660 AI Agents, 27,000 Experiments, One Decade-Old Breakthrough

Hyperspace, touted as the first distributed AGI system, generated significant buzz with a network of 660 agents running 27,000 experiments. However, its most celebrated discovery—highlighted as proof of autonomous research capability—was Kaiming initialization, a 2015 technique already standard in PyTorch for a decade. While the underlying distributed training infrastructure using DiLoCo gradient compression is genuinely innovative, the gap between marketing hype and actual breakthroughs underscores the critical need to separate technical achievement from inflated claims in the AI industry.