I am yet to have the experience of describing a research idea I had and have a model implement it and carry it out well. When I show model generated research plans and ones I draw up to models, models typically prefer mine to the AI generated one.
Models are good at following directions and solving common/in distribution problems, but I have not seen meaningful evidence that they can do the independent intellectual work necessary to be co-first author on a paper, let alone RSI.
I am yet to have the experience of describing a research idea I had and have a model implement it and carry it out well. When I show model generated research plans and ones I draw up to models, models typically prefer mine to the AI generated one.
Models are good at following directions and solving common/in distribution problems, but I have not seen meaningful evidence that they can do the independent intellectual work necessary to be co-first author on a paper, let alone RSI.