本文由 AI 分析生成
建立時間： 2025-12-27 來源： https://x.com/aakashgupta/status/2004713516930855284
Summary
A commentary on Andrej Karpathy’s confession that he’s “never felt this much behind as a programmer.” The author argues that if someone who built the underlying neural networks feels behind, it signals that raw intelligence and deep technical knowledge no longer guarantee mastery. The new skill is orchestrating a “zoo” of 15+ stochastic primitives (agents, subagents, prompts, contexts, memory, modes, permissions, tools, plugins, skills, hooks, MCP, LSP, slash commands, workflows) that didn’t exist 18 months ago and evolve weekly.
這是對 Andrej Karpathy 自白「從未感覺如此落後」的評論。作者主張，如果連打造底層神經網路的人都感覺落後，就代表原始智力與深厚技術知識已不再保證精通。新的技能是駕馭一整群 15 種以上的隨機性原語（agents、subagents、prompts、contexts、memory、modes、permissions、tools、plugins、skills、hooks、MCP、LSP、slash commands、workflows），這些在 18 個月前都還不存在，且每週都在演進。
Key Points
- Karpathy — who built the networks inside coding assistants, taught deep learning at Stanford, ran AI at Tesla — says he feels dramatically behind.
- The new stack is about orchestration of stochastic systems, not understanding transformers or writing elegant algorithms.
- 15+ new primitives appeared in ~18 months, each evolving weekly.
- Traditional engineering is deterministic; the new entities are “stochastic, fallible, unintelligible and changing.”
- “Alien tool with no manual”: everyone is reverse-engineering capabilities in real time; 3-month-old best practices are already wrong.
Insights
The “magnitude 9 earthquake already hit; aftershocks are the new normal” line reframes the discomfort as a permanent condition, not a transition to be waited out. The deeper implication is a leveling effect: when documentation is perpetually stale and expertise decays in months, the advantage shifts from accumulated knowledge to fast empirical adaptation — echoing the same Context-Engineering-over-credentials thesis appearing elsewhere.
Connections
Raw Excerpt
The new stack isn’t about understanding transformers or writing elegant algorithms. It’s about orchestrating a zoo of stochastic systems that nobody fully controls.