Sam Altman: The Future of OpenAI, ChatGPT's Origins, and Building AI Hardware

Altman says starting OpenAI was uncertain: in 2015, AGI seemed like a distant goal, and DeepMind appeared far ahead. The early team had eight people, no product or revenue, and was trying to produce useful research. He says a distinctive mission helped attract researchers who believed in the goal.

Altman says there is a large gap between what reasoning models can do and the products built to use them. He expects model costs to fall and describes an upcoming open-source model that could run powerful systems locally. He sees ChatGPT memory, data connections, and persistent background assistance as steps toward a service that can proactively help users.

Altman describes agents such as Codex and Deep Research as junior employees that can handle computer-based tasks for a few hours. He imagines AI interfaces that can act with less interruption, alongside new devices and eventually robots that do useful physical work. He also connects AI and robotics to a possible revival of domestic manufacturing.

Altman advises founders not to build a copy of ChatGPT’s core assistant, but to explore the much wider range of problems AI can address. He says companies often earn time to develop defensibility by doing something distinctive before competitors converge. He also argues that faster, cheaper iteration gives startups an advantage when technology changes an industry’s pace.

Altman says AI could help individuals and small teams accomplish more by reducing coordination costs. He is especially interested in AI for science, which he sees as a path to faster discoveries and long-term improvements in people’s lives. He links that ambition to energy, which he describes as a fundamental constraint on expanding AI.