4 Hours, 118 Responses: Liang Wenfeng’s Internal Q&A Addresses Everything
**DeepSeek Founder Liang Wenfeng's Candid Reflections on the Company's Path to AGI**
DeepSeek has recently completed its first external funding round, raising over 500 billion RMB (approx. $74B) at a pre-money valuation of 3.675 trillion RMB ($543B). Founder Liang Wenfeng personally invested 200 billion RMB. This marks a strategic shift from its initial "no financing, no IPO, no commercialization" principle.
In a recent investor Q&A, Liang articulated DeepSeek's core philosophy and roadmap. The company is driven by a powerful, unwritten vision for beneficial AGI rather than pure commercial maximization. He emphasizes "strategic restraint"—avoiding unnecessary conflicts, prioritizing long-term AGI success over short-term gains, and maintaining an open, cooperative stance even with competitors.
Liang outlined the AGI technical roadmap: current focus on Agent capabilities, followed by solving "continual learning," which he sees as the key to unlocking models that can learn and adapt like humans. This could lead to a gradual "singularity" where AI accelerates its own research, and eventually to embodied intelligence. DeepSeek will strictly focus on this "AGI mainline," avoiding distractions like video generation which, while commercially viable, don't directly advance core intelligence.
He identifies team stability as the single most critical factor for success, now bolstered by the recent funding. While talent is not a bottleneck, the primary constraint compared to the US is compute resources. Liang is optimistic about domestic AI chips, stating that Nvidia's CUDA moat is eroding and that within a year, the viability of the Chinese chip ecosystem will be proven, with Huawei's offerings being key. The main issue is production capacity.
On competition, Liang believes the final differentiators will be cost, time-to-market, and user experience. He foresees Chinese companies playing a major role by offering systematically lower-cost AI services globally. DeepSeek's commercialization strategy involves offering API services at a "reasonable profit" and focusing on coding Agents. He remains committed to open-sourcing even their strongest models, seeing no downside as the barriers to effective deployment remain high.
The company operates with a unique dual management structure combining top-down direction with significant bottom-up, unstructured research time for employees. Data quality and post-training are identified as major current challenges, with half of core researchers involved in data labeling efforts. Liang concludes that DeepSeek aims to be one of several trillion-dollar companies in the AI era, achieved through extreme focus on its chosen path.
链捕手Há 25m
