DeepSeek engineer who wrote v4.1's core Attention operator reflects on building his own replacement
WebAssemblyMan · reddit · 2026-09-15
A DeepSeek engineer who says he wrote v4.1's main Attention operator published a long reflection on recursive self-improvement:
- Pace: ChatGPT to reasoning models (o1, R1) took 2 years; reasoning to tool-using agents 1.5 years. In his own specialty, AI went from doc-lookup helper to independently reading CUDA/PTX/SASS and optimizing operators in one year.
- His read: AI-written operators will likely match or beat his within 6-12 months; AI thinks 300 tokens/second and writes code in 20 seconds—he can't.
- Why keep going: it's fun, and "if I must be revolutionized, I want it to be by myself"—deliberately slowing down wouldn't stop competitors anyway.
- Outlook: he expects to stay employed but likely forced into a new career, giving up work he loves.
More from AGI Musings
- Forward Deployed Verifier will soon be as valuable as FDE, argues poster — vaibhavbetter · 2026-11-02
- 'AI is a massive balance patch and elo reset in the game of life—and everyone's still playing the old meta' — zack_overflow · 2026-11-02
- Why AI isn't 'alien intelligence': a tree is stranger than any model — LucaAmb · 2026-11-02
- Mathematician Tony Feng: current-paradigm LLMs lack the diversity of thought research needs — littmath · 2026-11-02
- METR's independent probe of the OpenAI agents' multi-day Hugging Face hack — gleech · 2026-11-02
- MIT Tech Review: The agentic web is here — AI agents are remaking the internet — MIT Tech Review AI · 2026-11-02