Keep pulling the thread on Greg Brockman.
During the Dota 2 project, OpenAI observed that doubling the compute cores consistently resulted in a 2x performance improvement for the AI agent.
Bill Gates described the demo of GPT-4 as the best technology demonstration he had seen since visiting Xerox PARC.
Greg Brockman predicts that an AI will make a novel advance in mathematics or science, such as solving a Millennium Prize problem, within two to five years.
OpenAI is building large data centers in Abilene, Texas, in partnership with Oracle.
Greg Brockman believes energy availability is a very possible bottleneck for future AI progress, which is why OpenAI advocates for increased power generation.
Greg Brockman asserts that for the United States to remain competitive in AI, it must significantly increase its power generation capacity.
Greg Brockman predicts that AI for coding will evolve into a "full AI coworker," allowing a human to act as a manager directing a team of software engineering agents.
OpenAI's initial strategy was to pursue technological advancements without a predefined problem or application, which is the reverse of a typical startup approach.
A key management lesson from the Dota 2 project for OpenAI was to set milestones based on inputs, such as experiments to run, rather than on outcomes.
Greg Brockman asserts that AI has not yet passed the strict version of the Turing test, which requires an AI to be fully indistinguishable from a human.
OpenAI decided to build a product based on GPT-3 primarily to secure the necessary funding to continue its research.
OpenAI's first product strategy for GPT-3 was to release an API and allow the market to discover the use cases, a project Greg Brockman felt was "totally doomed" at the time.