55k ICLR submissions this year. Up from ~1k in 2018.
AI will do what the printing press did.
The printing press didn’t make every book better. It made publishing cheap enough that vastly more people could produce and distribute ideas.
That meant more junk, more mediocre work, more propaganda, but also far more science, philosophy, literature and technical knowledge.
AI pushes the same dynamic one layer upstream. It dramatically lowers the cost of creating intellectual work itself: papers, code, analysis, experiments, designs, arguments, hypotheses.
So yes, a lot of it will be slop. That’s the bell curve. More low quality, more mid, more high quality.
The thing is - at this volume you get far more edge cases, unexpected combinations, failed experiments, strange ideas and breakthroughs.
Edge cases. Remember that. Flying was an edge case once. We need more edge cases that stretch the envelope and push us forward. This is how we get there.
1600 submissions to ICLR.
Up from about 1000 last year.