Backcasting AI speedup feels strange to me. Below a certain usefulness threshold, the researchers don’t use AI assistance for research. It doesn’t actively slow them down below that threshold, so it bottoms out at whatever their unassisted rate is.
Backcasting AI speedup feels strange to me. Below a certain usefulness threshold, the researchers don’t use AI assistance for research. It doesn’t actively slow them down below that threshold, so it bottoms out at whatever their unassisted rate is.