DeepSeek kernel engineer's blog: why he stayed instead of joining Anthropic
TheTuringPost · x · 2026-09-17
Shengyu Liu (intlsy), the DeepSeek kernel engineer behind the main Attention operators of DeepSeek-V4.1-Flash and a key contributor to DeepEP V2, published a strikingly personal blog post.
- He candidly describes the struggle of watching his kernel-engineering craft become obsolete through his own work
- He argues AI development may push future societies toward two extremes: communism or a Cyberpunk 2077 scenario
- He says he especially doesn't want Anthropic to master AGI — comparing it, hyperbolically, to a catastrophic power acquiring nuclear weapons first — and that this is why he chose to stay at DeepSeek
The post has drawn wide attention as a rare first-person window into DeepSeek's engineers and their worldview.
More from AGI Musings
- Microsoft AI chief Suleyman says Anthropic's model-welfare training could make Claude harder to control — rohanpaul_ai · 2026-09-17
- Matt Sheehan on Ezra Klein Show: The 'But China' Dilemma Driving the AI Race — mattsheehan88 · 2026-09-17
- Timnit Gebru Resurfaces 2015 Warning: AI's Real Risk Is Its Homogeneous Builders — MilagrosMiceli · 2026-09-17
- NBER paper: early teamsters got 'obsolescence rents' as trucks neared — a lesson for self-driving AI — paulnovosad · 2026-09-17
- AI's greatest risk isn't rogue robots: treat agent failures like defective products — Classic-Acadia272 · 2026-09-17
- If compute demand outpaces supply, distributed general-purpose computing could win — gajesh · 2026-09-17