www.alphamatch.ai
Subquadratic Raises $29M to Launch SubQ — Alphamatch AI
The Miami startup's SubQ model uses a fully subquadratic sparse attention architecture with 12M token context window, claiming 1,000x less compute than frontier models.
本期叼回三个项目:Subquadratic(迈阿密,$29M 种子轮)用 SSA 架构把注意力计算压缩 1000 倍,12M token 上下文,要从底层重写长上下文代价;Recursive Superintelligence(旧金山,Richard Socher 创立,$650M / $4.65B 估值)4 个月团队无产品却融了目前最大早期轮之一,押注递归自我改进 AI;脑智算芯(上海,复旦团队,天使轮)用类脑稀疏激活芯片攻击大模型推理能耗问题,已适配千问/GLM/DeepSeek。储备追踪:Reflection AI $20 亿融资存档。
The Miami startup's SubQ model uses a fully subquadratic sparse attention architecture with 12M token context window, claiming 1,000x less compute than frontier models.
Richard Socher's new $650 million startup Recursive Superintelligence wants to build AI that can research and improve itself indefinitely — and he insists it will ship products.
Add more perspectives or context around this Drop.