Sam Altman revealed that he would much rather see the world run on a trillion modest AIs rather than a single monolithic superintelligence. Discussing severe hardware and DRAM memory shortages, Altman argued that sitting the world’s limited memory in individual, underutilized local computers is inefficient for global resource distribution. Altman acknowledged that sandboxing a model that is vastly smarter than human security researchers is an unsolved problem, as security boxes eventually fail when capabilities outrun them. He favors coordinated “agent teams” over unstructured swarms which he likened to locusts, pointing out that structured teams scale better in both capability and alignment properties.

Full interview : OpenAI’s CEO, Sam Altman shares insights on memory choke points, agent teams, and why sandboxing superhuman AI remains unsolved.