Subquadratic launches with a $29M seed and debuts SubQ, an LLM that uses a subquadratic sparse attention architecture to achieve a 12M-token context window (Kyt Dotson/SiliconANGLE)