当 AI 做完你的毕生课题，价值会转移到哪里？

Revenue headlines can look contradictory when reports use different gross, net, and annualized measures rather than the same accounting basis.

A technology-heavy survey can reveal how early adopters use AI without proving that the same returns apply across the whole economy.

Using open models does not imply abandoning proprietary models, because organizations can choose different systems for different tasks.

AI-generated mathematical results should be treated as claims requiring validation rather than as accepted proofs merely because they were produced quickly.

A correct statement becomes part of living mathematics when people understand it, connect it to other ideas, and learn how to use it.

As mathematical output expands, attention and taste may become scarcer than the ability to produce another technically correct result.

Researchers may experience AI progress as both an identity threat and an opportunity to pursue questions that were previously out of reach.

Fields with clear verification may support faster AI feedback loops than fields constrained by physical experiments, shifting conditions, or ambiguous judgments.

A theoretical advance should not be confused with practical performance or a broken security system unless implementation and independent evidence support that conclusion.

The useful next question is how people reorganize research around new capabilities, not only whether those capabilities feel impressive or unsettling.