I’d take your side. In 5 years we could be in low-medium automation, high automation, or full automation, and all of these seem more likely to have <100 papers.
In low-medium automation there’s no way to get 100 papers a year of output at current quality bar.
In high or full automation AIs will be doing ~all the research, but papers are designed for humans. At 100 papers per year, the average paper will probably be read by less than 5 humans. So I’d expect standards to rise and work to be consolidated into fewer papers rather than the number of papers exploding. AIs that need to communicate could use some other format. It would be weird but not impossible for the median to hit 100.
In full automation, we have the same issues as high automation, plus the median postdoc will be basically obsolete and only the ones with lots of compute will contribute.
I’d take your side. In 5 years we could be in low-medium automation, high automation, or full automation, and all of these seem more likely to have <100 papers.
In low-medium automation there’s no way to get 100 papers a year of output at current quality bar.
In high or full automation AIs will be doing ~all the research, but papers are designed for humans. At 100 papers per year, the average paper will probably be read by less than 5 humans. So I’d expect standards to rise and work to be consolidated into fewer papers rather than the number of papers exploding. AIs that need to communicate could use some other format. It would be weird but not impossible for the median to hit 100.
In full automation, we have the same issues as high automation, plus the median postdoc will be basically obsolete and only the ones with lots of compute will contribute.