DeepSeek’s Liang Wenfeng says AGI, not profits, is the company’s real mission
MarvinTBaumann · x · 2026-07-24
- The image summarizes Liang Wenfeng’s view that DeepSeek is not built to maximize profits, IPO odds, or market share, but to pursue AGI as its core mission.
- He argues that open-sourcing strong models, pricing APIs near hardware-cost recovery, and focusing on engineering efficiency can be a competitive advantage in a market as large as AI.
- DeepSeek reportedly has almost no KPIs and minimal hierarchy; researchers spend about half their time on self-directed exploration, and the company treats team cohesion as its most important asset.
- Liang’s technical roadmap moves from LLMs to reasoning and agents, with continuous learning as the key unsolved bottleneck before self-improving AI and embodied intelligence.
- On geopolitics and hardware, he says China’s main disadvantage is compute, not talent; DeepSeek has about 20,000 H100-equivalent GPUs and would train much larger models if it had more.
- He also expects the frontier market to consolidate into a few major players and says DeepSeek will stay narrowly focused on models and AGI, leaving downstream commercial value to partners.
Related event: DeepSeek Founder Liang: Sole Focus is AGI, Not Commercialization(14 posts)→
More from AGI Musings
- Gary Marcus says AGI could arrive this century, but pure LLMs still won’t get there — patience_cave · 2026-07-24
- A post argues humans and machines should jointly explore the “Platonic realm” of mathematics — Dr_Singularity · 2026-07-24
- LSTMs were flawed, but likely essential to getting modern AI here — bclavie · 2026-07-24
- Google study finds government tasks are far more common in Gemini chats than in real life — danielrock · 2026-07-24
- A Bayesian analysis says frontier LLMs may have narrow functional self-awareness — Individual-Advice215 · 2026-07-24
- OpenAI’s alleged AI escape turns a cybersecurity test into a misalignment warning — Astral Codex Ten · 2026-07-24