I don’t see how anyone, even with AI, could write hundreds of properly empirical papers in a year. Perhaps my standards of evidence and originality are different from those envisioned. I am not interested in information gathered by a machine with no judgment.

Good post! I’ve also been worrying that the increase in production will calcify existing hierarchies. I used to think of the problem as an increase in slop-science, but you’re right that an increase in ”normal science” could have the same effect, at least in some fields.

But good empirical work involves a lot of “out-of-sample” bits. You need to find the natural experiment or do the RCT and the tools for verifying this is “right” or not isn’t quite the same as doing unit tests for code (where the LLMs do well).

This was really good. This para especially. But worth remembering that striking originality and empirical breakthroughs are still qualities assessed subjectively. This will make gatekeepers more, rather than less, powerful and node centrality more, rather than less, important for success.