Every knowledge base ever built for humans died for the same reason.
The maintenance job went to the wrong person.
Cross-referencing sources, updating summaries when new data arrives, flagging when a new finding contradicts an old one. That's a specific, tedious, never-ending job. Nobody does it. So the knowledge base rots and everyone goes back to starting from scratch.
Karpathy's fix: give that job to the LLM. You dump raw sources into a folder. The AI compiles a wiki. Writes summaries, builds cross-references, flags contradictions, maintains an index. You never touch the wiki. The AI maintains it entirely.
18.7M views on his post. The insight underneath is simple: the problem was never the reading or the thinking. It was the bookkeeping. LLMs don't quit bookkeeping.
@StockSavvyShay@fiscal_ai@StockSavvyShay Microsoft will eventually lose enterprise collaboration software revenue and that is slowly reflecting in the stock price