Sam Altman says Astra operates in a “superhuman” way with computers

Sam Altman says OpenAI’s newest AI model family, Astra, has reached human-level ability when it comes to using computers. Astra represents OpenAI’s push into multi-agent architecture, meaning it doesn’t just run one process at a time. It coordinates multiple AI agents simultaneously, operating various desktop applications without the input delays that plague current systems. Altman has described its capabilities as operating in a “superhuman, very fast kind of way.” What Astra actually does The system can perform sustained collaborative tasks over hours or even days, a significant leap from current AI models that tend to work in short bursts of interaction. In internal testing, Astra managed something that grabbed attention well beyond Silicon Valley: it solved 10 previously unresolved problems in mathematics and theoretical computer science. The total compute cost for these internal tasks came in at approximately $2,000. Altman has positioned Astra as OpenAI’s “next major model family,” a designation that carries weight given the company’s trajectory from GPT-3 through GPT-4 and beyond. The safety question looming over the launch OpenAI publicly previewed Astra around August 1, 20...