DeepSeek at funding meeting: Liang Wenfeng discusses large language model technology and business philosophy

At a fundraising meeting, DeepSeek founder Liang Wenfeng presented a detailed analysis of large language model technologies and the company's commercial strategy. He discussed key architectural decisions, including the use of sparse attention and mixture of experts to improve training efficiency. Liang emphasized that DeepSeek is committed to openness, publishing models and technical reports to foster community development. He also touched on competition in the AI market, noting that China should focus on fundamental research rather than just applications. During the meeting, the prospects of multimodal models and the reduction of computing costs through hardware and software optimization were addressed.