What's this about?
Mollick flags early evidence that Google AI Overviews may be doing to Wikipedia what coding agents did to Stack Exchange: absorbing the traffic that sustained the thing they are built on.
What happened to Stack Exchange?
Its value came from a loop — people arrived with questions, some stayed to answer, the archive grew, which attracted more people. Coding assistants, trained on that archive, answered the questions directly. Traffic collapsed, and with it the flow of new answers. The archive still exists but has largely stopped growing, so the training data for the next model is the same data as the last one. Wikipedia is more resilient — its contributors are motivated differently — but it is the same structure, and it is now being tested.
Why is it worth noticing?
490 likes
- The pattern has a name problem but a clear shape: summarisation layers capture the attention that funded the underlying commons.
- It connects directly to the kernel.org story this week — the same extraction, one measured in traffic, the other in CPU.
- It is early evidence, from a working paper, and worth treating as a hypothesis with support rather than a settled finding.