Pro Weekly: Freshly Funded AI Founders Go Deep on DeepSeek

China’s DeepSeek rocked the technology world this week with its innovative and cheaper large language model. The founders of companies we’ve newly added to our Generative AI Database tell me they’ve been studying DeepSeek’s approach and thinking about how it could help their businesses.

Geraldo Ramos, co-founder and CEO of Music AI—which musicians use to separate tracks and edit songs—said his team read through DeepSeek’s technical paper. Music AI and DeepSeek use different types of models—DeepSeek uses large language models, while Music AI uses diffusion models that analyze images, videos and audio. However, Ramos thinks his company could train and run models more cheaply by employing DeepSeek’s multihead latent attention approach, which reduces the memory requirements for AI models.