases) to GPT-3.5 (foundational for new startups) to GPT-4 (reaching a threshold of reliability where it could perform complex, reliable workflows).

Altman attributes the success of Y Combinator and his own career to the "peer group effect," noting that surrounding oneself with ambitious founders is more critical than formal education; he explicitly left Stanford to join this environment.
He describes his personal identity not as "formidable," but as someone driven by a refusal to accept the status quo and a belief in acting on first principles.
The "adults in the room" myth was debunked for Altman; he now believes no one possesses the answers and that success requires iterating quickly without perfect information.
He advises founders to find a peer group early and warns that while conviction is necessary to start, it must be discarded immediately upon encountering contradictory data.
Altman recounts being told as a teenager not to work on neural networks at the AI lab he visited, highlighting the historical resistance to the very path that eventually succeeded.

Altman defines a five-level roadmap for AI capabilities:
- Level 1: Chatbots and basic conversational interfaces.
- Level 2: Reasoning models capable of complex problem-solving (achieved with O1).
- Level 3: Agents capable of executing multi-step, long-term tasks with external interaction and self-correction.
- Level 4: Innovators, where AI systems can independently explore undiscovered phenomena and conduct scientific research.
- Level 5: Full organizational execution, where AI manages and operates entire companies or large-scale industrial processes.
He notes that Level 4 (Innovator stage) may be reachable sooner than anticipated, citing a hackathon example where a startup used AI to iteratively design and optimize an airfoil for competitive lift.
Altman suggests that future companies may operate with a "one person plus 10,000 GPUs" structure, leveraging AI to achieve enterprise-scale output with minimal human headcount.
He warns startups against the misconception that the "laws of business" (e.g., moats, competitive advantage) do not apply simply because they utilize AI; enduring value must still be built.
Altman identifies the startup advantage in the current landscape as speed and the ability to react to technology shifts within days rather than quarters or years, a luxury large corporations lack.

OpenAI recently underwent significant structural changes and leadership transitions following the ousting of Sam Altman, which he describes as the result of a company "speed-running" its growth arc from zero to massive scale in under two years.
He expresses confidence that the company now has clear alignment on research, infrastructure, and product paths, enabling faster execution toward AGI.
Altman mentions he is most personally excited about 2025 for the birth of his child, while professionally he remains focused on the realization of AGI and its implications for human prosperity.