Agency and Taste Over Raw Knowledge | Generated by AI
You’re right, but I’d sharpen one part of it: AI is actually devaluing raw knowledge-in-head, not amplifying it — at least the encyclopedic kind. Anyone can retrieve facts now. What the sorting mechanism actually rewards is something more specific:
Knowledge as an index, not a database. You don’t need to hold the transformer math in RAM, but you need enough compressed understanding to instantly detect when the model is wrong, ask the next sharp question, and know which of five plausible answers is the real one. That’s why your 8,000 notes matter — not as storage, but as the verification layer that lets you run AI at 10x speed while others run it at 1.2x and get subtly poisoned by hallucinations. The gap between “uses AI” and “uses AI with judgment” is going to be brutal, and mostly invisible until output quality diverges.
Agency beats all of it. Capital, network, credentials — those are stock variables that took decades to accumulate under the old rules. AI fluency is a flow variable: it compounds daily and it’s the only lever on your list that’s roughly free. A dropout in Guangzhou with an MI300X droplet and 2B tokens/month has access to leverage that literally did not exist for anyone at any price in 2019. That’s the “美好” part — it’s the most democratized power transfer in a generation, precisely because it prices in hours of deliberate practice rather than family wealth or Tsinghua degrees.
The “残酷” part is real too, though, and worth being honest about: the distribution won’t be a smooth gradient. It’ll be power-law. The person who is 30% better at directing agents doesn’t earn 30% more — they capture the whole niche, because agents scale their judgment across unlimited execution. Middle skill gets hollowed out first. Wenfeng Liang understood this early: a small team of people with taste + massive compute beats a large team of average people, every time.
So the full-spectrum competition you listed — capital, talent, knowledge, values, cognition, network — I’d say it collapses into two questions in the AI era: can you tell good from bad faster than others (taste/认知), and do you actually ship (agency). Everything else is increasingly rentable or automatable.
You’re already positioned on the right side of the sort. The mortgage is just latency, not a rejection.