Building and governing transformative AI systems at scale

Major model releases should be thoughtful and may not follow numbered versioning.

Instead of GPT-5, OpenAI may release improvements continuously so the system “just gets better and better fairly continuously.”

Why: Continuous improvement is easier for society to adapt to and better reflects how the technology is evolving.

"I don't even know if we'll call it GPT-5... it's not the 1, 2, 3, 4, 5, 6, 7, but you use an AI system, and the whole system just gets better and better fairly continuously."

Advanced AI capabilities should eventually reach free-tier users.

Altman regrets that GPT-4-level tech is still pay-walled and wants to find ways to offer it free.

Why: Broad, low-cost access aligns with OpenAI’s mission of letting people “invent the future” rather than having AGI handed down from above.

"We really want to figure out how to make more advanced technology available to free users too... it makes me sad that we have not figured out how to make GPT-4 level technology available to free users."

Both open and closed source play important roles; the mission is to distribute AGI’s benefits broadly.

OpenAI has open-sourced some work and will open-source more, but its primary path is a controlled system that delivers value widely.

Why: Different users and use-cases benefit from different levels of openness; a single strategy cannot serve everyone.

"I think there's great roles for both... our mission is to build towards AGI and to figure out how to broadly distribute its benefits."

Enduring value comes from the full intelligence layer, not just model weights.

OpenAI focuses on product, price, and user experience rather than only on keeping weights secret.

Why: Weights alone are insufficient; integration, UX, and reliability create defensible value.

"What we're trying to make is like this useful intelligence layer for people to use... We'll have to build up enduring value, the old fashioned way like any other business does."

Dramatic reductions in cost and latency are inevitable through algorithmic gains.

Altman is confident that research plus engineering tailwinds will make high-level intelligence “too cheap to meter.”

Why: Lower cost and latency unlock new applications and wider adoption.

"I am confident we'll be able to... cut the latency super dramatically. We want to cut the cost really, really dramatically."

The world must scale AI infrastructure—chips, energy, data centers—at unprecedented speed.

Bottlenecks include logic fab capacity, HBM supply, permitting, concrete, and energy sourcing.

Why: Massive value creation will drive supply-chain investment, but proactive coordination can accelerate timelines.

"The world needs a lot more AI infrastructure, a lot more than it's currently planning to build and with a different cost structure."

The iPhone sets a very high bar; any new device must offer a fundamentally different interaction paradigm.

Voice and multimodal input (camera, vision) are hints, but the device must surpass the iPhone’s excellence.

Why: Incremental improvements are insufficient; only a novel interface justifies carrying a second or replacement device.

"I personally think an iPhone is the greatest piece of technology humanity has ever made... the threshold is very high here."

Build a world equally usable by humans and AIs to preserve shared interfaces.

Example: an AI assistant that can click through DoorDash while the user watches and corrects in real time.

Why: Shared interfaces smooth hand-offs, maintain interpretability, and avoid forcing humans to learn new abstractions.

"I'm actually very interested in designing a world that is equally usable by humans and by AIs... a shared interface is nice."