So AI will come up with a way to extract resources from the environment automatically, transport them to facilities to refine, create and fabricate, engineer and build the testing equipment, perform the experiments en masse somehow faster than current time requires? It seems like a small part of the equation will be sped up but it will be interesting to see if anything else will change right away .. It will also be interesting to see what kind of usefulness these LLMs will have in uncharted territory. They are great so far with information humans have already learned and developed, but who knows if stacking transformer layers on an LLM will actually benefit invention and innovation.. since you can't train on data that doesn't exist, RLHF is probably not going to help much, etc. Maybe I'm wrong, we will see!